Nonstandard Regularization for Selected Image Processing and Computer Vision Problems

In this thesis, a unified approach which is based on regularization theory has been applied in solving surface reconstruction, edge detection, edge integration, image restoration, image coding and compression problems of image processing and computer vision. It is observed that, edge detection performance of the weak membrane modeling is superior than that of the adaptive smoothing. A nonstandard regularization based image restoration algorithm is developed and it is observed that the blurring can be successfully removed. A new image compression scheme is introduced where sparse information along discontinuities is used. The DORS representation is used in two dimensions to obtain multiscale edge images from two regularized solutions, and the multiscale edges are integrated by a multiscale edge integration algorithm which has been expanded to two dimensions.