Image Retrieval System Using Kirsch based Local Ternary Pattern

Abstract

This paper addresses the challenge imposed by the tremendous growth of digital data to retrieve relevant images. In this paper, a novel feature design methodology is proposed to represent images efficiently for Content Based Image Retrieval (CBIR). Gener- ally, local patterns are computed directly on images and hence, directional information of the images is ignored. In the proposed feature, Kirsch operators are used to highlight eight major directional changes in the image and further, Kirsch Ternary Local Pat- tern (KLTP) is extracted by analysing local inten- sity variations in the neighbourhood. In KLTP, global color information is also incorporated to make it robust and perform well on variety of images. Experiments on natural and texture databases are done to verify the performance, as compared to the available features in the literature.