SECOND ORDER DERIVATIVE CROSS DIAGONAL MATRIX APPROACH FOR CBIR

Abstract

In this paper, we propose two descriptors based on second-order derivative tetra patterns for efficient content based image retrieval (CBIR). The proposed descriptors initially encode the first-order derivative along 00 and 900 over the center pixel and also on each of the sampling pixel or referred pixel of a 3x3 window. These derivatives derive ternary directions. This paper computes the relationship between the first-order ternary derivative of center pixel and each referred pixel. This process derives the second-order derivative tetra pattern ( ) for each referred pixel. This paper then computes unit for each 3x3window of tetra patterns. This paper initially computes the histograms of individual color planes of HSV model. The proposed first descriptor integrates individual histograms of HSV plane with , and this derives color based ( ). To reduce the bin size of , this paper divides the 3x3 into the cross and diagonal units and derives the second-order derivative cross and diagonal matrix ( ). The GLCM features are computed on . The proposed descriptors and are tested with popular databases and also compared with the recent methods of CBIR.