基于BM3D预处理的纹理光照不变特征提取算法  

Texture Illumination Invariant Feature Extraction Algorithm Based on BM3D Pre-processing

作　　者：占俊杰 尚赵伟[1] 

机构地区：[1]重庆大学计算机学院,重庆400030

出　　处：《计算机与现代化》2015年第10期50-54,共5页Computer and Modernization

基　　金：国家自然科学基金资助项目(61173130);重庆市自然科学基金资助项目(CSTC-2010BB2217)

摘　　要：为消除光照变化对图像结构信息的影响,提出基于三维块匹配(BM3D)预处理的纹理光照不变特征提取算法。基于BM3D算法的良好降噪特性,该方法首先对图像各颜色通道采用BM3D降噪,利用小波变换得到各颜色通道对数域的低频和高频分量,然后对低、高频分量分别运用小波和Bayes-Shrink算法降噪,并构造光照不变量,最后采用主成分分析(PCA)降低特征维度,取得特征向量,并利用K-最近特征线分类器进行图像分类。在Outex_TC_00014纹理数据库的实验结果表明,该算法具有较好的分类效果。In order to eliminate the effects of changing illumination on image structure information,a three dimensional blockmatching（ BM3D） pre-processing-based texture illumination invariant feature extraction algorithm is proposed. With the excellent denoising feature of BM3 D algorithm,this method firstly uses BM3 D denoising method to denoise each color channel,and uses wavelet transform to get the low and high frequency component of logarithmic domain of each color channel of image,secondly,uses the wavelet denoising method and Bayes-Shrink denoising method to denoise the low and high frequency components respectively,and thirdly constructs illumination invariant,using the principal component analysis（ PCA） to reduce the feature dimension,obtains the feature vector,and lastly uses the K-NFL classifier to classify image. Experimental results based on Outex_TC_00014texture database show that the proposed method delivers good classification performance.

关 键 词：三维块匹配(BM3D) 小波变换 Bayes-Shrink 主成分分析(PCA) K-最近特征线分类器 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]