基于NIB2DPCA的彩色图像过完整分块特征抽取方法

NIB2DPCA-based Color Image Feature Extraction Method with Over-complete Divided

在小空间占用的快速彩色图像的特征抽取方法和模块化FPCA(M-FPCA)彩色图像特征提取方法的基础上,结合最新的过完整表示思想,提出基于NIB2DPCA的彩色图像过完整分块特征抽取新方法。该方法对彩色图像进行过完整分块,然后对子图像模块从R、G、B三个信道用NIB2DPCA方法进行特征提取、重构,并进行多模块融合,最终获得分类的特征矩阵。该方法提取的信息量远大于原图像,提高了彩色图像的识别率。通过在FEI和CVL标准彩色人脸数据库上的对比实验表明,所提出方法的人脸识别准确率比文献[14]中的小空间占用的快速彩色图像特征抽取方法提高约4%,比文献[19]中的彩色图像M-FPCA方法提高约5%。

Based on color image feature extraction method with small space occupying and fast speed and color image feature ex- traction method with Modular Factored Principal Components Analysis （color M-FPCA）, combined with the latest over-complete representation idea, a novel method named NIB2DPCA-based color image feature extraction method with over-complete divided was proposed. This method can make the color image over-complete divided into smaller modular images, then NIB2DPCA is employed to extract feature information from three channels of a given sub color image module respectively. Then the three extracted feature matrices are reconstructed and multi module integrated. Finally, the classification feature matrix is gotten. The informa- tion amount extracted by the novel method is much larger than the original color image by this method and it improves the recogni- tion accuracy of color image. The results of contrast experiments on CVL and FEI color face databases show that, the proposed method can obtain a higher accuracy by about 4% than color image feature extraction method with small space occupying and fast speed in the reference[ 14] and also obtain a higher accuracy by about 5% than color M-FPCA in the reference[ 19].