计算机与现代化
計算機與現代化
계산궤여현대화
Computer and Modernization
2015年
10期
50-54
,共5页
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
三維塊匹配(BM3D)%小波變換%Bayes-Shrink%主成分分析(PCA)%K-最近特徵線分類器
삼유괴필배(BM3D)%소파변환%Bayes-Shrink%주성분분석(PCA)%K-최근특정선분류기
three dimensional block-matching(BM3D)%wavelet transform%Bayes-Shrink%principal component analysis(PCA)%K-nearest feature line classifier
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。
為消除光照變化對圖像結構信息的影響,提齣基于三維塊匹配( BM3D)預處理的紋理光照不變特徵提取算法。基于BM3D算法的良好降譟特性,該方法首先對圖像各顏色通道採用BM3D降譟,利用小波變換得到各顏色通道對數域的低頻和高頻分量,然後對低、高頻分量分彆運用小波和Bayes-Shrink算法降譟,併構造光照不變量,最後採用主成分分析( PCA)降低特徵維度,取得特徵嚮量,併利用K-最近特徵線分類器進行圖像分類。在Outex_TC_00014紋理數據庫的實驗結果錶明,該算法具有較好的分類效果。
위소제광조변화대도상결구신식적영향,제출기우삼유괴필배( BM3D)예처리적문리광조불변특정제취산법。기우BM3D산법적량호강조특성,해방법수선대도상각안색통도채용BM3D강조,이용소파변환득도각안색통도대수역적저빈화고빈분량,연후대저、고빈분량분별운용소파화Bayes-Shrink산법강조,병구조광조불변량,최후채용주성분분석( PCA)강저특정유도,취득특정향량,병이용K-최근특정선분류기진행도상분류。재Outex_TC_00014문리수거고적실험결과표명,해산법구유교호적분류효과。
In order to eliminate the effects of changing illumination on image structure information, a three dimensional block-matching (BM3D) pre-processing-based texture illumination invariant feature extraction algorithm is proposed.With the excellent denoising feature of BM3D algorithm, this method firstly uses BM3D 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, u-ses the wavelet denoising method and Bayes-Shrink denoising method to denoise the low and high frequency components respec-tively, 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_00014 texture database show that the proposed method delivers good classification performance.