Content-Based Image Retrieval Using Templates of Medical Images

This paper proposes an efficient and novel content–based image retrieval framework, formalized using the higher level patterns present in the image. This novel approach combines the color and texture features of the image with its patterns to increase the retrieval rate. The low level wavelet sub band coefficients are clustered by an iterative approach to form the higher level semantically meaningful patterns. The optimum of horizontal or vertical forward and reverse histogram of the query that has best match with the database image histogram is treated as another reference feature to provide directionally invariant retrieval performance. The Sum of Squared Distance (SSD) similarity measure determines the close similarity between the database images and the query image. Experiments were performed on a diverse collection of reference images, and have shown that the proposed scheme can be efficiently and effectively applied for Content Based Image Retrieval (CBIR) from large databases. In view of the fact that medical images generally exhibit a standard pattern, the proposed framework could show better retrieval performance for medical images rather other natural images. The retrieval results demonstrated significant improvement in precision and recall with increased average retrieval rate, up to 97%.