Edge detection and filtering of images corrupted by nonstationary noise using robust statistics

Images are a type of data widely used, processed and analyzed in CAD and telecommunication systems. To retrieve useful information from images, they are often subject to different kinds of preprocessing that commonly include edge detection and filtering. These operations can be performed by standard means if noise type and statistics are known in advance. In this paper we address situations when such a priori information is not available. Two local operators based on robust statistics calculated in spatial and spectral domains are proposed and analyzed for edge detection application. Then we show how the obtained edge maps can be exploited in locally adaptive filtering based on discrete cosine transform (DCT).