摘要： 二进小波变换利用模局部极大值来提取图像的多尺度边缘信息。该文研究了二进小波系数的统计特性，提出并验证了二进小波系数直方图服从于指数分布。分析了二进小波系数的一阶、二阶统计矩(共生矩阵)特性并将其应用于纹理特征提取。理论分析和实验说明，采用Manjunath方法和二阶统计矩方法的二进小波在纹理图像检索方面优于单小波。

关键词: 二进小波变换, 纹理, 共生矩阵, 统计特征, 纹理分类

Abstract: The dyadic wavelet transform utilizes local modulus maximum of a wavelet transform to obtain multiscale edges information. This paper analyzes the statistical characteristics of the wavelet detail coefficients of the dyadic wavelet transform, and brings forward the wavelet coefficients histogram of nature texture image which can be modeled by a family of exponentials. It studies the ways to extract the texture features using the first-order and second-order (co-occurrence) based on the dyadic wavelet transform. Experimental results show that methods of Manjunath and second-order of the dyadic wavelet are better than that of scalar wavelet on texture image retrieval.

Key words: dyadic wavelet transform, texture, co-occurrence matrix, statistical characterization, texture retrieval

中图分类号: 

• TP391.40