Uniform Extended Local Ternary Pattern for Content Based Image Retrieval

Abstract

The retrieval of images depending on content is a recurrent research topic in medical imaging. Most CBIR systems are designed to help physicians in the diagnostic of the pathologies. Image retrieval according to the content of the texture features can be performed through various methods developed so far. Local texture features are very beneficial for the analysis of the texture, thus, they are extensively used in image retrieval. The original LBP is improved in this paper with a new addition for CBIR called uniform extended local ternary pattern (UELTP). The method decomposes the image into objects; local texture features are extracted and stored into n-dimensional texture feature vectors. Then, the images are frequently obtained from a huge database dedicated for images using these vectors. In this paper, the performance of LBP descriptor, LTP and ELTP are evaluated for CBIR. According to the results, uniform extended local ternary pattern more accurate than other descriptors in terms of image retrieval.