Tag Based Image Ranking

Abstract

Abstract-The extent in the number of images has seen a prominent rise in the past decade. Image annotation with its research traits has gained much importance in the research field. At times image annotation is considered as multi label classification problem. On the contrary the main demerit is the requirement of a large number of training images with clean and complete semantics. This shortcoming is served by a combined approach of tag ranking and matrix recovery. The problem of making a binary decision for every tag is avoided in this approach and instead we rank the tags in descending order of their relevance. In addition, the proposed system also collects the aggregated data collected from the proposed models for different tags into a matrix and casts it into a matrix recovery problem. The matrix trace norm is assigned explicit control for controlling the model complexity, for establishing a reliable prediction model.