Advanced Local Direction Cross Diagonal Matrix

Abstract

This paper derives a new frame work for CBIR by considering edge responses on a local neighborhood, cross diagonal texture features of the edge responses and statistical features of textures. In the literature Local Directional Patterns [1] is derived to overcome the noise related disadvantages of LBP [2]. The main disadvantage of LDP is, in selecting the number of top edge responses of the sampling points. The LDP features vary based on the selection (chosen number) of top edge responses. This paper derives a new operator called Advanced LDP (ALDP) that overcomes the above disadvantage. This paper derives a ternary pattern on ALDP instead of traditional binary pattern and splits the ALDP into cross and diagonal matrix and derives GLCM features. These derived features represent the feature vector for CBIR. The proposed Advanced Local Direction Cross Diagonal Matrix (ALDCDM) is experimented with various types of databases and compared with state of art methods. The results indicate the efficiency of the proposed descriptor.