Local Septenary Patterns for Breast Density Classification in Mammograms

This paper presents the local septenary patterns (LSP) operator, which is a variant of the local binary patterns (LBP) texture descriptor inspired from local ternary patterns (LTP) and local quinary patterns (LQP). We introduce a seven-encoding system approach to capture more texture details and resulting in more discriminant features. Unlike the LTP and LQP operators, we introduce an automatic approach to determine the operator's threshold values by computing the first-order statistical values of the central pixel's neighbourhood. Experimental results suggest that the proposed approach produces competitive classification accuracy based on 322 mammograms taken from the Mammographic Image Analysis Society (MIAS). The proposed approach produced classification accuracy of 82.77% compared to 76.64%, 79.58% and 81.35% maximum accuracy, for LBP, LTP and LQP, respectively.