CAJ | 학술논문

采用组稀疏表示分类方法时，同类样本同时参与对测试样本的表示，忽略了类内样本间的相关性。提出了一种改进方法，该方法在块正交匹配追踪算法基础上，将样本间的相干系数作为参数，设置适当的阈值，对每次选取的样本进行判别，剔除与测试样本相关性较差的样本，优化算法的重建性能。在Yale B和ORL的数据库上的实验表明，与原有方法相比，改进后的方法得到的识别率较高，实验结果证明了该方法的有效性。
채용조희소표시분류방법시，동류양본동시삼여대측시양본적표시，홀략료류내양본간적상관성。제출료일충개진방법，해방법재괴정교필배추종산법기출상，장양본간적상간계수작위삼수，설치괄당적역치，대매차선취적양본진행판별，척제여측시양본상관성교차적양본，우화산법적중건성능。재Yale B화ORL적수거고상적실험표명，여원유방법상비，개진후적방법득도적식별솔교고，실험결과증명료해방법적유효성。
When the group sparse representation is used to face recognition, the same samples take part in representation of the test sample at the same time. The original method ignores the correlation between the samples. To solve this problem, an improved block orthogonal matching pursuit algorithm is presented. The presented algorithm uses the coherent coefficient of the samples as a parameter, setting the proper threshold value to select sample discrimination. Therefore, the reconstruction of the algorithm is optimized. Experiments on the Yale B database and the ORL database show that the recognition rate of improved algorithm is higher than the original one. The experiment results verify the validity of the proposed algorithm.