Haar-like Local Ternary Pattern for Image Retrieval

Abstract

In this paper a novel Haar- like local ternary pattern (HLTP) is introduced for content-based image retrieval. Many variants of local patterns like LBP, LTP etc. ignore the high pass information present in an image. The proposed HLTP feature not only extracts this information but the best suited Haar-like filter is also selected to represent the high pass information. Selection of only the best filter reduces the complexity of the feature. Then, in order to capture the structural similarity within the image, local ternary edges are computed in 3x3 neighborhood for each pixel of the dominant filter image. Hue and saturation histograms are concatenated with the HLTP feature to make it robust against colour variations. Experiments are conducted on two diversified datasets and performance of proposed method is compared with the existing methods.