Combining Local Binary Pattern and Speeded-Up Robust Feature for Content-Based Image Retrieval

Abstract

Large number of digital image libraries containing huge amount of images have made the task of searching and retrieval tedious. Content-Based Image Retrieval (CBIR) is a field which finds solution to this problem. This paper proposes CBIR a technique which extracts interest points from texture feature at multiple resolutions of image. Local Binary Pattern (LBP) has been used to perform texture feature extraction and interest points are gathered through Speeded-Up Robust Feature (SURF) descriptors. The multiresolution decomposition of image is done using Discrete Wavelet Transform (DWT). DWT coefficients of gray scale image are computed followed by computation of LBP codes of resulting DWT coefficients. The interest points from texture image are then gathered by computing SURF descriptors of resulting LBP codes. Finally, feature vector for retrieval is constructed through Gray-Level Co-occurrence Matrix (GLCM) which is used to retrieve visually similar images. The performance of the proposed method has been tested on Corel-1 K dataset and measured in terms of precision and recall. The experimental results demonstrate that the proposed method performs better than some of the other state-of-the-art CBIR techniques in terms of precision and recall.