基于DCT域加权最大类间边缘准则的模式识别

Pattern recognition based on DCT domain weighted pairwise maximum margin criterion

本文提出一种加权最大类间边缘准则(WPMMC)特征提取算法,它是最大类间边缘准则(MMC)的一般形式;同时证明了这种新的特征提取方法在DCT域的提取结果与空间域是相同的。对于JPEG格式的压缩域图像,WPM-MC直接在DCT域应用,可以省去计算IDCT的过程。最后,在人脸识别系统中检验这种基于DCT和WPMMC的特征提取方法。实验结果表明WPMMC方法比PCA、LDA和MMC方法提取的特征更为有效;保留部分DCT系数的识别结果略高于保留全部DCT系数和在空间域应用WPMMC方法的识别结果。

In this paper, a new feature extraction method of weighted pairwise maximum margin criterion （WPMMC） was introduced, it was the general form of maximum margin criterion （MMC）. It was proved that WPMMC can be directly implemented in the DCT domain and the results were＇exactly the same as the one ob- tained from the spatial domain. For compressed JPEG images, WPMMC can be directly implemented in the DCT domain so the process of calculating DCT can be omitted. At the end of this paper, the proposed WPM- MC method and its application in the DCT domain were tested in face recognition. Experimental results showed that the performance of WPMMC was more efficient than that of PCA, LDA, and MMC. The performance of WPMMC remaining parts of the DCT coefficients in the DCT domain can reach a better result than using the whole DCT coefficients or in the spatial domain.