Article ID Journal Published Year Pages File Type
494465 Neurocomputing 2016 11 Pages PDF
Abstract

In this paper, an efficient local operator, namely the Local Quantization Code (LQC), is proposed for texture classification. The conventional local binary pattern can be regarded as a special local quantization method with two levels, 0 and 1. Some variants of the LBP demonstrate that increasing the local quantization level can enhance the local discriminative capability. Hence, we present a simple and unified framework to validate the performance of different local quantization levels. In the proposed LQC, pixels located in different quantization levels are separately counted and the average local gray value difference is adopted to set a series of quantization thresholds. Extensive experiments are carried out on several challenging texture databases. The experimental results demonstrate the LQC with appropriate local quantization level can effectively characterize the local gray-level distribution.

Related Topics
Physical Sciences and Engineering Computer Science Artificial Intelligence
Authors
, , , ,