Texture classification model based on adaptive local ternary pattern
The success of image processing applications heavily relies on accurate texture analysis. Local Binary Patterns (LBP) is one of the texture models that has found its way into huge application in image texture analysis due to its simplicity and efficiency. However, LBP is more sensitive to noise and can easily classify different pattern into the same class therefore reducing its discriminating property. Local Ternary Pattern was proposed as an extension of LBP and was found to be more resistant to noise. However, LTP and its variants use a static threshold that lacks statistical relationship with the pixel values of an image and which makes it dynamically inappropriate to all images of a dataset or different datasets. This research proposes Adaptive Local Ternary Pattern (ALTP) that uses a dynamically calculated threshold to determine texture pattern for an image region. Experiments conducted on complex texture datasets (KTHTIPS2b, Stex and Describable Textures Dataset (DTD)) showed that ALTP outperformed state-of-the-art texture feature descriptors.