Saliency Detection by MICCLLR

Saliency detection means detecting visually attractive regions in images. It is an aspect of exploring visual attention from a computer vision. Each image is segmented to get bags. Features are extracted from each bag. Features, including low-, mid-,and high-level, are incorporated into the learning and testing process. They are position, color, texture, scale, center prior, and boundary. From these features, meta-instance is calculated for each bag. For detecting the salient region, a classifier is learned with meta-instance . In the existing system, saliency value is calculated using EM-DD algorithm and learned using multiple-instance learning. The idea of EMDD is to model the label of each bag with a hidden variable, which is estimated by the expectation-maximization (EM) algorithm. In the proposed system for improving the accuracy of saliency map, algorithm called MILCCLLR is used for metainstance creation. It is a Generalized Multiple-Instance Learning Algorithm Using Class Conditional Log Likelihood Ratio that converts the MI data into a single meta-instance data allowing any propositional classifier to be applied. Each image is tested with the learned model. Experiments shows that, MICCLLR algorithm is better than the EM-DD algorithm and will give better saliency map. Keywords— Saliency, saliency map, meta-instance, machine learning, computer vision.