摘要
在图像纹理分类方面,近些年局部二值模式(Local Binary Patterns,LBP)以及其变体已经被证明有其独有的优势,但是它仍然存在两个明显缺陷:连续旋转不变性问题以及对噪声鲁棒性差的问题。针对这两个缺陷,提出了一种新的纹理描述符SI-LCCP进行改进。该方法结合了图像表面形状指数(shape index,SI)具有旋转不变的特性的特点,对原有LBP进行补充。同时,将该算法与噪声鲁棒性强的局部凹凸模式(local concave-convex pattern,LCCP)进行算法融合。实验证明提出的算法有效地改善了上述缺陷,在四个常用的纹理图像数据集上均取得了更高的识别效果。
In the aspect of image texture classification, the unique advantages of local binary patterns(LBP)and its variants have been proved in recent years, but it still has two obvious defects: continuous rotation invariance and poor robustness to noise. A new texture descriptor SI-LCCP is proposed to improve the two defects. This method combines the characteristics of image surface shape index(SI)with rotation invariance, and complements the original LBP. At the same time, the algorithm is fused with the local convex and convex pattern(LCCP)with strong noise robustness. Experimental results show that the proposed algorithm can effectively overcome the shortcomings, and achieve higher recognition results on four commonly used texture image data sets.
作者
申柯
陈熙
张云飞
Shen Ke;Chen Xi;Zhang Yunfei(School of Big Data and Computer Science,Guizhou Normal University,Guiyang 550025)
出处
《现代计算机》
2022年第24期71-77,共7页
Modern Computer
关键词
图像纹理识别
LBP
形状指数
连续旋转不变性
image texture recognition
LBP
shape index
invariance of continuous rotation
