Extraction of Semantic Coherent Regions Using Bayesian Nearest Neighbor Search

In image processing, mining of semantic objects from images is one of the most significant and demanding problems in image examination. Nowadays, numerous schemes are there where this type of processing is absolutely desired. A first request of these thoughts is compression where diverse objects are implied with diverse quality which permits the exploitation and communication with the objects in the image. For semantic image segmentation, the common potts model (2) is used for spatial coherency that extends the higher order spatial semantic coherent class labels. However, the extraction of semantic regions is not so obvious and devours more time to develop the semantic regions. To overcome this, in this paper, we are going to present a scheme for extracting and clustering the semantic coherent regions obtained from the semantic connected coherent criteria discussed in the previous paper. The segregated semantic coherent regions are clustered based on Bayesian nearest neighbor search with neighborhood pixels. After clustering the semantic coherent regions of image, the segmentation of semantic regions alone takes place by adapting the cluster object purity obtained through the semantic connected coherent regions criteria. Then the clustered image regions are post processed with linear noise filters. An experimental evaluation is conducted with the set of images to estimate the performance of the proposed extraction of semantic coherent regions using Bayesian nearest neighbor search (ESCRBNN) in terms of cluster object purity, coherent region search efficiency, and computational complexity.