This PDF file contains the front matter associated with SPIE Proceedings Volume 8005, including the Title Page, Copyright information, Table of Contents, and Conference Committee listing.

Efficient algorithm for blind image deconvolution and its high-speed implementation is of great value in practice. Further optimization of SeDDaRA is developed, from algorithm structure to numerical calculation methods. The main optimization covers that, the structure's modularization for good implementation feasibility, reducing the data computation and dependency of 2D-FFT/IFFT, and acceleration of power operation by segmented look-up table. Then the Fast SeDDaRA is proposed and specialized for low complexity. As the final implementation, a hardware system of image restoration is conducted by using the multi-DSP parallel processing. Experimental results show that, the processing time and memory demand of Fast SeDDaRA decreases 50% at least; the data throughput of image restoration system is over 7.8Msps. The optimization is proved efficient and feasible, and the Fast SeDDaRA is able to support the real-time application.

We propose an adaptive edge detection algorithm for LOG operator based on a biological perspective solving the problem of parameter setting. The algorithm can survive in different kinds of images with different imaging qualities. We introduce the concept of Global Optimal Observation Scale that the best scale parameter for LOG lie at the global observation location in scale space. Experimental results demonstrate strong capacity of the algorithm.

In order to mosaic neighboring and partly overlapping orthophotos of a scene into one large image, the paper proposes a large block orthophoto mosaicking method. In our method, seam lines firstly are delineated through overlap areas among orthophotos according to an optimal geometrical criterion. Then a network of mosaicking is built based on these delineated seam lines, and related topology information may be easily abstracted from the mosaicking network. In the second stage of our method, each seam line is optimized again by a modified snake algorithm. The algorithm makes every seam line meet the requirements of maximum color similarity of the images and maximum texture similarity. In order to searching an optimal seam line in a large overlap area as fast as possible, a hierarchical strategy is adapted. In that way, an optimized path through the overlap area is found, where the color and texture of the two images are similar. The still remaining jumps in hue and the differences in intensity and saturation have to be leveled by smooth interpolation in the vicinity of the seam line. After having processed all overlapping areas, a large block of orthophotos are automatically merged to a final large image.

The W Transforms are widely used in digital signal processing. This paper proposes a novel approach to compute Discrete W Transforms via computation of the first-order moments without multiplications. A scalable and efficient systolic array is designed to implement this approach. Compared with the other existing methods, the proposed algorithms is simpler and more applicable.

In this paper, we present a co-design method for parallel image processing accelerator based on DSP and FPGA. DSP is used as application and operation subsystem to execute the complex operations, and in which the algorithms are resolving into commands. FPGA is used as co-processing subsystem for regular data-parallel processing, and operation commands and image data are transmitted to FPGA for processing acceleration. A series of experiments have been carried out, and up to a half or three quarter time is saved which supports that the proposed accelerator will consume less time and get better performance than the traditional systems.

In this paper a novel MDCT without multiplications has been presented via transforming it into the computation of the first-order moment. Then the method proposed for computing moments has been adapted to implement the arbitrary-length MDCT efficiently. A very simple and scalable systolic array without multipliers and ROMs has also been designed to perform the MDCT, which can result in the efficient VLSI implementation easily. And the comparison with some existing methods shows the superiority of our method.

By analyzing the characteristics of infrared focal plane array image, an improved implementation of infrared focal plane image enhancement algorithm based on FPGA is proposed, with limited FPGA memory resources for gray-scale stretching. Experiment results show that the implementation is easy on FPGA with low FPGA memory without extra memory devices. Moreover, it is flexible and effective for improving gray contrast of the interested region of the image, and proved to meet the requirements of infrared focal plane detector for image enhancement showing great utility value.

Retrieving the parameters in water quality with multispectral data using neural network is increasingly popular, however, the training process with large amount samples and calculation with large-volume data are a time-consuming work. Many emergency pollution events need quick responses for practical use. In this paper, an improved membrane computing strategy is presented. This strategy is a hybrid one combining the framework and evolution rules of P systems with active membranes and neural networks, and it involves a dynamic structure including membrane fusion and division, which helpful to enhance the information communication and beneficial to reduce the computation. Then, a parallel implementation with the training result is discussed. Experiments with Landsat datasets to obtain suspended sediment are carried out to demonstrate the practical capabilities of this introduced strategy.

A special designed VLSI chip for template matching fundamentally used in automatic target recognition is proposed in this paper, it adopts normalized cross correlation algorithm. Parallelism inherent in the operation is explored to reduce the huge needed external bandwidth. As much as 8 large binary templates could be configured into four operation modes of eight 1-bit, four 2-bit, two 4-bit and one 8-bit templates using partial product scheme and they are processed in parallel. It takes 13.23ms to execute 120x160 template matching with 256x320 image, therefore is suitable for real-time applications. The prototype of the chip is emulated on FPGA and also synthesized with Design Compiler, die area is 3mm x 3.1mm and power consumption is 114.1 mw when operate at 108 MHz.

This paper has designed and realized a coarse-grained, unbalanced, modularized parallel embedded software system for ATR. According to the characteristics of ATR algorithms, some control modules such as system monitoring, task assignment and hierarchical algorithm modules are realized in our system. There are different design principles for different modules. The task assignment module combines different modules into clusters based on mutually exclusive modules, and assigns them to different processors. The principle of combination is the minimum variance of load on different processors. The system satisfies the requirement of real-time performance due to this reasonable strategy for task assignment, with the flexibility and scalability significantly improved.

A novel line-based streaming labeling algorithm with its VLSI architecture is proposed in this paper. Line-based neighborhood examination scheme is used for efficient local connected components extraction. A novel reversed rooted tree hook-up strategy, which is very suitable for hardware implementation, is applied on the mergence stage of equivalent connected components. The reversed rooted tree hook-up strategy significant reduces the requirement of on-chip memory, which makes the chip area smaller. Clock domains crossing FIFOs are also applied for connecting the label core and external memory interface, which makes the label engine working in a higher frequency and raises the throughput of the label engine. Several performance tests have been performed for our proposed hardware implementation. The processing bandwidth of our hardware architecture can reach the I/O transfer boundary according to the external interface clock in all the real image tests. Beside the advantage of reducing the processing time, our hardware implementation can support the image size as large as 4096*4096, which will be very appealing in remote sensing or any other high-resolution image applications. The implementation of proposed architecture is synthesized with SMIC 180nm standard cell library. The work frequency of the label engine reaches 200MHz.

In the complex situation of the battlefield, COA (course of action) places an important role. It is required to coordinate many resources in some actions to achieve the desired purpose in the battle. The main goal of COA is to arrange the action in the right order and to put the right resource in the right action. The task which is composed of many actions is always extremely complex. Therefore, COA is actually NP-Hard and a multi-objective optimization problem. It is difficult to solve this problem by common methods. In this paper, a mechanism of co-evolutionary is introduced to solve the problem of COA. It deals well with the problems of resource management and action scheduling.

According to practical organization of disaster relief goods and materials dispatching activities in China, a novel relief goods and materials dispatching model for multi-depot and multi-disaster area is established. This model is based on the multi-level depot priority model and is formalized as a multi-objective optimization problem with the earliest emergency start time and the least number of participated depots. To approach this model, the particle swarm optimization algorithm is adopted. We also integrated this model into the disaster relief goods dispatching application successfully and applied for programming the simulated dispatching plan for Yushu earthquake disaster. The results prove the efficiency of the proposed model and its intelligent solution approach.

Steel code location is the key point to realize billet detection and recognition in production line scene with complex illumination. However, due to high temperature and complex scene in the rolling line, the steel code location at the end of billet is quite different from optical character location with simple background and vehicle license plate location. In the process of billet detection and recognition, how to determine steel code target location at the end of billet from the complex illumination scene is first necessary in steel intelligent recognition system. In order to solve this problem, a novel method for steel code location is proposed in this paper. First of all, production line scene image is restrained by Mean Shift filtering and iterative segmentation filter, and then candidate character region can be found by clustering character connected domain with same features. At last, the quantitative model is established for candidate region and the statistical decision algorithm can be used to complete screening object region. The experimental results show that the proposed location method is very precise in most different scenes.

The segmentation of sperm image exerts a profound influence in the analysis of sperm morphology, which plays a significant role in the research of animals' infertility and reproduction. To overcome the microscope image's properties of low contrast and highly polluted noise, and to get better segmentation results of sperm image, this paper presents a multi-scale gradient operator combined with a multi-structuring element for the micro-spermatozoa image of white rat, as the multi-scale gradient operator can smooth the noise of an image, while the multi-structuring element can retain more shape details of the sperms. Then, we use the Otsu method to segment the modified gradient image whose gray scale processed is strong in sperms and weak in the background, converting it into a binary sperm image. As the obtained binary image owns impurities that are not similar with sperms in the shape, we choose a form factor to filter those objects whose form factor value is larger than the select critical value, and retain those objects whose not. And then, we can get the final binary image of the segmented sperms. The experiment shows this method's great advantage in the segmentation of the micro-spermatozoa image.

This paper presents an efficient template matching which is adapted to both the rotated mono- and multi- sensor images. The proposed method uses the dominant orientation (DO) to estimate the rotation and, use the hill climbing to search the translation. First, to find the global optimum, climbers are placed all over the reference image with a constant interval, and each climber is assigned to a unique DO, by which the template rotation can be estimated at each climber. Then a class-adaptive clustering is introduced and all the climbers/DOs are clustered into several classes. In each class the template is rotated only once, so the total rotation operations can be reduced significantly. After the rotation, the hill climbing can be conducted and the global optimum can be achieved by the highest climber. Our method need not any presetting of the parameters, however, it is robust and efficient, as shown in experiments.

In this paper we propose an image super-resolution algorithm based on Gaussian Mixture Model (GMM) and a new adaptive image decomposition algorithm. The new image decomposition algorithm uses local extreme of image to extract the cartoon and oscillating part of image. In this paper, we first decompose an image into oscillating and piecewise smooth (cartoon) parts, then enlarge the cartoon part with interpolation. Because GMM accurately characterizes the oscillating part, we specify it as the prior distribution and then formulate the image super-resolution problem as a constrained optimization problem to acquire the enlarged texture part and finally we obtain a fine result.

Medical images have the characteristics of high noise and blurred edges, which makes them difficult to segment using traditional segmentation methods. The level set algorithm, which is a commonly used method for medical image segmentation, is restricted in use mainly due to the extremely intensive computation during the iterative contour evolution. The paper proposes some criteria of loop iteration break for the level set algorithm, making it possible to adaptively adjust the number of iterations to the specific characteristics of various medical images, so that the contour evolution can be terminated appropriately. Meanwhile, we change the step length of the iteration according to the previous loop iteration result, making it possible to decrease the number of iterations needed. To decrease the computational workload, we also restrict the iteration to a certain part of the image instead of the whole image.

This paper presents a fast, precise method for segmenting white blood cell(WBC) based on visual salient features, which is a two-stage algorithm consisting of adaptive WBC location based on salient map(AWLSM) by simulating the process of human perception with bottom-up strategies and extract precise cell structure in cell salient attention window(CASW) using parameter controlled adaptive salient mechanism (PCASM), the first step locates several CASWs in the blood cell image and the second step is to extract nucleus and cytoplasm accurately in each CASW. The experimental results demonstrate that the proposed method has sufficient accuracy and speediness to be used in the automatic blood analyzer.

TIRF microscopy is becoming increasingly popular in cell and molecular biology and opens a challenging computer vision application domain. However, time-lapse images, acquired by TIRF microscopy imaging system suffer the problem of intensity loss due to photobleaching. In this paper, TIRF images were segmented by a Gaussian mixture model into foreground and background. Parameters of the model were estimated through the expectation maximization. Finally, the restored image was wrapped backward to reference frame with the help of foreground parameters. The experimental results showed that the corrected images were effectively compensated and maintained a relatively constant intensity along time.

Currently, locating the tumor ROI is the prerequisite of feature extraction. However, due to the low contrast and complex background of ultrasound images it is hard to obtain the accurate tumor ROI. Other organizations often been wrongly extracted as a tumor region, result in multi-ROI (non-tumor, tumor) in one image. As the result, the performance of tumor classification algorithms will be poor. In such case, ability to discriminate non-tumor and tumor area of classifier is of the most important. This paper proposed bag structure constructor on the basis of multi-ROI and multiple instance learning (MIL) classification algorithm is introduced to solve the above problem that has ability to discriminate non-tumor and tumor area to some extent. Experiments show that accuracy of the proposed method in such problems is 10% more than the traditional ultrasonic classification of breast tumor.

A new approach of multi-modal medical image registration is proposed to overcome the drawbacks of mutual information as taking no consideration of the space information, taking all intensities without distinction, and being sensitive to noise. The proposed method firstly extracts the phase congruencies of the reference and floating image, secondly, it computes quantitative-qualitative mutual information with the phase congruency mappings, finally, the geometric transform is optimized by Particle Swarm Optimization. The quantitative-qualitative mutual information used in our algorithm select the pixels whose utility are larger than the threshold of 1. In addition, Mutual information incorporating phase congruency assimilates the information of both intensity and space. Experiment results show that our approach is more robust in suppressing noise and can achieve higher accuracy.

MDP (Dirichlet Process Mixtures) model is applied to segment medical images in this paper. Segmentation can been automatically done without initializing segmentation class numbers. The MDP model segmentation algorithm is used to segment natural images and MR (Magnetic Resonance) images in the paper. To demonstrate the accuracy of the MDP model segmentation algorithm, many compared experiments, such as EM (Expectation Maximization) image segmentation algorithm, K-means image segmentation algorithm and MRF (Markov Field) image segmentation algorithm, have been done to segment medical MR images. All the methods are also analyzed quantitatively by using DSC (Dice Similarity Coefficients). The experiments results show that DSC of MDP model segmentation algorithm of all slices exceed 90%, which show that the proposed method is robust and accurate.

Based on mean preserving bi-histogram equalization (BBHE), an adaptive image histogram equalization algorithm for contrast enhancement is proposed. The threshold is gotten with adaptive iterative steps and used to divide the original image into two sub-images. The proposed Iterative of Brightness Bi-Histogram Equalization overcomes the over-enhancement phenomenon in the conventional histogram equalization. The simulation results show that the algorithm can not only preserve the mean brightness, but also keep the enhancement image information effectively from visual perception, and get a better edge detection result.

Computed tomography angiography (CTA) is widely used to assess heart disease, like coronary artery disease. In order to complete the auto-segmentation of cardiac image of dual-source CT (DSCT) and extract the structure of heart accurately, this paper proposes a hybrid segmentation method based on k clustering and Graph-Cuts (GC). It identifies the initial label of pixels by this method. Based on this, it creates the energy function of the label with the knowledge of anatomic construction of heart and constructs the network diagram. Finally, it minimizes the energy function by the method of max-flow/min-cut theorem and picks up region of interest. The experiment results indicate that the robust, accurate segmentation of the cardiac DSCT image can be realized by combining Graph-Cut and k clustering algorithm.

This paper presents a novel segmentation method for extracting coronary artery tree from angiogram, which is based on multiscale Gabor filtering and transition region extraction. Firstly the enhanced image is obtained after multiscale Gabor filtering, then the transition region of the enhanced image is extracted using the local complexity algorithm, and the final segmentation threshold is calculated, finally the image segmentation is achieved. To evaluate the performance of the proposed approach, we carried out experiments on various sets of angiographic images, and compared its effects with those of the improved top-hat segmentation method. The experiments indicate that the proposed method outperforms the latter method about better extraction of small vessels, more background elimination, better visualized coronary artery tree and continuity of the vessels.

Traditional deformable models provide a global method for image analysis, but these is easily relapsed into a local optimal in a high noise image and invalid for the image contour with deeply narrow concavities. In this paper, we proposed a novel deformable model to extract the contour of interested object in medical images in medical images. In the procedure of the evolvement of contour curve, by introducing the designed image transform operator to derive the region force from the region information included in the interested object, our method could improve the capacity to alleviate the sensitivity to image noise and converge into complex boundary. Experiments were performed with synthetic and medical images and the feasibility and robustness of our method was demonstrated.