The characteristics of multispectral (MS) and panchromatic (P) image fusion, or pansharpening, methods are investigated. Depending on the way spatial details are extracted from P, such methods can be broadly labeled into two main classes, roughly corresponding to component substitution (CS), also known as projection substitution, and methods based on multiresolution analysis (MRA), i.e. on digital filtering. Theoretical and experimental results carried out on QuickBird and Ikonos data sets evidence that CS-based fusion is far less sensitive than MRA-based fusion to registration errors, i.e. spatial misalignments between MS and P images, possibly originated by cartographic projection and resampling of individual data sets, and aliasing occurring in MS bands and deriving from a modulation transfer function (MTF) of each MS channel that is excessively broad relatively to the spatial sampling interval. Simulated misalignments carried out at full scale by means of a suitable quality evaluation protocol have evidenced the quality-shift tradeoff of the two classes: MRA methods yield a slightly superior quality in the absence of misalignments, but are more penalized, whenever shifts between MS and P are present, than CS methods producing a slightly lower quality in the ideal case, but that are intrinsically more shift tolerant.

Super-resolution mapping is used to produces thematic maps at a scale finer than the source images. This paper presents a new super-resolution mapping approach that exploits the typically fine temporal resolution of coarse spatial resolution images as it input and an adoption of an active threshold surface using Hopfield neural network as a means to map land cover at a sub-pixel scale. The results demonstrated that the proposed technique is slightly more accurate than the existence technique in terms of site specific accuracy and produce better visualization on individual land cover map.

The purpose of this work is to review and evaluate the performance of several algorithms which have been designed for satellite imagery in a geographic context. In particular we are interested in their performance with low-baseline image pairs like those which will be produced by the Pleiades satellite. In this study local and global state of the art algorithms have been considered and compared: CARMEN, MARC, MARC2 and MICMAC. This paper aims also at proposing a new benchmark to compare stereo algorithms. A set of simulated stereo images for which the ground truth is perfectly known will be presented. The obtained accuracy for the ground truth is more than a hundredth of pixel. The existence of an accurate ground truth is a major improvement for the community, allowing to quantify very precisely the disparity error in a realistic setting.

In this paper we propose Alternating Sequential Attribute Filters, which are Alternating Sequential Filters (ASFs) computed with Attribute Filters. ASFs are obtained by the iterative subsequent application of morphological opening and closing transformations and process an image by filtering both bright and dark structures. ASFs are widely used for achieving a simplification of a scene and for the removal of noisy structures. However, ASFs are not suitable for the analysis of very high geometrical resolution remote sensing images since they do not preserve the geometrical characteristics of the objects in the image. For this reason, instead of the conventional morphological operators, we propose to use attribute filters, which are morphological connected filters and process an image only by merging flat regions. Thus, they are suitable for the analysis of very high resolution images. Since the attribute selected for use in the analysis mainly defines the effects obtained by the morphological filter, when applying attribute filters in an alternate composition (as the ASF) it is possible to obtain a different image simplification according to the attribute considered. For example, if one considers the area as attribute, an input image will be processed by progressively removing both larger dark and bright areas. When using an attribute that measures the homogeneity of the regions (e.g., the standard deviation of the values of the pixels) the scene can be simplified by merging progressively more homogeneous zones. Moreover, the computation of the ASF with attribute filters can be performed with a reduced computational load by taking advantage of the efficient representation of the image as min- and max-tree. The proposed alternating sequential attribute filters are qualitatively evaluated on a panchromatic GeoEye-1 image.

This paper addresses the problem of semi-automatic extraction of road or hydrographic networks in satellite images. For that purpose, we propose an approach combining concepts arising from mathematical morphology and hydrology. The method exploits both geometrical and topological characteristics of rivers/roads and their tributaries in order to reconstruct the complete networks. It assumes that the images satisfy the following two general assumptions, which are the minimum conditions for a road/river network to be identifiable and are usually verified in low- to mid-resolution satellite images: (i) visual constraint: most pixels composing the network have similar spectral signature that is distinguishable from most of the surrounding areas; (ii) geometric constraint: a line is a region that is relatively long and narrow, compared with other objects in the image. While this approach fully exploits local (roads/rivers are modeled as elongated regions with a smooth spectral signature in the image and a maximum width) and global (they are structured like a tree) characteristics of the networks, further directional information about the image structures is incorporated. Namely, an appropriate anisotropic metric is designed by using both the characteristic features of the target network and the eigen-decomposition of the gradient structure tensor of the image. Following, the geodesic propagation from a given network seed with this metric is combined with hydrological operators for overland flow simulation to extract the paths which contain most line evidence and identify them with the target network.

The qualified Earth Observation (EO) images sometimes may present unexpected artifacts. These perturbations and distortions can make more difficult the analysis of images and may decrease the efficiency of interpretation algorithms because the information is distorted. Thus is necessary to implement methods able to detect these artifacts regardless of the model which are formed, i.e. parameter free. In this article, we propose and present a method based on data compression, whether lossy compression or lossless compression for detecting aliasing, strips, saturation, etc.

The generation of agro-statistics in tropical regions, using remote sensing data, is a difficult task due to the presence of high variety of crops at a given time over the same area that are not restricted by a crop calendar. The proposed segmentation is developed into two phases: spatial and spectral. The spatial segmentation is based in using multidimensional mathematical morphology for the delimitation of the objects that define each of the production segments or plots. The basic multi-dimensional morphologic operators (dilatation, erosion and gradient), were defined to be adapted to the study of agricultural areas using multi-spectral imagery. Meanwhile, the spectral segmentation is performed assigning the dominant spectral signature to each spatial segment to reduce its number using a hierarchical grouping method.

Hyperspectral instruments acquire electromagnetic energy scattered within their ground instantaneous field view in hundreds of spectral channels with high spectral resolution. Very often, however, owing to low spatial resolution of the scanner or to the presence of intimate mixtures (mixing of the materials at a very small scale) in the scene, the spectral vectors (collection of signals acquired at different spectral bands from a given pixel) acquired by the hyperspectral scanners are actually mixtures of the spectral signatures of the materials present in the scene. Given a set of mixed spectral vectors, spectral mixture analysis (or spectral unmixing) aims at estimating the number of reference materials, also called endmembers, their spectral signatures, and their fractional abundances. Spectral unmixing is, thus, a source separation problem where, under a linear mixing model, the sources are the fractional abundances and the endmember spectral signatures are the columns of the mixing matrix. As such, the independent component analysis (ICA) framework came naturally to mind to unmix spectral data. However, the ICA crux assumption of source statistical independence is not satisfied in spectral applications, since the sources are fractions and, thus, non-negative and sum to one. As a consequence, ICA-based algorithms have severe limitations in the area of spectral unmixing, and this has fostered new unmixing research directions taking into account geometric and statistical characteristics of hyperspectral sources. This paper presents an overview of the principal research directions in hyperspectral unmixing. The presentations is organized into four main topics: i) mixing models, ii) signal subspace identification, iii) geometrical-based spectral unmixing, (iv) statistical-based spectral unmixing, and (v) sparse regression-based unmixing. In each topic, we describe what physical or mathematical problems are involved and summarize state-of-the-art algorithms to address these problems.

In the past few years, spectral analysis of data collected by hyperspectral sensors aimed at automatic anomaly detection has become an interesting area of research. In this paper, we are interested in an Anomaly Detection (AD) scheme for hyperspectral images in which spectral anomalies are defined with respect to a statistical model of the background Probability Density Function (PDF).The characterization of the PDF of hyperspectral imagery is not trivial. We approach the background PDF estimation through the Parzen Windowing PDF estimator (PW). PW is a flexible and valuable tool for accurately modeling unknown PDFs in a non-parametric fashion. Although such an approach is well known and has been widely employed, its use within an AD scheme has been not investigated yet. For practical purposes, the PW ability to estimate PDFs is strongly influenced by the choice of the bandwidth matrix, which controls the degree of smoothing of the resulting PDF approximation. Here, a Bayesian approach is employed to carry out the bandwidth selection. The resulting estimated background PDF is then used to detect spectral anomalies within a detection scheme based on the Neyman-Pearson approach. Real hyperspectral imagery is used for an experimental evaluation of the proposed strategy.

This paper addresses the unmixing of hyperspectral images, when intimate mixtures are present. In these scenarios the light suffers multiple interactions among distinct endmembers, which is not accounted for by the linear mixing model. A two-step method to unmix hyperspectral intimate mixtures is proposed: first, based on the Hapke intimate mixture model, the reflectance is converted into single scattering albedo average. Second, the mass fractions of the endmembers are estimated by a recently proposed method termed simplex identification via split augmented Lagrangian (SISAL). The proposed method is evaluated on a well known intimate mixture data set.

Algorithms have been developed to support quantitative analysis of a gas plume using down-looking airborne hyperspectral long-wave infrared (LWIR) imagery. The resulting gas quantification "GQ" tool estimates the quantity of one or more gases at each pixel, and estimates uncertainty based on factors such as atmospheric transmittance, background clutter, and plume temperature contrast. GQ uses gas-insensitive segmentation algorithms to classify the background very precisely so that it can infer gas quantities from the differences between plume-bearing pixels and similar non-plume pixels. It also includes MODTRAN-based algorithms to iteratively assess various profiles of air temperature, water vapour, and ozone, and select the one that implies smooth emissivity curves for the (unknown) materials on the ground. GQ then uses a generalized least-squares (GLS) algorithm to simultaneously estimate the most likely mixture of background (terrain) material and foreground plume gases. Cross-linking of plume temperature to the estimated gas quantity is very non-linear, so the GLS solution was iteratively assessed over a range of plume temperatures to find the best fit to the observed spectrum. Quantification errors due to local variations in the camera-topixel distance were suppressed using a subspace projection operator. Lacking detailed depth-maps for real plumes, the GQ algorithm was tested on synthetic scenes generated by the Digital Imaging and Remote Sensing Image Generation (DIRSIG) software. Initial results showed pixel-by-pixel gas quantification errors of less than 15% for a Freon 134a plume.

This paper is concerned with the segmentation of very high spatial resolution panchromatic images. We propose a method for unsupervised segmentation of remotely sensed images based on texture information and evidence theory. We first perform a segmentation of the image using a watershed on some coefficients issued from a wavelet decomposition of the initial image. This yields an over-segmented map where the similar objects, from a textural point of view, are aggregated together in a step forward. The information of texture is obtained by analyzing the wavelet coefficients of the original image. At each band of the wavelet decomposition, we compute an indicator of similarity between two objects. All the indicators are then fused using some rules of evidence theory to derive a unique criterion of similarity between two objects.

The automatic detection, recognition, and segmentation of object classes in remote sensed images is of crucial importance for scene interpretation and understanding. However, it is a difficult task because of the high variability of satellite data. Indeed, the observed scenes usually exhibit a high degree of complexity, where complexity refers to the large variety of pictorial representations of objects with the same semantic meaning and also to the extensive amount of available details. Therefore, there is still a strong demand for robust techniques for automatic information extraction and interpretation of satellite images. In parallel, there is a growing interest in techniques that can extract vector features directly from such imagery. In this paper, we investigate the problem of automatic hierarchical segmentation and vectorization of multispectral satellite images. We propose a new algorithm composed of the following steps: (i) a non-uniform sampling scheme extracting most salient pixels in the image, (ii) an anisotropic triangulation constrained by the sampled pixels taking into account both strength and directionality of local structures present in the image, (iii) a polygonal grouping scheme merging, through techniques based on perceptual information, the obtained segments to a smaller quantity of superior vectorial objects. Besides its computational efficiency, this approach provides a meaningful polygonal representation for subsequent image analysis and/or interpretation.

This work presents a morphological-based segmentation approach for coastline detection based on a waterfall hierarchical scheme. Hierarchical waterfall is constrained my markers in each step on the hierarchical tree. In this manner, a map of the waterfall minimum persistency is created to identify the coastline. The proposed algorithm was tested on Envisat-ASAR and TerraSAR-X images acquired over the Lisbon region.

High resolution optical remote sensing images allow to produce accurate land-cover maps. This is usually achieved using an ad-hoc mixture of image segmentation and supervised classification. The main drawback of this approach is that it does not scale for real world complete scenes. In this paper we present a framework which allows to implement this kind of image analysis without scale issues.

Band selection methods often suppose normal distribution or, at least, a significant number of samples per class to compute statistical parameters. In this paper, we propose a band selection technique that needs very few training samples per class to be effective. To take into account the spectral variability inside the classes and to be independent on statistic parameters, we propose to use a criterion which measures the separability between two classes. This criterion is based on the extension of Spectral Angle Mapper (SAM) to a SAM within- and between-classes. The proposed selection method consists in eliminating spectral bands from the original set, thanks to the previous criterion that check the increase of separability measure between classes on the remaining bands subset. We use a stochastic algorithm to choose, at each step, which band to eliminate. We proceed by successive elimination until we reach the number of desired bands or the maximum of the criterion. This top-down method allows taking simultaneously into account all the interesting bands during the whole process, instead of selecting them one by one. The method provides, at the end, the selected spectral bands for a pair of classes. We expand this two-class selection technique to the multiclass band selection. We improve the method by adding a pre-selection of interesting bands considering a measure of a spectral signal on noise ratio. Some examples are given to show the effectiveness of the method: as one main application is classification, we compare the results of classification achieved after the data reduction made by different methods. We check the efficiency according to the number of training samples.

Hyperspectral imaging is a continuously growing area of remote sensing. Hyperspectral data provide a wide spectral range, coupled with a very high spectral resolution, and are suitable for detection and classification of surfaces and chemical elements in the observed image. The main problem with hyperspectral data for these applications is the (relatively) low spatial resolution, which can vary from a few to tens of meters. In the case of classification purposes, the major problem caused by low spatial resolution is related to mixed pixels, i.e., pixels in the image where more than one land cover class is within the same pixel. In such a case, the pixel cannot be considered as belonging to just one class, and the assignment of the pixel to a single class will inevitably lead to a loss of information, no matter what class is chosen. In this paper, a new supervised technique exploiting the advantages of both probabilistic classifiers and spectral unmixing algorithms is proposed, in order to produce land cover maps of improved spatial resolution. The method is in three steps. In a first step, a coarse classification is performed, based on the probabilistic output of a Support Vector Machine (SVM). Every pixel can be assigned to a class, if the probability value obtained in the classification process is greater than a chosen threshold, or unclassified. In the proposed approach it is assumed that the pixels with a low probabilistic output are mixed pixels and thus their classification is addressed in a second step. In the second step, spectral unmixing is performed on the mixed pixels by considering the preliminary results of the coarse classification step and applying a Fully Constrained Least Squares (FCLS) method to every unlabeled pixel, in order to obtain the abundances fractions of each land cover type. Finally, in a third step, spatial regularization by Simulated Annealing is performed to obtain the resolution improvement. Experiments were carried out on a real hyperspectral data set. The results are good both visually and numerically and show that the proposed method clearly outperforms common hard classification methods when the data contain mixed pixels.

A key factor for the success of supervised remote sensing image classification is the definition of an efficient training set. Suboptimality in the selection of the training samples can bring to low classification performance. Active learning algorithms aim at building the training set in a smart and efficient way, by finding the most relevant samples for model improvement and thus iteratively improving the classification performance. In uncertaintybased approaches, a user-defined heuristic ranks the unlabeled samples according to the classifier's uncertainty about their class membership. Finally, the user is asked to define the labels of the pixels scoring maximum uncertainty. In the present work, an unbiased uncertainty scoring function encouraging sampling diversity is investigated. A modified version of the Entropy Query by Bagging (EQB) approach is presented and tested on very high resolution imagery using both SVM and LDA classifiers. Advantages of favoring diversity in the heuristics are discussed. By the diverse sampling it enhances, the unbiased approach proposed leads to higher convergence rates in the first iterations for both the models considered.

Many remote sensing data processing problems are inherently constituted by several tasks that can be solved either individually or jointly. For instance, each image in a multitemporal classification setting could be taken as an individual task but relation to previous acquisitions should be properly considered. In such problems, different modalities of the data (temporal, spatial, angular) gives rise to changes between the training and test distributions, which constitutes a difficult learning problem known as covariate shift. Multitask learning methods aim at jointly solving a set of prediction problems in an efficient way by sharing information across tasks. This paper presents a novel kernel method for multitask learning in remote sensing data classification. The proposed method alleviates the dataset shift problem by imposing cross-information in the classifiers through matrix regularization. We consider the support vector machine (SVM) as core learner and two regularization schemes are introduced: 1) the Euclidean distance of the predictors in the Hilbert space; and 2) the inclusion of relational operators between tasks. Experiments are conducted in the challenging remote sensing problems of cloud screening from multispectral MERIS images and for landmine detection.

A typical tendency in modern remote sensing (RS) is to apply multichannel systems. Images formed by them are in more or less degree noisy. Thus, their pre-filtering can be used for different purposes, in particular, to improve classification. In this paper, we consider methods of multichannel image denoising based on discrete cosine transform (DCT) and analyze how parameters of these methods affect classification. Both component-wise and 3D denoising is studied for three-channel Landsat test image. It is shown that for better determination of different classes, DCT based filters, both component-wise and 3D variants are efficient, but with a different tuning of involved parameters. The parameters can be optimized with respect to either standard MSE or metrics that characterize image visual quality. Best results are obtained with 3D denoising. Although the main conclusions basically coincide for both considered classifiers, Radial Basis Function Neural Network (RBF NN) and Support Vector Machine (SVM), the classification results appear slightly better with RBF NN for the experiment carried out in this paper.

This paper addresses the problem of the classification of very high resolution (VHR) SAR amplitude images of urban areas. The proposed supervised method combines a finite mixture technique to estimate class-conditional probability density functions, Bayesian classification, and Markov random fields (MRFs). Textural features, such as those extracted by the greylevel co-occurrency method, are also integrated in the technique, as they allow to improve the discrimination of urban areas. Copulas are applied to estimate bivariate joint class-conditional statistics, merging the marginal distributions of both textural and SAR amplitude features. The resulting joint distribution estimates are plugged into a hidden MRF model, endowed with a modified Metropolis dynamics scheme for energy minimization. Experimental results with COSMO-SkyMed and TerraSAR-X images point out the accuracy of the proposed method, also as compared with previous contextual classifiers.

Nowadays, an ever increasing number of multi-temporal images is available, giving the possibility of having with high temporal frequency information about the land-cover evolution on the ground. In general, the production of accurate land-cover maps requires the availability of reliable ground truth information on the considered area for each image to be classified. Unfortunately the rate of ground truth information collection will never equal the remote sensing image acquisition rate, making supervised classification unfeasible for land-cover maps updating. This problem has been faced according to domain adaptation methods that update land-cover maps under the assumption that: i) training data are available for one of the considered multi-temporal acquisitions while they are not for the others and ii) set of land-cover classes is same for all considered acquisitions. In real applications, the latter assumption represents a constraint which is often not satisfied due to possible changes occurred on the ground and associated with the presence of new classes or the absence of old classes in the new images. In this work, we propose an approach that removes this constraint by automatically identifying whether there exist differences between classes in multi-temporal images and properly handling these differences in the updating process. Experimental results on a real multi-temporal remote sensing data set confirm the effectiveness and the reliability of the proposed approach.

In this paper, a Synthetic Aperture Radar Automatic Target Recognition approach based on Gaussian process (GP) classification is proposed. It adopts kernel principal component analysis to extract sample features and implements target recognition by using GP classification with automatic relevance determination (ARD) function. Compared with k-Nearest Neighbor, Naïve Bayes classifier and Support Vector Machine, GP with ARD has the advantage of automatic model selection and hyper-parameter optimization. The experiments on UCI datasets and MSTAR database show that our algorithm is self-tuning and has better recognition accuracy as well.

The iteratively re-weighted multivariate alteration detection (IR-MAD) algorithm may be used both for unsuper- vised change detection in multi- and hyperspectral remote sensing imagery as well as for automatic radiometric normalization of multi- or hypervariate multitemporal image sequences. Principal component analysis (PCA) as well as maximum autocorrelation factor (MAF) and minimum noise fraction (MNF) analyses of IR-MAD images, both linear and kernel-based (which are nonlinear), may further enhance change signals relative to no-change background. The kernel versions are based on a dual formulation, also termed Q-mode analysis, in which the data enter into the analysis via inner products in the Gram matrix only. In the kernel version the inner products of the original data are replaced by inner products between nonlinear mappings into higher dimensional feature space. Via kernel substitution, also known as the kernel trick, these inner products between the mappings are in turn replaced by a kernel function and all quantities needed in the analysis are expressed in terms of the kernel function. This means that we need not know the nonlinear mappings explicitly. Kernel principal component analysis (PCA), kernel MAF and kernel MNF analyses handle nonlinearities by implicitly transforming data into high (even in¯nite) dimensional feature space via the kernel function and then performing a linear analysis in that space. In image analysis the Gram matrix is often prohibitively large (its size is the number of pixels in the image squared). In this case we may sub-sample the image and carry out the kernel eigenvalue analysis on a set of training data samples only. To obtain a transformed version of the entire image we then project all pixels, which we call the test data, mapped nonlinearly onto the primal eigenvectors. IDL (Interactive Data Language) implementations of IR-MAD, automatic radiometric normalization and kernel PCA/MAF/MNF transformations have been written which function as transparent and fully integrated extensions of the ENVI remote sensing image analysis environment. Also, Matlab code exists which allows for fast data exploration and experimentation with smaller datasets. Computationally demanding kernelization of test data with training data and kernel image projections have been programmed to run on massively parallel CUDA-enabled graphics processors, when available, giving a tenfold speed enhancement. The software will be available from the authors' websites in the near future. A data example shows the application to bi-temporal RapidEye data covering the Garzweiler open pit mine in the Ruhr area in Germany.

In this paper we present a technique for the detection of multiple changes in multitemporal and multispectral remote sensing images. The technique is based on: i) the representation of the change detection problem in polar coordinates; and ii) a 2-step decision strategy. First of all the change information present in the multitemporal dataset is represented taking advantage from the framework for change detection in polar coordinates. Within this representation the Bayesian decision theory is applied twice: the first time for distinguishing changed from unchanged pixels; and the second one for discriminating different kinds of change within changed pixels. The procedure exploits the Expectation-Maximization algorithm and is completely automatic and unsupervised. Experiments carried out on high and very high resolution multispectral and multitemporal datasets confirmed the effectiveness of the proposed approach.

In object-based change detection approaches using specified object features as change measures, segmentation is the crucial step, especially when also shape changes are considered. In this paper we present an enhanced segmentation procedure based on the multiresolution segmentation. The procedure segments the first image using the multiresolution segmentation. The segmentation is then applied to the second image and checked for its consistency. If a segment is found to be inconsistent with the second image, it is split up. The performance of the proposed procedure is demonstrated based on simulated and real image data.

This paper presents a novel unsupervised clustering scheme to find changes in two or more coregistered remote sensing images acquired at different times. This method is able to find nonlinear boundaries to the change detection problem by exploiting a kernel-based clustering algorithm. The kernel k-means algorithm is used in order to cluster the two groups of pixels belonging to the 'change' and 'no change' classes (binary mapping). In this paper, we provide an effective way to solve the two main challenges of such approaches: i) the initialization of the clustering scheme and ii) a way to estimate the kernel function hyperparameter(s) without an explicit training set. The former is solved by initializing the algorithm on the basis of the Spectral Change Vector (SCV) magnitude and the latter is optimized by minimizing a cost function inspired by the geometrical properties of the clustering algorithm. Experiments on VHR optimal imagery prove the consistency of the proposed approach.

The objective of this paper is twofold: first, to presents a generic approach for the analysis of Radarsat-1 multitemporal data and, second, to presents a multi classifier schema for the classification of multitemporal images. The general approach consists of preprocessing step and classification. In the preprocessing stage, the images are calibrated and registered and then temporally filtered. The resulted multitemporally filtered images are subsequently used as the input images in the classification step. The first step in a classifier design is to pick up the most informative features from a series of multitemporal SAR images. Most of the feature selection algorithms seek only one set of features that distinguish among all the classes simultaneously and hence a limited amount of classification accuracy. In this paper, a class-based feature selection (CBFS) was proposed. In this schema, instead of using feature selection for the whole classes, the features are selected for each class separately. The selection is based on the calculation of JM distance of each class from the rest of classes. Afterwards, a maximum likelihood classifier is trained on each of the selected feature subsets. Finally, the outputs of the classifiers are combined through a combination mechanism. Experiments are performed on a set of 34 Radarsat-1 images acquired from August 1996 to February 2007. A set of 9 classes in a forest area are used in this study. Classification results confirm the effectiveness of the proposed approach compared with the case of single feature selection. Moreover, the proposed process is generic and hence is applicable in different mapping purposes for which a multitemporal set of SAR images are available.

A new approach to the problems of statistical and structural pattern recognition, a signal processing and image analysis techniques has been considered. These problems are extremely important for tasks being solved by airborne and space borne remote sensing systems. Development of new remote sensors for image and signal processing is inherently connected with a possibility of statistical processing of images. Fundamentally new optoelectronic sensors "Multiscan" have been suggested in the present paper. Such sensors make it possible to form directly certain statistical estimates, which describe completely enough the different types of images. The sensors under discussion perform the Lebesgue-Stieltjes signal integration rather than the Cauchy-Riemann one. That permits to create integral functionals for determining statistical features of images. The use of the integral functionals for image processing provides a good agreement of obtained statistical estimates with required image information features. The Multiscan remote sensors allows to create a set of integral moments of an input image right up to high-order integral moments, to form a quantile representation of an input image, which provides a count number limited texture, to form a median, which provides a localisation of a low-contrast horizon line in fog, localisation of water flow boundary etc. This work presents both the description of the design concept of the new remote sensor and mathematical apparatus providing the possibility to create input image statistical features and integral functionals.

Orthorectification that corrects the perspective distortion of remote sensing imagery, providing accurate geolocation and ease of correlation to other images is a valuable first-step in image processing for information extraction. However, the large amount of metadata and the floating-point matrix transformations required to operate on each pixel make this a computation and I/O (Input/Output) intensive process. As result much imagery is either left unprocessed or loses timesensitive value in the long processing cycle. However, the computation on each pixel can be reduced substantially by using computational results of the neighboring pixels and accelerated by special pipelined hardware architecture in one to two orders of magnitude. A specialized coprocessor that is implemented inside an FPGA (Field Programmable Gate Array) chip and surrounded by vendorsupported hardware IP (Intellectual Property) shares the computation workload with CPU through PCI-Express interface. The ultimate speed of one pixel per clock (125 MHz) is achieved by the pipelined systolic array architecture. The optimal partition between software and hardware, the timing profile among image I/O and computation, and the highly automated GUI (Graphical User Interface) that fully exploits this speed increase to maximize overall image production throughput will also be discussed. The software that runs on a workstation with the acceleration hardware orthorectifies 16 Megapixels per second, which is 16 times faster than without the hardware. It turns the production time from months to days. A real-life successful story of an imaging satellite company that adopted such workstations for their orthorectified imagery production will be presented. The potential candidacy of the image processing computation that can be accelerated more efficiently by the same approach will also be analyzed.

This paper evaluates the potential use of nonlinear retrieval methods to derive cloud, surface and atmospheric properties from hyperspectral MetOp-IASI and MTG-IRS spectra. The methods are compared in terms of both accuracy and speed with the current IASI and IRS L2 PPFP implementation, which consists of a principal component extraction, typically referred as to Empirical Orthogonal Functions (EOF), and a subsequent canonical linear regression. This research proposes the evaluation of some other methodological advances considering 1) other linear feature extraction methods instead of EOF, such as (orthonormalized) partial least squares, and 2) the linear combination of nonlinear regression models in the form of committee of experts. The nonlinear regression models considered in this work are artificial neural networks (NN) and kernel ridge regression (KRR) as nonparametric multioutput powerful regression tools. Results show that, in general, nonlinear models outperform the linear retrieval both in the presence of noise and noise-free settings, and for both IASI and IRS synthetic and real data. The combination of models makes the retrieval more robust, improves the accuracy, and decreases the estimated bias. These results confirm the validity of the proposed approach for retrieval of atmospheric profiles.

Ground reference data are important for understanding and characterizing angular effects on the images acquired by satellite sensors with off-nadir capability. However, very few studies have considered image-based soil reference data for that purpose. Compared to non-imaging instruments, imaging spectrometers can provide detailed information to investigate the influence of spatial components on the bidirectional reflectance distribution function (BRDF) of a mixed target. This research reported in this paper investigated soil spectral reflectance changes as a function of surface roughness, scene components and viewing geometries, as well as wavelength. Soil spectral reflectance is of particular interest because it is an essential factor in interpreting the angular effects on images of vegetation canopies. BRDF data of both rough and smooth soil surfaces were acquired in the laboratory at 30° illumination angle using a Specim V10E imaging spectrometer mounted on the University of Lethbridge Goniometer System version 2.5 (ULGS-2.5). The BRDF results showed that the BRDF of the smooth soil surface was dominated by illuminated pixels, whereas the shaded pixels were a larger component of the BRDF of the rough surface. In the blue, green, red, and near-infrared (NIR), greater BRDF variation was observed for the rough than for the smooth soil surface. For both soil surface roughness categories, the BRDF exhibited a greater range of values in the NIR than in the blue, green, or red. The imaging approach allows the characterization of the impact of spatial components on soil BRDF and leads to an improved understanding of soil reflectance compared to non-imaging BRDF approaches. The imaging spectrometer is an important sensor for BRDF investigations where the effects of individual spatial components need to be identified.

Images issued from a SAR (Synthetic Aperture Radar) sensor are effected by a specific noise called speckle; therefore, many studies have been dedicated to modulate this noise with the aim to be able to reduce its effects. But, studies in the area of polarimetric SAR (PolSAR) images despeckling are still poor and don't take advantage correctly of polarimetric information. In this way, this paper describes an original and efficient method of despeckling PolSAR images in order to improve the visualization and the extraction of planimetric features. The proposed filter, takes into account all polarization modes for each polarization mode despeckling. So, for a pixel in a single polarization mode, the modification of its radiometric value will be supervised by it adjacent pixels in the same polarization mode and also by their equivalent pixels in other polarization modes. Furthermore, to avoid error propagation, the filter will be very cautious in modification of radiometric values in such way that it runs in many iterations modifying the less ambiguous pixels firstly and leaves the rest of the pixels for the next iterations for a possible modification. To combine the information resulting from each polarization mode and make a decision, the proposed filter calls some rules of the Evidence Theory. The experimentation was done on Radarsat-2 images of the Arctic and Quebec regions of Canada, and the results show clearly the benefit and the high performance of this despeckling approach.

An algorithm for waterline extraction from SAR images is presented based on the estimation of the geodesic path, or minimal path (MP) between two pixels on the waterline. For two given pixels, geodesic time is determined in terms of the time shortest path, between them. The MP is determined by estimating the mean value for all pairs of neighbor pixels that can be part of a possible path connecting the initial given pixels. A MP is computed as the sum of those two geodesic image functions. In general, a MP is obtained with the knowledge of two end pixels. Based on the 2-dimensional spreading of the estimated geodesic time function, the concepts of propagation energy and strong pixels are introduced and tested for the waterline extraction by marking only one pixel in the image.

This paper extends the range of radar remote sensing applications by considering the application of remote sensing radar images for site-specific land clutter modelling. Data fusion plays a central role in our approach, and enables effective combination of remote sensing radar measurements with incomplete information about the Earth's surface provided by optical sensors and digital terrain maps. The approach uses airborne remote sensing radar measurements to predict clutter intensity for different terrain coordinates and utilises an empirical backscattering model to interpolate radar measurements to grazing angles employed by land-based radar sensor. The practical aspects of the methodology application for real-life remote sensing data and generation of a land clutter map of the test site at X-band are discussed.

In last decades many interferometric Synthetic Aperture Radar (SAR) applications have been developed aiming at measuring terrain morphology or deformations. The geodesic information is carried by the interferometric phase. However, this can be observed only in the principal interval giving the so called wrapped phase. To extract the interesting information, e.g., height surface or terrain deformation, the absolute phase should be estimated from the wrapped phase observations. In this work we present an approach to solve PU relying on the local analysis of the wrapped phase signal gradients to recover fringes and phase jumps. The proposed fringe detection algorithm is tested on both synthetic and real data. Synthetic phase surfaces are generated characterized by different signal-to-noise ratio and using a real topographic scene. Real interferograms obtained by processing ENVISAT and TerraSAR-X SAR images are also used to test the above algorithm. First results show that the proposed approach is able to recognize and reconstruct fringes in noisy interferograms.

In this work, the integration between data provided by Time-of-Flight cameras and a multi-image matching technique for metric surveys of architectural elements is presented. The main advantage is given by the quickness in the data acquisition (few minutes) and the reduced cost of the instruments. The goal of this approach is the automatic extraction of the object breaklines in a 3D environment using a photogrammetric process, which is helpful for the final user exigencies for the reduction of the time needed for the drawing production. The results of the performed tests on some architectural elements will be reported in this paper.

The analysis of changes occurred in multi-temporal images acquired by the same sensor on the same geographical area at different dates is usually done by performing a comparison of the two images after co-registration. When one considers very high resolution (VHR) remote sensing images, the spatial information of the pixels becomes very important and should be included in the analysis. However, taking into account spatial features for change detection in VHR images is far from being straightforward, due to effects such as seasonal variations, differences in illumination condition, residual mis-registration, different acquisition angles, etc., which make the comparison of the structures in the scene complex to achieve from a spatial perspective. In this paper we propose a change detection technique based on morphological Attribute Profiles (APs) suitable for the analysis of VHR images. In greater detail, this work aims at detecting the changes occurred on the ground between the two acquisitions by comparing the APs computed on the image of each date. The experimental analysis has been carried out on two VHR multi-temporal images acquired by the Quickbird sensor on the city of Bam, Iran, before and after the earthquake occurred on Dec. 26, 2003. The experiments confirm that the APs computed at different dates show different behaviors for changed and unchanged areas. The change detection maps obtained by the proposed technique are able to detect changes in the morphology of the correspondent regions at different dates regardless their spectral variations.

Currently, there is much interest in developing electro-optic and infrared stationary and moving object acquisition and tracking algorithms for Intelligence, Surveillance, and Reconnaissance (ISR) and other applications. Many of the existing EO/IR object acquisition and tracking techniques work well for goodquality images, when object parameters such as size are well-known. However, when dealing with noisy and distorted imagery many techniques are unable to acquire stationary objects nor acquire and track moving objects. This paper will discuss two inter-related problems: (1) stationary object detection and segmentation and (2) moving object acquisition and tracking in a sequence of images that are acquired via an IR sensor mounted on both stationary and moving platforms. 1. A stationary object detection and segmentation algorithm called "Weighted Adaptive Iterative Statistical Threshold (WAIST)" will be described. The WAIST algorithm takes any intensity image and separates object pixels from the background or clutter pixels. Two common image processing techniques are nearest neighbors clustering and statistical thresholding. The WAIST algorithm uses both techniques iteratively, making best use of both techniques. Statistical threshold takes advantage of the fact that object pixels will exist above a threshold based on the statistical properties of the known noise pixels in the image. The nearest neighbor technique takes advantage of the fact that when many neighboring pixels are known object pixels, the pixel in question is more likely to be a object pixel. The WAIST algorithm initializes the nearest neighbor parameters and statistical threshold parameters and adjusts them iteratively to converge to an optimal solution. Each iteration of the algorithm conservatively declares a pixel to be noise as the statistical threshold is raised. This algorithm has proven to segment objects of interest from noisy backgrounds and clutter. Results of the effort are presented. 2. For moving object detection and tracking we identify the challenges that the user faces in this problem; in particular, blind geo-registration of the acquired spatially-warped imagery and their calibration. For moving object acquisition and tracking we present an adaptive signal/image processing approach that utilizes multiple frames of the acquired imagery for geo-registration and sensor calibration. Our method utilizes a cost function to associate detected moving objects in adjacent frames and these results are used to identify the motion track of each moving object in the imaging scene. Results are presented using a ground-based panning IR camera.

In this article, we de-blur one of the out of focus image among several multispectral (MS) remote sensing images by total variation method. The no blur images are used as priors in the restoration of the out of focus image. Although the distributions of the pixel intensity of the multimodal image of different CCD sensors are greatly different form each other, the directions of their edges are very similar. Then, these similar structures and edge information are used as the important priors or constraints in the total variation image restoration. The steps are: first, the PAN (panchromatic) image is denoted approximately as the weighted sum of all the bands of MS images, and the weight parameters of the relationship between the PAN image and the MS images are computed by least square method; Second by the relationship and the weight parameters, an initial estimation of the out of focus image is calculated; third, the total variation image restoration is local linearized by fixed point iterative method; fourth, the initial estimation for the out of focus image in the third step is brought to the fixed point iteration. At last, by introduce the new priors from the relationship between MS and PAN image, the new total variation image restoration frame is constructed. The edge and gradient information from the no blur images of other channels make the total variation regularization better suppress the noise in de-convolution. The comprehensive experiments are done by using different images with different level of noise. The higher PSNR is acquired by proposed method when it is compared with some other state of art methods. Experiments confirm that the algorithm is very effective especially when the noise in blur remote sensing image is relative large.

Three different methods to realize point correspondences in 3D measurement systems based on fringe projection are described and compared concerning accuracy, sensitivity, and handling. Advantages and disadvantages of the three techniques are discussed. A suggestion is made to combine the principles in order to achieve an improved completeness of the measurements. The principle of a virtual image point raster which is the basis of the combination of the methods is explained. A model to describe the random error of a 3D point measurement for the three methods is established and described. Simulations and real measurements confirm this error model. Experiments are described and results are presented.

The modeling and simulation of SAR clutter is important for Radar system design and signal processing. This paper deals with this problem in high-resolution ground clutter. Based on the theory of clutter modeling, comparisons between different probability distribution function (PDF) models and autocorrelation function (ACF) models are made, and the best model are found taking TerraSAR-X data as example. Finally simulation is made through Memoryless Non-Linear Transformation (MNLT) method based on this best fitting model. The result reveals that a mix of Rayleigh and LogNormal distribution can fit homogeneous, heterogeneous and extremely heterogeneous clutter regions. The simulation shows that MNLT method works well, and PDF and ACF model can represent SAR clutter well based on the homogeneity of texture.

This paper presents a new method based on Semantic Structure Tree (SST) for remote sensing image segmentation, in which, the semantic image analysis is used to construct the SST of the image. The leaves of the SST represent the semantics of the image and serve as human semantic understanding of the image. The root of the tree is the whole image. The SST uses grammar rules to construct a hierarchy structure of the image and gives a complete high-level semantics contents description of the image. Experimental results show that the tree can give efficient description of the semantic content of the remote sensing image, and can be well used in remote sensing image segmentation.

The optical joint transform correlator (JTC) is an effective way to measure the image motion that is hard to be eliminated for space camera. In this manuscript, the principle of JTC is briefly introduced and then a static experiment is designed to demonstrate the suitability of JTC for space camera application. The results demonstrate that the RMS (root-mean-square) error of motion determination can be controlled below 0.15 pixels in both the horizontal and vertical direction, which is good enough to satisfy the space camera requirements. Besides that, the performance of two methods used to locate the cross-correlation peak is also evaluated. Compared with the traditional centroid computing method, Intensity Weighting Centroiding (IWC) is superior because it can not only reduce the sensitiveness of cross-correlation peak to window size selection, but also improves measurement accuracy.

Receiving accurate and comprehensive knowledge about the conditions of roads after earthquake strike are crucial in finding optimal paths and coordinating rescue missions. Continuous coverage of the disaster region and rapid access of high-resolution satellite images make this technology as a useful and powerful resource for post-earthquake damage assessment and the evaluation process. Along with this improved technology, object-oriented classification has become a promising alternative for classifying high-resolution remote sensing imagery, such as QuickBird, Ikonos. Thus, in this study, a novel approach is proposed for the automatic detection and assessment of damaged roads in urban areas based on object based classification techniques using post-event satellite image and vector map. The most challenging phase of the proposed region-based algorithm is the segmentation procedure. The extracted regions are then classified using nearest neighbor classifier making use of textural parameters. Then, an appropriate fuzzy inference system (FIS) is proposed for road damage assessment. Finally, the roads are correctly labeled as 'Blocked road' or 'Unblocked road' in the road damage assessment step. The proposed method was tested on QuickBird pan-sharpened image of Bam, Iran, concerning the devastating earthquake that occurred in December 2003. The visual investigation of the obtained results demonstrates the efficiency of the proposed approach.

Assessing the damage caused by natural disasters requires fast and reliable information. Satellite imagery, especially high-resolution imagery, is recognized as an important source for wide-range and immediate data acquisition. Disaster assessment using satellite imagery is required worldwide. To assess damage caused by an earthquake, house changes or landslides are detected by comparing images taken before and after the earthquake. We have developed a method that performs this comparison using vector data instead of raster data. The proposed method can detect house changes without having to rely on various image acquisition situations and shapes of house shadows. We also developed a houseposition detection method based on machine learning. It uses local features including not only pixel-by-pixel differences but also the shape information of the object area. The result of the house-position detection method indicates the likelihood of a house existing in that area, and changes to this likelihood between multi-temporal images indicate the damaged house area. We evaluated our method by testing it on two WorldView-2 panchromatic images taken before and after the 2010 earthquake in Haiti. The highly accurate results demonstrate the effectiveness of the proposed method.

The dust on the camera's lens will leave dark stains on the image. Calibrating and compensating the intensity of the stained pixels play an important role in the airborne image processing. This article introduces an automatic compensation algorithm for the dark stains. It's based on the theory of flat-field correction. We produced a whiteboard reference image by aggregating hundreds of images recorded in one flight and use their average pixel values to simulate the uniform white light irradiation. Then we constructed a look-up table function based on this whiteboard image to calibrate the stained image. The experiment result shows that the proposed procedure can remove lens stains effectively and automatically.

Extraction of urban land-use information is base step of urban change detection. However, challenges remain in automatic delineation of urban areas and differentiation of finer inner-city land cover types. The extraction accuracy of built-up area is still unsatisfactory. This is mainly due to the heterogeneity nature of urban areas, where continuous and discrete elements occur side by side. Another reason is the mixed pixel problem, which is particularly serious in an urban environment. The built-up areas in arid areas may confuse with nearby bare soil and stony desert, which present very similar spectral characteristics as construction materials such as concrete, while they are often surrounded by farmland. This study focuses on improving urban land use and land cover classification approach in typical city of China's west arid areas using multi-sensor data. Pixel-based classification of the NDBI and Maximum Likelihood Classification (MLC) and object-oriented image classification were used in the study and the classification dataset including Landsat ETM (1999), CBERS (2005), and Beijing-1 (2006). The accuracy is assessed using high-resolution images, aerial photograph and field investigation data. The traditional pixel-based classification approach typically yield large uncertainty in the classification results. Object-oriented processing techniques are becoming more popular compared to traditional pixel-based image analysis.

A methodology to highlight changes in the landscape based on satellite image classification has been developed involving unsupervised and supervised approaches. With past acquisitions, ground truth data are in general not known, therefore the classification can only be unsupervised. These classifications provide labels but not surface types. The main difficulty lies in the interpretation of these classes. An automatic interpretation method has been developed to allocate semantics to classes thanks to a radiometric value catalogue. However, it requires radiometrically comparable images. After radiometric correction, the images are not free from defects; this is why a normalization method has been developed. We propose a specific methodology to evaluate changes consisting in regrouping classes of the same theme, smoothing and eroding contours without taking "mixels" into account and comparing the classified images to provide statistics and image changes. The different steps of the process are essential to avoid false changes and to quantify land cover change with a high degree of accuracy. Various statistical results are given: changes or no changes, types of changes, and crop rotations over several years. Land use /cover change (LUCC) can provide an estimate of carbon capture and storage. Reforestation, changing land use and best practices can increase carbon sequestration in biomass and soils for a period of several decades, which may constitute a significant contribution to the fight against the greenhouse effect. Deforestation, conversely, can lead to significant levels of CO2 emission. By application to the South-West region of Toulouse, we observe significant land cover changes over 11 years (1991- 2002). The crop rotations are given for 4 years (year per year 2002-2005).

This paper is a part of developing a program that generates IR images by considering the surface emitted radiance from objects. The spectral radiance received by a remote sensor is consisted of the self-emitted component directly from the target surface, the reflected component of the solar irradiance at the target surface, and the scattered component by the atmosphere without ever reaching the object surface. The radiance self-emitted from a surface can be calculated by using the temperature and optical property of the surface together with the spectral atmospheric transmittance. The thermal modeling is essential for identifying objects on the scenes obtained from the satellites. And the temperature distribution on the object is necessary to obtain their infrared images in contrast to the background. We considered the composite heat transfer modes including conduction, convection and spectral solar radiation for objects within a scene to calculate the surface temperature distribution. The object considered is assumed to be consisted of several different materials with different properties, such as conductivity, absorption, density, and specific heat etc. We get the weather conditions (air temperatures, wind directions, wind velocity, relative humidity and atmospheric pressure), solar irradiance and surface temperatures of the test plates. The measured diurnal emitted radiance from the test plate with several different surface properties are fairly well compared with the modeled results obtained from the software developed in this study.

The fraction of surface covered by green vegetation is an essential biophysical parameter for addressing land surface processes in the terrestrial climate system. This parameter has been used in many agronomic, ecological and meteorological applications, e.g. for modelling weather predictions, estimation of urban vegetation abundance, etc. In this work, we collect several measures of vegetal cover indexes and relate them to the reflected light signal. The aim is to estimate vegetal cover from hyperspectral images using Support Vector Machine regression of vegetal cover in which every pixel represents an index of this measure from hyperspectral data.

In this study SPOT-panchromatic image with 10 m spatial resolution was fused with ASTER-band ratio images with 30 m spatial resolution. The fusion of SPOT image with ASTER band-ratio data using PC, Brovey, HPF and IHS transform techniques proved to be excellent for both lithological and structural mapping as it preserves the spectral information of ASTER and SPOT data. By visually comparing of these data fusion, the HIS and CNT, methods produce high color distortion with respect to the original image, while it preserve a perfect spatial resolution. The PCA fusion method produce very low color distortion but it dose not preserve all the spatial information. The HPF fusion method produce very low color distortion as well as preserve all the spatial information which look sharper than the other images. This study revealed that, the HPF fusion method looks the best method comparing with the other methods in terms of the quality of spectral and spatial information. By quantitatively analyzed using the correlation coefficient, The CC is ranging from 0.406 to 0.455 using HIS fusion method. While by using Brovey transform, the CC ranging from 0.955 to 0.988. Wherever, The CC between the multispectral input data and the output fused image is ranging from 0.988 to 0.996 using automatic PCA fusion technique. By using manual PCA fusion technique, the CC is ranging from 0.978 to 0.997, so there is no big different between the automatic and the manual PCA methods. The best CC between the multispectral input data and the output fused image is ranging from 0.989 to 0.999 using HPF fusion technique.

It is expected, that ship detection and classification in SAR satellite imagery will be part of future downstream services for various applications, e.g. surveillance of fishery zones or tracking of cargo ships. Due to the requirements of operational services and due to the potential of high resolution SAR (e.g. TerraSAR-X), there is a need for composing, optimization, and validation of specific fully automated image processing chains. The presented processing chain covers all steps from land masking, screening, object segmentation, feature extraction to classification and parameter estimation. The chain is base for experiments with both open sea and harbor scenes for ship detection and monitoring. Within this chain, a classification component for SAR ship and non-ship decision is investigated. Based on many extracted image features and numerous image chips for training and test, some promissing results are presented and discussed. Since the classification can reduce the false alarms of the screening component, the processing chain is expected to work on images with less good weather and signal conditions and to extract ships with lower reflexions.

In this article, we apply the total variation method to remote sensing image restoration. A new method to calculate the regularization parameter by using an improved Generalized Cross-Validation (GCV) method is proposed. Classical GCV can not be directly used in total variation regularization due to the nonlinearity of the total variation. In our method, the GCV method and the fixed point iterative method are combined. In order to use the GCV method, we separate a new linear regularization operator from the definition of the fixed point iterative method. Based on the linear regularization operator, we change the form of the classical GCV function and make it suitable for total variation regularization. A new GCV function suitable for total variation regularization is constructed. By using the new GCV method, the regularization parameter is automatically changing in total variation remote sensing image restoration and a higher signal-noise-ratio is acquired. Experiments confirm that the adaptability and the stability of the total variation remote sensing image restoration are improved.