Authors : Ch. Sudha Sree and M.V.P. Chandra Sekhara Rao
Abstract: Rotation invariant texture classification is an important issue in image analysis. For more than a decade Local Binary Pattern (LBP) variants have been proven to be successful methods in wide applications of rotation invariant texture classification. However, these invariant patterns are not absolutely rotation invariant and some of these are noise sensitive/insensitive. Till date, no ternary LBP variant is found as rotation invariant and noise in sensitive. This study proposes a rotation invariant and noise insensitive texture descriptors called, Local Ternary Count (LTC) and Completed Local Ternary Count (CLTC). The two descriptors characterize the textures using local ternary gray scale difference by avoiding the micro-structure. The proposed CLTC is a set of three new operators defined for sign, magnitude and central pixel components. Experiments are conducted on three well known benchmark databases Outex, UIUC and CUReT. The performance of the proposed method is analysed by comparing with the various existing LBP variants. It is observed that, CLTC exhibits significant improvement in classification accuracy and is more robust to noise when compared with LBP variants at different Signal-to-Noise Ratio (SNR) values.
Ch. Sudha Sree and M.V.P. Chandra Sekhara Rao, 2018. Completed Local Ternary Count for Rotation Invariant Texture Classification. Journal of Engineering and Applied Sciences, 13: 3633-3641.