快速和尺度稳健的纹理图像识别

传统的纹理图像识别方法很难同时获得较好的识别精度、实时性和尺度稳健性,不利于实际的工程应用.因此,提出一种快速和尺度稳健的纹理图像识别方法.该方法首先利用高斯滤波构造一个纹理图像的多尺度空间,然后利用完备的局部二值计数(Completed Local Binary Count,CLBC)算法对多尺度空间中的每个图像提取局部二值特征,并跨尺度提取局部二值特征的最大值,再将多个分辨率的特征相结合作为纹理图像的最终特征描述,最后利用最近子空间分类器(Nearest Subspace Classifier,NSC)判定纹理图像的类别.在基准纹理图像数据库上的实验表明,该方法在识别精度、实时性和尺度稳健性方面获得了很好的综合性能,有利于实际的工程应用.

A Fast and Scale-robust Texture Image Recognition Method

Traditional texture image recognition methods can't achieve high recognition accuracy,high efficiency and scale robustness simultaneously,which are difficult to be applied in real-world scenarios. To address the above problem,a fast and scale-robust texture image recognition method is proposed. First,a multi-scale texture image space is constructed by Gaussian filtering. Second,the completed local binary count( CLBC) algorithm is used to extract the local binary pattern of each image in the image space. Third,the maximum CLBC pattern is acquired by taking the maximum value of each pattern across different scales. Fourth,the maximum CLBC patterns in multi-resolutions are concatenated to describe the feature of original texture image. Finally,the nearest subspace classifier( NSC) is used to discriminate the class label of the texture image. The experimental results on the benchmark texture databases demonstrate that the proposed method can achieve high performance in terms of recognition accuracy,scale robustness and efficiency,which can be used in many practical applications.