Image Detail-preserving Filters Research Based On Switching Median Filtering

During the forming, transferring and memorizing of the digital images, the images are often corrupted by different kinds of noises because of the defection of the imaging system, transferring medium and memorizing equipment. Therefore, in the fields of pattern recognition, computer vision, image analyzing and video coding, the early vision processing is very important. The result of it will directly affect the quality and outcoming of the later processing.Different from the linear filter, the non-linear filter can not only remove noise effectively but also keep the details of the digital images, so images can be clearer and more vivid. As an effective processing technology, the non-linear filter is widely used in the digital image processing. Median filters are representative. The typical median filter can remove the impulse noise, but it also corruptes some very important details of the images. There are many improved median filters firstly suggested to overcome the problems. Three new median filters are proposed in this paper, such as the threshold decomposition multistage median(TDMM) filter and modified threshold decomposition multistage median filter; pre-segmentation binarized switching median(PBSM) filter; ordering threshold switching median(OTSM) filter.In this paper, based on the order statistics and image segmentation and under the switching-based median(SBM) filter structure, the digital image is classified into different areas with pre-processing. With the characters of these areas, the median filter not deals with the edges and details areas but removes the noise pixels to obtain good subjective results. Firstly, to deal with impulse noise with very big values, TDMM filter seperates the image into noise-densed parts, then uses SBM filter to remove noise; To deal with more complex noise, PBSM filter separates the image into part gray level images, then uses TDMM filter to dispose them; For more general impulse noise, OTSM filter classes the pixels into several types, and uses SBM filter to treat with them differently. From many noise models simulation, it shows these three methods are effective.