Improved dominant local binary pattern texture features

Autor: Shahera Hossain, Niraj P. Doshi, Gerald Schaefer
Rok vydání: 2016
Předmět:
Zdroj: 2016 5th International Conference on Informatics, Electronics and Vision (ICIEV).
DOI: 10.1109/iciev.2016.7760181
Popis: Texture features are important in many computer vision applications. LBP is a simple yet powerful texture descriptor that is based on the concept of local binary patterns which describe the relationships of pixels to their local neighbourhood. These relationships are encoded in binary form, and the resulting patterns are then typically used to build histograms over an image or image region. It is observed that only relatively few of these patterns occur frequently in images. Dominant LBP (D-LBP) is a variant of LBP based on these dominant LBP patterns. D-LBP re-arranges the histogram bins in descending order of frequency and then selects the first few dominant patterns as texture features. By doing so, however, it discards the information of which patterns are selected. In this paper, we propose an improved Dominant LBP algorithm that preserves the pattern information and show it, based on an extensive set of experiments on several Outex benchmark datasets, to outperform D-LBP for texture classification.
Databáze: OpenAIRE