Study On Image Restoration Algorithms And FPGA Implementation Technology

Image processing techniques have developed to be a comprehensive subject with wide application and abundant contents. Its development bases on two research fields, which have different proprets, but compact relationship: image processing algorithms and corresponding circuit imlementation. The application requirement is the power that drives the development of image processing techniques. The goal of the research of image processing algorithm and its impelementation technique is to improve the visual effect of observed image, increase the processing speed and prepare data for the further image processing and image analysis. It brings the social benefit and economic returns.Image restoration is one of image processing techniques, which estimates the original image from the observed image and the degradation model. All the time, the researchs of image restoration algorithm and its implementation are important field of image processing, and they have wide application fields. The main contradication in image restoration is to smooth noise and preserve the details in image, simultaneously. The research contents in this dissertation include restoration algorithms with the capability of detail preservation based on iteration, circuit architectures and FPGA impelementation techniques.The geneal research background is given at the beginning of the dissertation. And a detailed stduy follows. This dissteration presents the past work and state of the art about image restoration from the three points of view: the computational processes, circuit architectures and FPGA impelementation techniques.The followings are the main propositions:First of all, the algorithm, nonlinear diffusion driven by local features for image restoration, is proposed. In order to remove the noise, simultaneously, preserve the edge information in an image, the features in an extended neighborhood are extracted, and the appropriate diffusion coefficients are established, such that the diffusion speeds are properly controlled according to the characteristic of an image local region. This algorithm breaks through the limitations of the traditional nonlinear diffusion methods and obtains the better balance between removing noise and preserving the edges. Secondly, the proposed image restoration algorithm is improved, such that it can be mapped to circuit architecture easily. The circuit architecture adopts the two-level buffers, parallel processing and pipeline techiniques. The image data and intermediate results are reused abundantly. Xilinx XC3S1000 FPGA is employed to implement the circuit. The experimental results show that this circuit has large throughout and strong computational power.Thirdly, a circuit design is proposed for image restoration problem, in which the observed image is contaminated by multiplicative noise. The circuit is implemented under resource constraints of FPGA. The design uses the resource of FPGA efficiently due to explore the symmetry of point spreat function. The circuit is also implemented by the use of Xilinx XC3S1000 FPGA, and it can finish the image restoration processing fleetly.Finally, an adaptive Hopfield neural network algorithm based on cooccurrence matrix analysis for image restoration is proposed. This algorithm computes the co-occurrence matrix of each image region, extracts texture feature and clusters the nonzero elements in co-occurrence matrix. A new concept, Detail Intensity, and its computation method are proposed. Detail Intensity distinguishes accurately flat and detail regions in image and adjusts adaptively Hopfield network weight coefficients through a nonlinear function, such that the weight coefficients are suitable for image texture features. The iterative process of image restoration and the updating process of weight coefficients of neural network are executed alternately. This algorithm removes noise in smooth regions and reveals details in detail regions. It conforms to human perceptual criteria. The comparative experimental results show, restored images, which are produced by this proposed algorithm, have higher SNR and better visual effect.Image restoration problem is ill-posed inverse problem, and it has dense operations in computional processes. Image restoration processing is one of difficult problems in image engineerng practices. The algorithms and circuit architectures proposed in this dissertation have remarkable theoretica significance and application value.