An Efficient Perceptual of CBIR System using MIL-SVM Classification and SURF Feature Extraction

Hasty increase in use of color image in recent years has motivated to the need of retrieval system for color image. Content Based Image Retrieval (CBIR) system is used to retrieve similar images from large image repositories based on color, texture and shape. In CBIR, the invariance to geometrical transformation is one of the most desired properties. Speeded Up Robust Feature (SURF) and Multiple Instance Learning Support Vector Machine (MIL-SVM) are proposed for extracting invariant features and improving the accuracy of image retrieval respectively. The proposed system consists of the following phases 1) Image Segmentation using Quad tree Segmentation 2) Extraction of features using SURF 3) Classification of images using MIL-SVM 4) Codebook design using Lindae-Buzo-Gray (LBG) algorithm 5)Measurement of Similarity between Query image and the database image using Histogram Intersection (HI). In comparison with the existing approach, the proposed approach significantly improves the retrieval accuracy from 74.5% to 86.3%.