Edge Extraction And Image Restoration Algorithm Based On2-D System Filtering

Digital image processing is a more and more rapidly evolving field in science andengineering. Digital image processing usually is defined by the number oftwo-dimensiona（l2-D） image processing computer and interpreted as two-dimensionaldata for any digital processing. Existing edge extraction and image restoration methodin grayscale image filtering mostly proposes1-D and other ways to complete the imagepreprocessing, which usually have a complex program design and low efficiency. The2-D system used in image processing and analysis program design has variousadvantages including flexible implementation in software and hardware and smallphysical size. This dissertation studies the edge extraction and image restorationalgorithm based on2-D system filtering.Edge extraction based on the regions of interest of color image is achieved, therelated color space transformation is adopted. And the multi-scale filtering is applied toeach coordinate component. Then, Sobel and Canny operator are proposed to achievetwo kinds of edge extraction for the utilization of the targeted image region. Theobtained image edge information become much richer with this method, and thus theperformance of edge continuity is much better.In this dissertation, the images are represented by2-D system mathematical model,and block-based2-D Kalman filtering for image restoration method is proposed toremove degraded image blur and noise. To improve the initial setting of the specificfilter correlation image, a hierarchical processing method is introduced. At the sametime, interpolation method based on edge protection is adopted to improve restorationeffect based on the protection of the edge of the image.Processing based on Kalman filter requires the noise statistical properties inadvance, but this condition can be satisfied in practice. For the image with the noisestatistical properties being unknown in advance, the robust H infinity filtering method isadopted to solve this problem. In tackling the noise negative definite inner product, theproblem is extended to Krein space for processing, while it can be resolved in theHilbert space with the constraints of the problem.