CBIR Systems Based Approaches for Improving Computational Efficiency

Computer Science & Engineering

The current work additionally centers on the chance of integrating tone and edge information with the interest-point based invariant descriptors in the production of visual bags for improving the retrieval performance in natural image databases. For this, a multi-combination approach is embraced, in which, the edge-shading features of the images are joined through early combination for making the vocabulary of visual words. The image histogram worked with the resultant vocabulary is then joined with histogram developed with vocabulary of invariant features through late combination to portray the image. The joining of extra information helps in giving a superior portrayal of the images and helps in improving the retrieval performance. Also, for multi-query CBIR frameworks, another feature swap algorithm is proposed for similarity computation that can offer better retrieval results without query refinement or feature reweighting. The algorithm decides the similarity among query and target images in the database by processing the total amount of the removals of the query-set images' centroid brought about by supplanting every component in the query image set with the up-and-comer images in the database; subsequently successfully amassing the dissimilarity between the images in query-set and the target dataset images. Content-based image retrieval is against customary idea based methodologies. "Content-based" implies that the hunt examines the contents of the image as opposed to the metadata, for example, keywords, labels, or portrayals related with the image.

This work is licensed under a Creative Commons Attribution-ShareAlike 4.0 International License.