一种多尺度灰度共生矩阵的纹理特征提取算法

纹理特征作为图像的一个重要特征,在国画分类识别中的地位十分重要,但现有的纹理提取算法大多基于灰度信息而忽略了颜色信息。针对国画分类识别中纹理提取算法存在的问题,本文提出了一种多尺度、多色域的纹理特征提取算法,该算法结合了轮廓波变换和灰度共生矩阵的优点。为了对国画进行特征提取,该算法首先将国画图像转变到HSI色彩空间。然后,提取色调、饱和度、强度这三个色彩分量进行分区域操作,即提取每一个色彩分量的纹理特征。最后,将提取的3个特征向量融合并进行主成分分析降维。实验证明,与灰度共生矩阵相比,本文算法在国画分类识别方面查准率提高了7.5%,查全率提高了8.7%。实验表明多尺度灰度共生矩阵算法优于传统的灰度共生矩阵算法。

Multi-scale algorithm of texture feature extraction based on gray-level co-occurrence matrix

As an important feature, texture feature is very important in the category of Chinese painting, but the majority of the existing texture extraction algorithms is based on gray-scale information. To solve the problems in Chinese painting texture extraction classification, a multi-scale, multi-color domain texture feature extraction algorithm has been proposed. This algorithm combines the advantages of Contourlet transform and Gray-level Co-occurrence Matrix. In order to extract Chinese painting features by the new algorithm, the image is first transformed into HSI color space. Then, the three color components of HSI is extracted to sub-regional operation, which is to extract the texture characteristics of each color component. Finally, three feature vectors are integrated and the dimensionality of matrix is reduced using Principal Component Analysis. Experimental results show that compared with Gray-level Co-occurrence Matrix, the algorithm improves the precision of 7.5%, re-check rate increased by 8.7% in terms of Chinese painting classification. The experiment show that the new algorithm presented in this paper is better than Gray-level Co-occurrence Matrix algorithm.