Abstract—Texture analysis is a crucial step in various computer vision and pattern recognition applications. A large number of techniques for describing, and retrieving texture images has been proposed during the last few decades. The conventional Local Binary Patterns (LBP) has proven to be an efficient technique for capturing important texture properties. Its simplicity and large success motivated researchers from computer vision and image processing community to study the descriptor and propose variants that overcome identified limitations. One of these limitations is the lack of spatial information in the LBP histograms. Recently, few co-occurrence-based methods have been introduced to overcome this drawback. These techniques improve significantly LBP performance but they are in general much slower than the original operator. In this paper we present two algorithms that make use of the local similarities of the binary patterns to improve the performance of the original LBP without dramatically increasing the execution time. The first algorithm combines the conventional LBP histogram with a histogram recording local similarities of the LBP patterns. The algorithm is almost as fast as the original LBP technique and yet outperforms the operator in terms of retrieval accuracy. The second algorithm records the co-occurrences of LBP codes with Local Similarities of the Patterns. Its retrieval performance is similar to those of the co-occurrence based techniques but with a significant gain in execution time. We have conducted several experiments on two popular datasets (Brodatz, and Outex) to demonstrate the efficiency of the proposed algorithms.

 

Index Terms—texture descriptor, LBP operator, co-occurrence matrices, local similarities of the patterns

Cite: Abdelhamid Abdesselam, "Local Similarities for Boosting the Performance of Local Binary Patterns Technique," International Journal of Signal Processing Systems, Vol. 4, No. 6, pp. 510-514, December 2016. doi: 10.18178/ijsps.4.6.510-514