摘要： 分块主成分分析(BPCA)方法忽视模块间特征向量的质量差异，在遮挡环境中的识别率较低。为此，提出基于识别信心决策融合的分块PCA人脸识别方法。该方法将人脸图像划分为子模块，利用PCA和最近邻分类器分别识别各模块，得到模块识别结果及其对应的识别距离，依据识别距离区分各模块识别信心的大小，最终决策结果判定为对应最大识别信心的模块识别结果。AR人脸库的实验结果表明，该方法在遮挡环境中的识别率明显优于PCA和BPCA方法，对遮挡环境的适应能力显著增强。

关键词: 人脸识别, 遮挡环境, 分块策略, 主成分分析, 决策融合, 识别信心

Abstract: The conventional Block Principal Component Analysis(BPCA) method ignores the difference in quality between extracted local features, and it gives a weak performance under condition of occlusion. A recognition confidence based decision fusion algorithm for block PCA is proposed. The approach divides images into blocks, and each block is processed by PCA and nearest neighbor classifier independently. The recognition confidence is proposed to measure the confidence of individual results, and the final decision is made in favor of the individual result with respect to maximum recognition confidence. Experimental results based on the benchmark AR database indicate that the proposed method can achieve better recognition rates than the conventional PCA and BPCA methods.

Key words: face recognition, occlusion condition, block strategy, Principal Component Analysis(PCA), decision making fusion, recognition confidence

中图分类号: 

• TP18