PSO based Automatic Image Annotation using Weakly Supervised Graph Propagation
R. Sumathi, D. Narmadha · 2013
Abstract: In this work, a weakly supervised graph propagation method is proposed to automatically assign the annotated labels at image level to those contextually derived semantic regions. The images can be segmented into different regions using color image segmentation. The graph is built with the over-segmented patches of the image pool as nodes. Image-level labels are carried out on the graph as weak supervision information over sub graphs, each of which relates to all patches of one image, and the contextual information from end to end different images at patch level are then mined to assist the process of label propagation from images of their descendent regions. The Weakly Supervised Graph propagation encodes two types of contextual information among image patches, i.e., consistency and incongruity. The decisive optimization problem is proficiently solved Particle swarm optimization (PSO) method. It iteratively tries to improve a candidate solution with regard to a given measure of quality for optimization. Experiments using the data sets show the effectiveness of the proposed method for the task of collective image parsing.