Salient Region Detection in Natural Images Using EM-Based Structure-Guided Statistical Textural Distinctiveness

The objective of salient region detection is to separate salient region from entire image. This salient region detection framework consists of a structure-guided statistical textural distinctiveness approach. This approach includes the five main stages: i) image decomposition ii) textural representation, iii) texture modeling, iv) matrix construction, and v) saliency map computation. In the image decomposition stage, decomposition of image into structural image elements to learn a structure-guided texture model. In second stage, define a rotational-invariant neighborhood based texture feature model that represents the underlying textural characteristics of regions in a local manner. In texture modeling stage, Sparse texture modeling is done using structure-guided texture learning. In matrix construction stage, characterize all pair-wise statistical textural distinctiveness within the sparse texture model of the image and construct a textural distinctiveness matrix. In the final stage, the saliency of a region can be computed as the expected statistical textural distinctiveness of the region in the given image .The proposed approach has been extensively evaluated on images from MSRA-1000 datasets. Citation:SRUJY KRISHNA A U, FEDERAL INSTITUTE OF SCIENCE AND TECHNOLOGY, ANGAMALY, ERNAKULAM, KERALA, INDIA; SHIMY JOSEPH ,FEDERAL INSTITUTE OF SCIENCE AND TECHNOLOGY, KERALA, INDIA. "Salient Region Detection in Natural Images Using EM-Based Structure-Guided Statistical Textural Distinctiveness." Global Research and Development Journal For Engineering 19 2016: 31 - 38.