用于纹理特征提取的改进的成对旋转不变共生局部二值模式算法

Improved pairwise rotation invariant co-occurrence local binary pattern algorithm used for texture feature extraction

针对用于纹理特征提取的成对旋转不变共生局部二值模式(PRICoLBP)算法计算特征维度大、旋转不变性较差、对光照变化敏感的问题,提出一种融合局部纹理信息的改进PRICoLBP算法。首先,分别最大化和最小化图像像素点的二值序列,得到两个邻域像素点的坐标,由中心像素点坐标和得到的邻域像素点坐标计算出共生点对的坐标;其次,利用完备二值模式(CLBP)算法提取图像的每个像素点的纹理信息。在相同分类器下,对Brodatz、Outex(TC10,TC12)、Outex(TC14)、CURe T和KTH_TIPS数据库的分类实验中,所提算法的识别率比PRICoLBP算法分别提高了0.17、0.24、2.65、2.39和2.04个百分点。实验结果表明,所提算法在处理纹理旋转变化、光照条件多样的图像时具有较好的识别效果。

The texture feature extraction algorithm of Pairwise Rotation Invariant Co-occurrence Local Binary Pattern （PRICoLBP） has characteristics of high computing feature dimension, poor rotation invariance and sensitivity to illumination change. In order to solve the issues, an improved PRICoLBP algorithm was proposed. Firstly, the coordinates of two neighboring pixels were obtained by respectively maximizing and minimizing the binary sequence of image pixels. Then, the position coordinates of co-occurred pixel points were calculated via the position coordinates of the center pixel and the two neighboring pixels. Secondly, the texture information of every image pixel was extracted through utilizing the Completed Local Binary Pattern （CLBP） algorithm. Compared with PRICoLBP, the recognition rate of the proposed method was improved respectively by the percentage points of 0.17, 0.24, 2.65, 2.39 and 2.04, on the image libraries of Brodatz, Outex（TCl0, TC12）, Outex（TC14）, CURET and KTH TIPS under the same classifier. The experimental results show that the proposed algorithm has better recognition effect for the images with texture rotation variation and illumination change.