Proceedings Volume 4555

Neural Network and Distributed Processing

Xubang Shen, Jianguo Liu
cover
Proceedings Volume 4555

Neural Network and Distributed Processing

Xubang Shen, Jianguo Liu
View the digital version of this volume at SPIE Digital Libarary.

Volume Details

Date Published: 20 September 2001
Contents: 3 Sessions, 29 Papers, 0 Presentations
Conference: Multispectral Image Processing and Pattern Recognition 2001
Volume Number: 4555

Table of Contents

icon_mobile_dropdown

Table of Contents

All links to SPIE Proceedings will open in the SPIE Digital Library. external link icon
View Session icon_mobile_dropdown
  • Neural Network
  • Hardware Parallel and Distributed Processing
  • Poster Session
Neural Network
icon_mobile_dropdown
Handwritten character recognition based on hybrid neural networks
Peng Wang, Guangmin Sun, Xinming Zhang
A hybrid neural network system for the recognition of handwritten character using SOFM,BP and Fuzzy network is presented. The horizontal and vertical project of preprocessed character and 4_directional edge project are used as feature vectors. In order to improve the recognition effect, the GAT algorithm is applied. Through the hybrid neural network system, the recognition rate is improved visibly.
Online graphic symbol recognition using neural network and ARG matching
Bing Yang, Changhua Li, Weixing Xie
This paper proposes a novel method for on-line recognition of line-based graphic symbol. The input strokes are usually warped into a cursive form due to the sundry drawing style, and classifying them is very difficult. To deal with this, an ART-2 neural network is used to classify the input strokes. It has the advantages of high recognition rate, less recognition time and forming classes in a self-organized manner. The symbol recognition is achieved by an Attribute Relational Graph (ARG) matching algorithm. The ARG is very efficient for representing complex objects, but computation cost is very high. To over come this, we suggest a fast graph matching algorithm using symbol structure information. The experimental results show that the proposed method is effective for recognition of symbols with hierarchical structure.
New wideband radar target classification method based on neural learning and modified Euclidean metric
Yicheng Jiang, Ping Cheng, Yangkui Ou
A new method for target classification of high-range resolution radar is proposed. It tries to use neural learning to obtain invariant subclass features of training range profiles. A modified Euclidean metric based on the Box-Cox transformation technique is investigated for Nearest Neighbor target classification improvement. The classification experiments using real radar data of three different aircraft have demonstrated that classification error can reduce 8% if this method proposed in this paper is chosen instead of the conventional method. The results of this paper have shown that by choosing an optimized metric, it is indeed possible to reduce the classification error without increasing the number of samples.
Improved pulse-coupled neural network for target segmentation in infrared images
This paper presents a new image segmentation algorithm based on the pulse coupled neural network (PCNN) and histogram method for infrared images. The proposed algorithm abandons entirely the mechanism of the time exponential decaying function and uses the results of the gray-level histogram analysis as the interior thresholds of PCNN, meanwhile, it keeps the advantage of briding small spatial gaps and minor intensity variations. Experiment results demonstrate that the proposed algorithm can get more complete region and edge information in infrared images. It is also of much lower complexity and of high speed than the original one.
Ultrasound Doppler tissue image analysis based on neural network
Shukui Zhao, Deyu Li, Lixue Yin, et al.
A new method for quantitative analysis of ultrasound Doppler tissue images (DTI) has been developed based on a neural network. The method aims to extract numerical data of velocity or acceleration from DTI images and analyze them quantitatively. A three-layered back propagation (BP) neural network is used to accomplish this task. The input of the network is the differences between the red, green and blue components of pixels and the output is the acceleration or velocity values. The network is trained with the color bars in the DTI images. The result of analyzing the movement of the left ventricle anterior free wall (LVAW) from DTA (DTI acceleration mode) image sequences is presented. The result of time-acceleration curve is highly correlated with the electrocardiogram (ECG) curve and gives us a quantitative and graphic description of the ventricle movement in cardiac cycles. It shows the movement characteristics of the left ventricle in cardiac cycles and also shows the excitation differences among the three layers of the myocardium. It is demonstrated that the method has great potential to characterize myocardial movement, which may provide a new way to characterize cardiac activities.
Bayesian networks for mapping salinity using multitemporal Landsat TM imagery
Dongming Huo, Jingxiong Zhang, Jiabing Sun, et al.
Bayesian networks are used for reasoning under uncertainty. This paper examines the use of Bayesian networks for integrating multi-temporal remotely sensed data with landform data derived from digital elevation models (DEM) and groundwater data to produce maps showing areas affected by salinity in the Yellow River Delta of China. Incorporating prior knowledge about the relationships between input attributes and their relationship with salinity, a conditional probabilistic network is used to impose a known relationship between input attributes and salinity status. The results are compared with maximum likelihood classification techniques using single-date Landsat TM imagery. They show a large improvement on the maximum likelihood classifier. The network is used to produce a time-series of landcover and salinity maps for the Yellow River Delta.
Obstacle detection for mobile vehicle using neural network and fuzzy logic
Huaijiang Sun, Jingyu Yang
In our mobile vehicle project, sensors for environment modeling are a CCD color camera and two line-scan laser range finders. The CCD color camera is used to detect road edges. The two line-scan laser range finders are used to detect obstacles. Only two line-scan laser range finders increase processing speed, but there are blind zones for low obstacles, especially near the vehicle. In this paper, neural network and fuzzy logic are used to cluster and fuse obstacle points provided by two line-scan laser range finders. There is an assumption that obstacles missed by laser radar in some instant must be detected previously. A circle Adaptive Resonance neural network algorithm is used to incrementally cluster obstacle points provided by laser range finders into candidate obstacles. Every candidate obstacle is expressed by a circle, and is assigned a belief by a fuzzy logic system. Inputs of the fuzzy logic system are radius and number of points. Fuzzy rules are provided by human and can be fine-tuned with training data. The final true obstacle is the nearest one chosen from candidate obstacles whose beliefs exceed a threshold. Experiment results indicate that our mobile vehicle can safely follow road and avoid obstacles.
Detection of microcalcifications ROI in digital mammograms using two stages of neural networks
Yang-suk Lee, Seung-Chul Lim, Dong-Sun Park
In this paper, we present an efficient algorithm to detect microcalcifications ROI (Regions of Interest) in digital mammograms using two stages of neural networks. To efficiently detect microcalcifications ROI, we used four sequential processes; preprocessing for breast area detection, modified multilevel thresholding, ROI selection using simple thresholding filters and final ROI selection with two stages of neural networks. In modified multilevel thresholding, the shape property of microcalcification resulted from the gray-level difference with surroundings is used. This algorithm separates microcalcifications from tissues by applying the half-toning technique for different gray-levels. The first selection process with simple thresholding filters defines the filter parameters using the statistically extracted shape property and then it eliminates tissues, which are obviously recognized, to reduce the processing overhead in the next step. The final selection process using neural networks is to detect the ROI in two steps. Through the two stages of neural networks, ROIs with microcalcifications are selected. Each neural network compares and analyzes recognition performance after training. The ROI detection method for microcalcification used in this paper is the first stage for a CAD system. The designed ROI detection methods efficiently find 98.06% of with microcalcifications.
Image registration and fusion of PET and MRI images using neural network
Weifu Wang, Frank Q.H. Ngo, Jyh-Cheng Chen
Multimodality image registration and fusion are widely used in clinical diagnosis and treatment planning because combining information from different modalities can offer more information than single modality. Several image fusion techniques have been developed for these purposes. Neural network technique is commonly used in image or pattern recognition, but up until now there are very few studies on image fusion using neural network. Magnetic resonance imaging (MRI) is an anatomical imaging with high spatial resolution while positron emission tomography (PET) provides biochemical and physiological information but with poor spatial resolution. In this paper, we present a neural network approach for the registration and fusion of PET and MRI images. In our study, we use a multi-layer backpropagation neural network to train spatial characteristic points and to obtain translational range and rotational angles. After image registration and fusion using this method, we show that fused image with transformation has better biochemically consistent result than the one without transformation. Since the method relies on anatomic information in the images rather than on external fiducial markers, ti can be applied retrospectively.
Nonlinear dynamic system modeling based on neural state space model
Yongji Wang, Qing Wu, Hong Wang
In this paper, an approach of nonlinear system modeling based on neural state space model is proposed. The neural state space model is of the quasi-linear characteristics of system, therefore, many linear system controller design approach can be extended to apply to the NNSP models. The EKF approach is adopted for parameter identification of neural state space models and a High-order correction method is then applied to test the validity of the neural state space model of nonlinear systems. The application of this method to dynamic modeling of typical chemical processes shows that the presented approach is effective.
Histogram-based image classification using fuzzy ARTMAP neural network
Chunhong Jiang, Zhe Chen
This paper presents a novel fast images classification method which based on image histogram features and using fuzzy ARTMAP neural network. Compared with the previous method, the edge, brightness, contrast and SNR feature of images are taken into account in this method, and it has so many advantages such as self-adaptive clustering, fast convergence, good real-time ability, high classification accuracy and high universality etc. It can be adopted in SMGS (scene matching guidance system) to auto-select real- time images and so as to improve the level of intelligence, reliability and real time ability of SMGS.
Recognition of lymph node metastasis malignancy tumor cells based on the multilayer error back-propagation (BP) neural network
Ling Kong, Chunping Liu, Peihua Shen, et al.
In the paper, one multi-layer BP neural network is applied to identify metastasis malignancy tumor cells in lymph node puncture image. The topology structure of the network is as following: the node number of input-layer is 9, which involves morphologic features and chroma information; the node number of first hide-layer and second hide-layer is defined respectively as 8 and 12; the node number of output-layer is 3, which is the category number of recognized objects. Experimental results show that the learning performance of multi-layer BP neural network is good, comparing with three-layer BP neural network, the recognition rate is improved, and the method can be as an assistant means to recognize lymph node metastasis malignancy tumor cells.
Hardware Parallel and Distributed Processing
icon_mobile_dropdown
Multiprocessor scheduling problem with machine constraints
Yong He, Zhiyi Tan
This paper investigates multiprocessor scheduling with machine constraints, which has many applications in the flexible manufacturing systems and in VLSI chip design. Machines have different starting times and each machine can schedule at most k jobs in a period. The objective is to minimizing the makespan. For this strogly NP-hard problem, it is important to design near-optimal approximation algorithms. It is known that Modified LPT algorithm has a worst-case ratio of 3/2-1/(2m) for kequals2 where m is the number of machines. For k>2, no good algorithm has been got in the literature. In this paper, we prove the worst-case ratio of Modified LPT is less than 2. We further present an approximation algorithm Matching and show it has a worst-case ratio 2-1/m for every k>2. By introducing parameters, we get two better worst-case ratios which show the Matching algorithm is near optimal for two special cases.
Design of a parallel RISC image processor based on PCI bus
Xianyang Jiang, Xubang Shen, Tianxu Zhang
Low-level image processing operations usually involve simple and repetitive operations over the entire input images, thus image processor may communicate with the memory system or each other frequently, hence the image processor would provide high throughput rate. In this article we present an architectural design and analysis of a parallel RISC image processor. The processor was based on PCI bus to speed up a range of image processing operations. The other characteristic of the processor is that a new three-port hostbridge is integrated into the processor. The implementation of commonly used image processing algorithms and their performance evaluation are also discussed.
Efficient method for hardware-based DCT/IDCT implementation
Xiantao Sun, Chengke Wu
Discrete Cosine Transformation (DCT) and Inverse Discrete Cosine Transformation (IDCT) are important parts of many image and video compression system. Unfortunately these operations are extremely computation-intensive in a coding system, it consumes a large amount of resources for computation, especially in a real-time video coding system. In this article an efficient method for hardware based DCT/IDCT implementation is proposed. We combine the vector processing with parallel processing using Distributed Arithmetic. At the same time the processing elements are pipelined to increase the processing speed and reduce the computation latency, which can also reduce the resource requirement and thus enhance the efficiency.
Real-time infrared target tracking system design and research based on DSP
Xianghui Xu, Ran Tao, Yue Wang
Based on the Digital Signal Processor (ADSP21060), a principal and subordinate structure parallel real time infrared target tracking system is established. The system is composed of image acquisition sub system, image processing sub- system, manual control subsystem and target indicates subsystem. A new fast two-dimensional (2-D) entropy algorithm and motion target detection algorithm is implemented in second DSP to detect infrared target. The mass centered tracking algorithm and correlative-tracking algorithm are used to tracking targets. A Kalman filter algorithm is used to predicting approach. And using ADSP21060 to achieve the Kalman filtering algorithm and satisfies the real-time need. In this tracking system the tracking state of infrared target is considered, detecting algorithm or tracking algorithm can be selected automatic. If tracking system is working in detecting mode and target is detected, then system will turn to tracking mode .If system is working in tracking mode and target is lost, then system will turn into detecting mode. As the ADSP-21060 offers powerful features to multi-processing DSP systems, it is easy to expand to multi subordinate DSP system. It is possible to use more complexes detecting or tracing algorithms in real time tracing system.
Two-dimensional mesh-connected parallel processor with complex processing elements
Chaoyang Chen, Xubang Shen, Zhong Wang, et al.
LS MPP is a massively parallel processor .It has fine-grained parallelism with up to 4096 processing elements arranged in a SIMD architecture .The processing elements are arranged in 64x64 two-dimensional mesh-connected array for low-level image processing .In this paper, the system architecture ,the components of processing element ,array controller ,memory organization of LS MPP processor are described .In the final ,we have discussed the performance of LS MPP.
Highly scalable interconnection network for parallel image processing
Hongyu Wang, Weikang Gu
In this paper, we introduce a new hierarchical interconnection network for massively parallel systems, named Fully Connected Cubic Network (FCCN). FCCN is able to emulate the popular Hypercube. FCCN has a constant nodal degree of 4 and it therefore eliminates the problem of large fanout in Hypercube. Moreover, the constant degree is an important requirement for efficiently fabricating an architecture in parallel image processing. FCCN is also a highly scalable architecture in that the existing links remain intact when new nodes are introduced. FCCN is maximally fault tolerant, and it enjoys reasonably low diameter, growth of the number of links and average internodal distance. At last, FCCN is used for parallel image processing system for interconnection. The computation results show that FCCN is a high efficient interconnection network for parallel image processing.
Poster Session
icon_mobile_dropdown
Hierarchical map-matching algortihm for quadtree image on MPP
Guangyuan Fu, Dong Miao, Wenjun Zhang, et al.
The problem of Map-matching is one key problem in the field of aircraft and vehicle guidance. It deals with the technologies of remote sensing, computer vision, image processing and pattern recognition, etc. Researchers are focusing on how to improve the system's performance, to reduce the searching times and error matching probability. With using an improved quadtree image representative method and the idea of the sequential similarity detection algorithm (SSDA), a hierarchical map-matching algorithm based on embedded MPP system is designed in this paper. The algorithm can greatly reduce matching times and improve locate accuracy.
Implementation of AHB interface in a communication-specific CPU based on ARM core
Xianyang Jiang, Hu Chen, Xubang Shen, et al.
The characteristics of design of embedded CPU based on IP core are analyzed, then provides the design hole problems existing in design of AHB interface in an ARM core based CPU .By modification on AHB protocol, hole-avoiding methods are presented. In the end, hardware implementation of AHB interface is discussed.
Pattern recognition of internal structural defects in industrial radiographic testing based on neural network
Ming Ming, Zheng Li
It is shown that an artificial neural network can be used to classify internal structural defects in radiographic nondestructive testing. We design a series of images presenting phantoms to simulate three different classes of defects: porosity, crack, and slag. Features of these defects are selected from domains of geometry, gray statistics, frequency spectrum, and etc. Some of them are especially suitable for pattern recognition in the case of radiographic image. A three-layered neural network trained with back-propagation rule is developed to carry out the classification. The training and testing data for the net are the features extracted from digitized radiographic images. Results are presented with satisfactory recognition rate.
Comparison of neuron selection algorithms of wavelet-based neural network
Xiaodan Mei, Sheng-He Sun
Wavelet networks have increasingly received considerable attention in various fields such as signal processing, pattern recognition, robotics and automatic control. Recently people are interested in employing wavelet functions as activation functions and have obtained some satisfying results in approximating and localizing signals. However, the function estimation will become more and more complex with the growth of the input dimension. The hidden neurons contribute to minimize the approximation error, so it is important to study suitable algorithms for neuron selection. It is obvious that exhaustive search procedure is not satisfying when the number of neurons is large. The study in this paper focus on what type of selection algorithm has faster convergence speed and less error for signal approximation. Therefore, the Genetic algorithm and the Tabu Search algorithm are studied and compared by some experiments. This paper first presents the structure of the wavelet-based neural network, then introduces these two selection algorithms and discusses their properties and learning processes, and analyzes the experiments and results. We used two wavelet functions to test these two algorithms. The experiments show that the Tabu Search selection algorithm's performance is better than the Genetic selection algorithm, TSA has faster convergence rate than GA under the same stopping criterion.
IP core design of template matching algorithm in image processing
Quanqing Zhu, Xuecheng Zou, Zhenzhong Dong, et al.
This paper presents the design and implementation of template matching IP cores for image processing. Enhanced Moment Preserving Pattern Matching (MPPM) algorithm of template matching was adopted for efficient hardware implementation. The cores were coded in Verilog HDL for modularity and portability. The IP cores were validated in a XC4052XL FPGA and XESS XS40 prototyping board.
High-speed aerial image processing system based on DSP
Haiju Lei, Dehua Li, Hanping Hu, et al.
This paper introduces a high-speed parallel system based on DSP to process aerial image. The system is of the master-slave architecture. The master system is composed of general computer. Its slave system based on high speed DSP TMS320C6201 array can process real time aerial image concurrently. The data communicates between the master system and the slave system through bus interface. In this thesis, on the basis of the fundamental characters of digital image processing, we adopt domain decomposition method suiting digital image processing. The mode can parallelize serial digital image-processing algorithms which process digital images on spatial domain. This algorithm applies in many image algorithms and can be realized easily. At the same time it ensures every DSP load balance in the course of processing aerial image. Experiments show the system is effective, simple and steady.
VLSI architecture for the Hadamard-transform-based fast VQ encoder
Shurong Cheng, Zhe-Ming Lu, Xiamu Niu
Vector Quantization (VQ) is an efficient image compression technique. In this paper, a new VLSI architecture for Vector Quantization (VQ) encoding based on the Hadamard Transform (HT) domain with the partial distance search (PDS) technique is proposed. The PDS algorithm is a simple and efficient algorithm, which allows early termination of the distortion calculation between an input vector and a codeword by introducing a premature exit condition in the search process. By using a codeword elimination criterion based on MSE in the Hadamard transform, presorted codebook and nearest search method, a large number of codewords can be rejected before computing MSE while the image quality remaining unchanged compared to the full-search VQ encoder. The proposed fast codeword search algorithm can reduce computation and is easier to be implemented by VLSI technology. Experimental results demonstrate the effectiveness of the proposed VLSI architecture.
Module of network RAID implemention on cluster computing
Ke Zhou, Jiangling Zhang
Network RAID is a good idea in mass storage field. It can reduce response time and increase data transferring speed of storage system, especially to network application. But it is a problem that how to use network RAID to improve the performance of parallel processing system. It is current to use cluster architecture to develop high performance parallel computer. Cluster is a set of complete computers that can run one task at the same time through a high-speed network. The normal cluster belongs to non-shared architecture. This architecture makes cluster consume most of time on communicating with each other. Network RAID has its natural characteristic that is connected to network directly. Using network RAID in cluster makes a disk-shared architecture of cluster. There are two modules to use network RAID in cluster, one is using network RAID as a shared storage space, the other is using network RAID as a private storage space. As a shared storage space, network RAID records the public data that will be used by some nodes. As a private storage space, each node is provided with a segment of data space in network RAID. The other important problem is how the node of cluster communicates with network RAID. Because network RAID has two channels for network communication, one is command channel, the other is data channel, and nodes must be modified to adapt this communication module. We use TCP/IP protocol, and define some data transferring rules in application level. At last, it must be paid more attention to performance evaluation of cluster system. In normal cluster architecture, load is distributed into each node. But in disk-shared cluster architecture, some part of load may be assigned to network RAID. How to evaluate the performance is due to the part of load assigned to network RAID. This paper discusses above three problems, that are using module, communication module and performance, and provides a new disk-shared architecture of cluster that uses network RAID.
Parallel algorithm implementation of MPEG-4 video decoder on DSP
Dongmei Li, Zhaohui Li
MPEG-4 is an international coding standard that aims at providing standardized core technologies allowing efficient storage, transmission and manipulation of video data in multimedia environments. As mobility has become one of the key requirements of the information society today, the next generation mobile will be a service-oriented industry, capable of delivering rich multimedia content, especially streaming video, to the palms of mobile subscribers. MPEG-4, with its superior compression, interactivity and systems capabilities, is the most promising future standard. And that small package high performance high speed DSPs are very suitable to be used in portable devices. This paper describes the implementation of MPEG-4 SVP(simple visual profile) video decoder on the TMS320C6201 DSP which is on the 'C6x EVM evaluation module(EVM) . The Texas Instruments TMS320C62x devices are fixed-point DSPs that feature the VelociTiTM architecture, which is a high-performance, advanced, very-long-instruction-word (VLIW) architecture. With this architecture, a high degree of parallelism can be exploited to meet real-time requirements of video processing such as compression and decompression. In this paperú¼ different aspects of the decoder , the decoding algorithm, decoder structure and memory requirements are discussed.
Unification of support vector machines and soft computing paradigms for pattern recognition
Ying Li, Licheng Jiao
This paper analyzes support vector machines (SVMs) and several commonly used soft computing paradigms for pattern recognition including neural and wavelet networks, and fuzzy systems. Bayesian classifiers, fuzzy partitions, etc and tries to outline the similarities and differences among them. Support vector machines provide a new approach to the problem of pattern recognition with clear connections to the underlying statistical learning theory. We try to bring SVMs into the framework of the unification paradigm called the weighted radial basis function paradigm. Unifying different classes of methods has enormous advantages, such as the ability to merge all such techniques within the same system. It is hoped that this paper would provide theoretical guides for the study and applications of support vector machine and soft computing paradigms.
Knowledge-based artificial neural network and the application of it in understanding remotely sensed images
Chunxiang Wu
The Artificial Neural Network (ANN) is an intelligent computer system bases on the empirical learning of the human being. Knowledge-Based Artificial Neural Networks (KBANN) effectively combines the knowledge learnt from theory with that of learnt from examples. This efficient combination of theory and data may result in efficient learning system. And the method of building a KBANN solves the problem that how to design the structure of the neural network. According to the way of building a KBANN, interpreting system of Remotely-Sensed images can be built.