A revised averaging algorithm for an effective feature extraction in component- Based Image Retrieval system

Abstract

Currently used description based image retrieval is not suitable for an effective image search with a large unstructured repository of images. Thus to overcome the issue the concept of Content Based Image Retrieval (CBIR) aroused, where image search is done using the features which can be extracted from the image content. Color, edge, shape and texture are the most common. In most of the CBIR systems, there are few sub functions as Feature Extraction, Clustering and Storing, Similarity Matching and Display results. Quality of the last result in CBIR heavily depends on Feature Extraction module which is time consuming. Proposed solution was designed for and Effective CBIR by improving the efficiency of Feature Extraction. In most of CBIR systems search image set must be inserted. Then query image can be inserted. As the output user can retrieve group of images alike the queried image. When image set is inserted, system extracts features of each image, average them, cluster them, indexed them and store them in appropriate clusters. Features of set of images are extracted by the Feature Extraction Module. Before clustering, image matrices are averaged to one dimensional array using a revised averaging algorithm to reduce the complexity of calculations and perform efficiency. Averaged features are clustered using K-mean algorithm and stored appropriately. When query image inserted, again extracts the features of it and compares the stored features and calculates a similarity value for each image in the nearest cluster. Finally it displays the result image set according to the order they are matched with the query image.