A high-throughput texture classification approach using a new descriptor

In this paper, we propose a simple construction approach (FR: features´ value range) as a high performance texture descriptor. The FR works based on local textural information. We show the throughput of texture classification can be improved using the FR. In the classification process, the FR is considered as a pre-classifier and selects a few candidate categories for an input texture. Using the proposed approach, comparison time of the main classifier is reduced. To evaluate of the FR in different situations, some criteria have been proposed. To implement of the proposed approach, the texture descriptors such as local binary pattern (LBP), Haralick, and circular Gabor filter (CGF) are considered. The experimental results are done by implementation of the FR approach on the Scene-13, Outex and UIUC data sets. The results show the throughput of texture classifiers improve up to 14.85×.