【摘要】 人脸图像的庞大信息量使其不适合于直接识别,根据矢量量化技术的特点,提出基于LBG算法的人脸识别方法。将训练样本拼接后分割子图像作为码书训练的输入矢量,用经典的LBG算法训练码书,根据测试样本对各码书的失真度对图像进行分类识别。ORL人脸图像数据库的实验结果表明,这一方法在不同分辨率、不同码书长度下都优于传统的PCA算法,在低分辨率下亦能获得很好的识别效果,能较好地解决人脸这样一类复杂的图像识别问题。更多还原

【Abstract】 The large amount of information in the face image is not suitable for direct identification. A new method of face recognition based on LBG algorithm is presented according to the characteristic of the vector quantization. The codebooks are tained by the classical LBG algorithm, with the training samples as the input vectors after mosaics segmentation sub-images. The images are categorised and recognized in accordance with the distortion of training samples respectively corresponding to their codebooks. The experiments implemented on the ORL face database demonstrate that this method is superior to the traditional PCA algorithm in both cases of different resolutions and different codebook lengths and that the method has a desirable performance in recognizing the complicated images such as human-face as well as in the low resolution.更多还原