计算机工程与应用 ›› 2021, Vol. 57 ›› Issue (1): 219-226.DOI: 10.3778/j.issn.1002-8331.1909-0015

• 图形图像处理 • 上一篇    下一篇

成对旋转不变的共生自适应完全局部三值模式

陈晓文,刘光帅,刘望华,李旭瑞   

  1. 西南交通大学 机械工程学院,成都 610031
  • 出版日期:2021-01-01 发布日期:2020-12-31

Pairwise Rotation-Invariant Co-occurrence Adaptive Complete Local Ternary Pattern

CHEN Xiaowen, LIU Guangshuai, LIU Wanghua, LI Xurui   

  1. School of Mechanical Engineering, Southwest Jiaotong University, Chengdu 610031, China
  • Online:2021-01-01 Published:2020-12-31

摘要:

针对成对旋转不变的共生局部二值模式(PRICoLBP)旋转不变性较差及其相关改进算法EPRICoELBP对光照变化和噪声干扰较为敏感的问题,提出了一种增强成对旋转不变的共生自适应阈值完全局部三值模式。通过自适应阈值局部三值模式(ALTP)将图像分成Upper和Lower模式;分别在两种模式中找出像素点LBP特征极大、极小值对应的邻域起始编码点,利用中心像素点与其LBP特征极大、极小值对应的邻域起始编码点作为方向矢量,来确定中心像素点的上下文共生点对;利用自适应阈值完全局部三值模式(ACLTP)提取Upper和Lower模式中共生点对的局部纹理信息;联合上下文共生点对的特征直方图训练卡方核支持向量机,进行纹理图像识别检测。在应用广泛的Brodatz、Outex(TC10、TC12-h、TC12-t、TC14)、CUReT、KTH_TIPS、UIUC标准纹理库中,该算法相较于原始的PRICoLBP算法和其他算法在分类准确率上均有一定的提升,且在添加了高斯噪声和椒盐噪声的KTH_TIPS纹理库中,该算法依旧保持了较高的分类准确率。实验结果表明,该算法对旋转、光照变化和噪声干扰具有较强的鲁棒性。

关键词: 共生点对, 局部三值模式, 成对旋转不变, 噪声鲁棒性, 自适应阈值

Abstract:

The texture feature extraction algorithm of Pairwise Rotation Invariant Co-occurrence Local Binary Pattern(PRICoLBP) has characteristics of poor rotation invariance, and its improved algorithm EPRICoELBP can enhance rotation invariance effectively, but it is sensitive to illumination change and noise. In order to solve the issues, an enhanced pairwise rotation-invariant co-occurrence adaptive complete local ternary pattern is proposed. Firstly, the image is divided into Upper pattern and Lower pattern by Adaptive Local Ternary Pattern(ALTP). Secondly, the neighborhood initial coding points corresponding to the maximum and minimum LBP features of the pixels are found in the Upper pattern and Lower pattern respectively. Thirdly, the context co-occurrence point pair of the central pixel point is determined by using the central pixel point and the neighborhood initial points of coding corresponding to the maximum and minimum LBP values of each pixel, respectively. Fourth, the local texture information of co-occurrence point pairs in Upper pattern and Lower pattern is extracted using Adaptive Complete Local Ternary Pattern(ACLTP). At last, a Chi-Square Kernel Support Vector Machine, which is trained with feature histogram of co-occurrence point pairs in context to detect the image texture categories. Compared with the original PRICoLBP algorithm and other algorithms, the proposed algorithm has a obvious improvement in classification accuracy on the Brodatz, Outex(TC10, TC12-h, TC12-t, TC14), CUReT, KTH_TIPS, UIUC texture databases, and on the KTH_TIPS texture databases with Gaussian noise and salt-and-pepper noise, the proposed algorithm still maintains a high classification accuracy. The experimental results show that the proposed algorithm is highly robust to rotation, illumination changes and noise.

Key words: co-occurrence point pairs, local ternary pattern, pairwise rotation-invariant, noise robustness, adaptive threshold