Saliency detection via outlier pursuit in compress domain (SDOPCD)

A saliency detection technique has been proposed in this article. The two major constituents, proximal gradient pursuit and compress sensing techniques, are the important requirement.The preliminery assumption is salient object adhered to minimal portion in an image and exist in large size along with other color based constraints. The concept of superpixel has been employed to divide the image into super pixel blocks. The similar colors within the predefined radius are used to reduce required color space. The blocks of super pixels are used as columns to constitute a matrix which is further compressed using compress sensing methods. The compressed matrix is used in outlier pursuit algorithm. The sparse matrix generated by outlier pursuit provides the required salient object without requiring any reconstruction method. The performance enhancement is verified by comparing with existing methods.