A novel texture descriptor Octagonal-based Directional Distance Local Binary Pattern (ODLBP) for image classification

Various statistical descriptors are available for extracing features from texture images, with the Local Binary Pattern being among the most widely used methods for uncovering the fundamental structure of a texture. Each variants have their own advantages; however, these algorithms provide some major issues that needs to address. In order to solve these issues, we introduce a novel texture descriptor known as the octagonal directional distance local binary pattern (ODLBP) for image classification. The descriptors explain new approaches to feature extraction, while the computational framework is very different and much superior to prior methods. This is a unique descriptor that makes a binary calculation with a thorough comparison of the central pixel and its adjacent and secondary neighboring pix- els, thus offering substantial improvement over the conventional techniques. Thus, ODLBP is highly recommended for improving the texture representation, as no pixel value is overlooked. This approach is thorough enough to allow for more complete and accurate feature extraction, providing more pronounced image representations compared to those obtained through previous methods. The efficiency of ODLBP descriptors has been checked quite thoroughly against popular datasets like KTH-TIPS, KTH- TIPS 2b, Caltech101, CUReT, Leeds Butterfly and Brodatz. The results have proved that ODLBP is very easy to realize and implement; on the other hand, it ensures superior performance in image classification.