CONTENT BASED IMAGE RETRIEVAL USING LOCAL FEATURE DESCRIPTORS

Abstract

The current research work presents design of various new feature descriptors which are useful for image retrieval. These either increase the. accuracy or decrease the computational complexity of content based image retrieval (CBIR) system. These features are either global or local, color or texture and spatial or transform domain. In this report, two new techniques are proposed for image indexing and retrieval. The first technique, magnitude line edge binary pattern operator (MLEBP), which is extension of line edge binary pattern (LEBP) is proposed for image indexing and retrieval. This method represents the local region of an image based on mean of distribution of line edges in contrast to LEBP. In the MLEBP, the central pixel is replaced by mean of neighbors' line edge values. The combination of both operators LEBP and MLEBP is proposed in this work. The retrieval results of the proposed method are tested on multimedia image databases i.e. Corel-1000, Corel-5000 and field database. The performances of the proposed method are measured in terms of precision, recall and average retrieval rate (ARR). The results after being investigated show a significant improvement in terms of precision and recall as compared to LEBP and local binary pattern (LBP). The second technique, directional line edge binary pattern (DLEBP) is proposed for texture image indexing and retrieval. This method is also the extension of line edge binary pattern (LEBP). The directional. line edges of neighbours for a particular center pixel are determined by eight directional windows. The 3 x 3 directional window is considered for this purpose. The retrieval results of the proposed method are tested on Brodatz image database. The performance of the proposed method is measured in terms of ARR. The results show a significant improvement in terms of ARR as compared to Gabor transform (GT), dual-tree complex wavelet transform (DT-CWT), dual-tree rotated complex wavelet transform (DT-RCWT), local ternary pattern (LTP), LEBP and LBP. The retrieval results of proposed methods are tested on multimedia image databases (i.e. Corel-1000, Corel-5000 and field database) for first method and Brodatz image database for second method.