Automatic image annotation based on vocabulary prior probability

Abstract

Automatic image annotation is an important and challenging task in computer vision. The existing models only use low-levels features of images to do the approximate calculation, without considering the influence of semantic information. This paper proposes a new automatic image annotation algorithm based on the vocabulary prior probability. It can solve the semantic gap to a certain extent. The algorithm is divided into two stages, first according to the existing generative model calculated the initial annotation word, and then calculated image similarity with considering the annotated words to improve the result of the annotation. The experiments over Corel5k images have shown the proposed method can effectively improve the rate of the annotation's accuracy and recall.