PROJECT TITLE :

Multi-view object extraction With fractional boundaries - 2016

ABSTRACT:

This paper presents an automatic technique to extract a multi-view object during a natural atmosphere. We assume that the target object is bounded by the convex volume of interest outlined by the overlapping area of camera viewing frustums. There are 2 key contributions of our approach. Initial, we gift an automatic methodology to identify a target object across different pictures for multi-read binary co-segmentation. The extracted target object shares the same geometric illustration in space with a distinctive color and texture model from the background. Second, we have a tendency to gift an algorithm to detect color ambiguous regions along the object boundary for matting refinement. Our matting region detection algorithm is predicated on the information theory, that measures the Kullback-Leibler divergence of local color distribution of different pixel bands. The local pixel band with the largest entropy is chosen for matte refinement, subject to the multi-read consistent constraint. Our results are high-quality alpha mattes consistent across all totally different viewpoints. We demonstrate the effectiveness of the proposed technique using various examples.