Feature Selection approach using KNN supervised learning for Content-Based Image Retrieval

Digital images have a serious influence over all the world, since, images have a very tangible importance, which impacts several people needs, this gave image retrieval researches this importance. So, Content-based image retrieval (CBIR) became one of the challenges in information retrieval and image processing fields, its services needs to apply for both business and science domains. In this paper we have proposed a CBIR approach that based on statistical methods feature selection via k-nearest neighbor (KNN) technique. Hence, supervised learning is employed for CBIR to train WANG database images, then the query image was the tested sample, and among predicted class we have filtered the final results. Number of randomly selected query images have been tested via this approach, with a variation of CBIR criteria, which have varied among image types, category query, and threshold distances. The average results have presented in three result groups, some RGB images results have showed good evaluation, in term of precisions measure, also grayscale and RGB images have varied values; on the evaluation of f-measures results, which have changed according some different CBIR factors. Proposed feature selection have made CBIR simpler and effective, this have been showed in high precision evaluation in final results.