基于Contourlet变换的加权小波特征抽取算法

特征提取是模式识剐中的一个关键问题.本文提出了一种基于Contourlet变换的特征抽取算法.Con-tourlet变换具有多方向性和各向异性,能以接近最优的方式描述图像的边缘和纹理.文中算法利用Contourlet变换各子带系数的统计特性,构造特征矢量.Contourlet变换获得的特征是图像的局部特征,图像不同子带特征的分类能力是不相同的,针对各子带数据的离散程度进行加权处理.为分类能力强的特征量赋予较大的权值.该算法充分利用样本的统计信息,简捷、高效,并具有一定的鲁棒性.将该算法应用于Brodatz图像库纹理图像的检索,验证了算法的有效性.

Weighted Wavelet Feature Extraction Algorithm Based on Contourlet Transform

Feature extraction is one of the key problems in pattern recognition system. A feature extraction algorithm is proposed based on the Contourlet transform. Contourlet transform can be used to effectively describe image edges and textures in both the location and the direction. The extracted feature vector has advantage of the statistical attribution of Contourlet transform coefficients. Local characters can be obtained with Contourlet transform and they have different discrimination qualities. The feature vector is weighted according to their degrees of the dispersion, and the feature with higher discrimination quality has bigger weight. The algorithm is used to texture retrieval and the promising results is obtained. In the retrieval experiments, a subset of the Brodatz image data is used. Finally, the experimental result shows improvements on the performance for various oriented textures.