Automation of multichannel image processing
Remote sensing (RS) by airborne and spaceborne instruments provides useful information for agriculture, forestry, hydrology, ecology, and other applications. Almost all modern RS sensors provide a great deal of data collected onboard and then transmitted via downlink and distributed to customers. This data is often not ready for use and must be processed. If there are many images to be processed, or an image has multiple channels (e.g., it is color, multi- or hyperspectral, or has dual or multi-polarization), it becomes necessary to perform processing such as filtering, lossy compression, restoration, edge detection, and segmentation automatically (that is, blindly).1
Automation saves time and effort in processing acquired RS data. It can also be helpful if processing should be done operatively (for example, when RS data is used to monitor a catastrophe, or pictures are made using digital cameras2) and a qualified expert is not available to choose the correct algorithm and/or set its parameters (e.g., if processing is performed onboard satellites). Automation is also desirable if the a priori information on the characteristics of noise and other distortions is limited and these characteristics can vary within rather wide limits depending upon imaging conditions.
In recent years, researchers have studied the automation of particular operations in multichannel image processing, such as onboard lossy compression of multi- and hyperspectral RS data. However, there is no unified approach to this work. In particular, it is difficult to predict how operations applied in earlier processing stages influence the performance of processing at later stages and how to reach the final goal. A set of methods and algorithms has recently been designed cooperatively by three teams from France, Finland, and Ukraine, respectively, for application in the practical situations mentioned above.
We propose the processing of RS data in several stages, as shown in Figure 1. In the first stage, blind determination of the type of noise and estimation of its statistical and spatial correlation characteristics is performed.1–4 These operations are based on component-wise processing of multichannel RS images using pre-segmentation, detection of homogeneous image regions, and joint robust processing of the data in them. On the basis of the results, other stages of multichannel image processing can be performed, including image pre-processing such as homomorphic transformations, normalization, and sub-band grouping to improve further processing.1, 5,6 An important stage is image filtering (denoising) using locally adaptive transform-based filters, both component-wise and vector (3D),5, 7 where the latter allow the exploitation of inter-band correlations between data. Then, lossy image compression follows at a rather high compression ratio (8 to 30), yielding compressed data of appropriate quality, either visual8 or adequate for classification. Finally, image post-processing is performed, if necessary, followed by classification, interpretation, object detection, and further interpretation.
Two core features of the proposed processing techniques are the discrete cosine transform and robust estimation based on order statistics. Fast algorithms and hardware exist for both of these techniques, enabling high computational efficiency.
Because the designed methods have to perform adequately for images and noise having different properties, our algorithms have been intensively verified. We have done this by noise characteristic estimation in real-life color images,2 test images from the Tampere Image Database 2008,9 and aerial photos (see Figure 2). For most images, the statistical characteristics of the noise were automatically estimated with relative errors of less than 20%. This is good enough to allow the parameters for filtering and compression techniques used in the next stage to be set in order to obtain close-to-optimal efficiency. We have tested the filtering, compression, and classification of data from the LandSat Thematic Mapper, ITRES Compact Airborne Spectrographic Imager, Specim AISA Eagle, Airborne Visible/Infrared Imaging Spectrometer, and Hyperion multi- and hyperspectral RS data1, 4,5,7 and synthetic aperture radar images from different instruments.1, 6
The obtained results have demonstrated that it is possible to make the processing of multichannel images fully automatic, or at least to carry out basic blind data processing at several stages. The main tasks that we will focus on in the future are blind estimation of signal-dependent noise parameters, vector filtering, and compression of multichannel images in the case of signal-dependent noise.
Vladimir Lukin graduated from Kharkov Aviation Institute (now NAU) in 1983 with his diploma with honors in radio engineering. Since then he has been with the Department of Transmitters, Receivers, and Signal Processing (DTRSP) at the university, where he is currently department vice-chairman and professor. He defended his Candidate of Technical Science thesis in 1988 and Doctor of Technical Science thesis in 2002 in digital signal processing for remote sensing. Since 1995, he has also worked in collaboration with Tampere University of Technology, Finland. His research interests include digital signal/image processing, remote sensing data processing, image filtering, and compression.
Sergey Abramov graduated from NAU in 2000 with a diploma with honors in radio engineering. Since then, he has been with the DTRSP, where he is currently associate professor and part-time senior researcher. He defended his Candidate of Technical Science thesis in digital signal processing for remote sensing in 2003. His research interests include digital signal/image processing, blind estimation of noise characteristics, and image filtering.
Nikolay Ponomarenko graduated from NAU in 1995 and received his diploma with honors in computer science. Since then he has been with the DTRSP, where he is currently senior researcher and part-time associate professor. He defended his Candidate of Technical Science thesis in digital signal processing for remote sensing in 2004. He also defended his Doctor of Technology thesis on image compression at Tampere University of Technology in 2005. His research interests include digital signal/image processing, remote sensing data processing, image filtering, and compression.
Mikhail Uss graduated from NAU in 2002 with a diploma with honors in radio engineering. Since then he has been with the Department of Aircraft Radioelectronic Systems Design at the university, where he is currently head of the department. He obtained the Candidate of Technical Science degree in digital signal processing for remote sensing from NAU in 2006 and the PhD degree from University of Rennes 1 (France) in 2011. His research interests include statistical theory of radio-technical systems, digital signal/image processing, blind estimation of noise characteristics, and theory of fractal sets, with applications to image processing and remote sensing.
University of Rennes 1
Benoit Vozel obtained the state engineering degree and the MSc degree in control and computer science in 1991 and the PhD degree in 1994 from École Centrale de Nantes (France). As a PhD student, he was with the Signal Processing Group at Institut de Recherche en Communication et Cybernétique de Nantes, where he worked on the detection of abrupt changes in signals. Since 1995, he has been with the ENSSAT, where he is currently with the signal and multicomponent/multimodal image processing research team (TSI2M) within the Institute of Electronics and Telecommunications of Rennes (IETR). His research interests generally concern blind estimation of noise characteristics, image filtering and restoration, and adaptive image and remote sensing data processing.
Kacem Chehdi received the PhD and ‘Habilitation à diriger des recherches’ degrees in signal processing and telecommunications from the University of Rennes 1, in 1986 and 1992, respectively. At the university, he was an assistant professor from 1986 to 1992, has been a professor of signal and image processing with the ENSSAT since 1993, was the head of the Laboratory of Systems Analysis for Information Processing (LASTI) from 1998 to 2003, and has been the head of the TSI2M in IETR since 2004. His research interests include adaptive processing at every level in the pattern recognition chain by vision, blind restoration, and blind filtering (in particular the identification of the physical nature of image degradation and the development of adaptive algorithms) and segmentation and registration topics (in particular the development of unsupervised, cooperative, and adaptive systems). The main applications currently under investigation are multispectral and hyperspectral image processing and analysis.
Karen Egiazarian received his PhD from Moscow M. V. Lomonosov State University, Russia, in 1986, and Doctor of Technology degree from Tampere University of Technology in 1994. He is a leading scientist in signal, image, and video processing, with about 300 refereed journal and conference articles, three book chapters, and a book published by Marcel Dekker. His main interests are in the field of multirate signal processing, image and video denoising and compression, and digital logic. He is a member of the DSP Technical Committee of the IEEE Circuits and Systems Society.
Jaakko Astola received his BS, MS, Licentiate, and PhD in mathematics from Turku University, Finland, in 1972, 1973, 1975, and 1978, respectively. From 1976 to 1977, he was with the Research Institute for Mathematical Sciences of Kyoto University, Japan. From 1979 to 1987, he was with Lappeenranta University of Technology (Finland). He has been at Tampere University of Technology since 1987 and is currently professor of signal processing and director of the Tampere International Center for Signal Processing, academy professor in the Academy of Finland, and IEEE fellow. His research interests include signal/image processing, statistics, and image coding.
