On integrated model for image filtering and segmenting based on Structure Statistic of Decomposable Markov Network
Jiannong Cao, Yong Fang · 2009
The removing of image noise, which is abnormity of pixels, is image filtering, and the key of problem is ascertaining the location of pixels with abnormity gray-level. The segmenting pixels with no-similar gray-level are image segmentation. Obviously, the abnormity gray-level is equal to no-similar gray-level in measurement of pixels. So a model integrated (namely Decomposable Markov Networks, for short, DMN), which not only can segment but also filter image, is put forward. The microcosmic configurations of DMN are obtained by computing pixels attribute (namely gray-level, texture and so on), and can firstly identify normal (namely including no-similar or similar gray-level) or abnormity gray-level (namely possible noise). The abilities of DMN identifying are realized by linking intension of networks, which derive a new uncertain complication (namely uncertain relations of microcosmic link) that is leaded by natural random factors of image data spatial distributing. So the macroscopical Structure Statistic of Decomposable Markov Network (SSDMN) can identify statistical abnormity gray-level (namely including no-similar [possible noise] and similar gray-level), and then filtering and segmenting image is implemented by a model integrated. Obviously, the DMN is facility of integration, and settles a difficult problem, which is uniting description of pixels numerical value and its spatial locations.