Content-Based Image Detection of Semantic Similarity

This paper present a new image similarity measures for fast retrieving latent semantic similarity images via clustering feature descriptors of images and indexing images. The approaches used in this paper is hierarchical cluster which is used to segment the collection of the image feature from image database and locality sensitive hashing which is used to index the similarity images. The proposed method introduces an `image tag library´ of the vector clustered region feature descriptors and an improved hash algorithm for images retrieval exploits. An approximate set intersection between image descriptors is used as a similarity measure. The paper proposed a convenient way of transforming sophisticated SIFT features into predefined `tag set´, and also proposed an efficient way of simplifying similarity measures based on hamming distance that have proven to be essential in image object retrieval. The procedures in the paper focused primarily on scalability to very large image databases, where fast query processing is necessary. The method requires only a small amount of data need be stored for each image. We demonstrate our method on the Ground Truth Database from University of Washington and also on challenging the USC-SIPI Image Database.