Semantic Image Segmentation and Web Supervised Visual Learning

Inferring user search goals is very important in improving search engine relevance and user experience. Queries may not exactly represent the need of the user because of polysemy in keywords. At some times user may tend to form short queries so that the sense of search may be messed up on the whole. So far image is been searched as product indent most probably based on query classification. Initially they select some tag words as textual suggestions to satisfy relatedness and informativeness; however it depends on the precision of tags. So as to leverage the semantic gap between image features and image semantics a strategy to focus on click content logs is proposed. Both strategies click content information and click session information works drastically forward to sum up the gap between image visual features and textual content. The image results are then re-ranked in a diverse manner to provide a better idea about the semantic correlation among images and its number search goals of each query can be obtained by a classification risk approach.