Abstract:

A novel content based image retrieval (CBIR) algorithm using relevant feedback is presented. The proposed framework has three major contributions: a novel feature descriptor called color spectral histogram (CSH) to measure the similarity between images; two-dimensional matrix based indexing approach proposed for short-term learning (STL); and long-term learning (LTL). In general, image similarities are measured from feature representation which includes color quantization, texture, color, shape and edges. However, CSH can describe the image feature only with the histogram. Typically the image retrieval process starts by finding the similarity between the query image and the images in the database; the major computation involved here is that the selection of top ranking images requires a sorting algorithm to be employed at least with the lower bound of O(n log n). A 2D matrix based indexing of images can enormously reduce the search time in STL. The same structure is used for LTL with an aim to reduce the amount of log to be maintained. The performance of the proposed framework is analyzed and compared with the existing approaches, the quantified results indicates that the proposed feature descriptor is more effectual than the existing feature descriptors that were originally developed for CBIR. In terms of STL, the proposed 2D matrix based indexing minimizes the computation effort for retrieving similar images and for LTL, the proposed algorithm takes minimum log information than the existing approaches.