Abstract: The Local Binary Pattern(LBP) method does not consider the contrast value between neighbor pixels, so the feature representation ability of LBP method is limited. According to the problem, this paper proposes a Local Multi-layer Contrast Pattern(LMCP) method to overcome this disadvantage. It limits the illumination variation and divides the contrast value range between neighbor pixels in local area into several layers. Every contrast value is mapped into a certain layer. The statistic histogram for every contrast layer is constructed using the same idea as LBP method. In order to solve the problem of expanded feature dimension, it adopts statistics mapping method to reduce the histogram bins. Experimental result based on Outex_TC_00012 texture database proves that the LMCP method has good classification effect.

Key words: Local Binary Pattern(LBP), Local Multi-layer Contrast Pattern(LMCP), contrast degree, illumination invariance property, multi-resolution

摘要： 经典局部二值模式(LBP)方法没有考虑像素之间的对比度，从而限制其描述纹理特征的能力。为此，提出一种局部多层对比模式(LMCP)方法，将其应用于纹理特征分类中。通过预处理把光照变化控制在一定范围内，将局部区域临近像素间的对比度分为若干个层次，使每个对比度值映射到某个层次中，按照LBP的类似方法构建每个层次的统计直方图，采用统计映射的方式降低特征维度。基于Outex_ TC_00012纹理数据库的实验结果表明，该LMCP方法具有较好的分类效果。

关键词: 局部二值模式, 局部多层对比模式, 对比度, 光照不变性, 多分辨率

CLC Number: 

• TP39