IMAGE: Region-Based Image Retrieval Toolbox

Abstract

This paper focuses on a low-dimensional color-based indexing technique for achieving efficient and effective retrieval performance. We have developed a region based image retrieval toolbox (IMAGE: Indexing Multidimensional data And Grouping for Efficient Retrieval) that aims at searching image databases for specific image regions that are similar to the given query image region using color feature. The mean shift segmentation algorithm, a robust clustering technique based on color is used to extract region of interest, thereby improving the quality of image retrieval. The region feature attribute is computed at the time of ingest into the database. The cluster (region) mode (color value) is used as representative of image in 3-D color space. To efficiently retrieve the images from database cluster-based R *-Tree indexing method is proposed. Here only representative features of clusters are used for indexing, thus reducing the time. The efficiency and accuracy of the proposed method is compared with R *-Tree and sequential search. Experimental results show the performance of the proposed method. The prototype is developed to retrieve images based on query-by-example in JAVA.