【作者】 郑振谦；

【导师】 张二虎；

【作者基本信息】 西安理工大学 ， 信号与信息处理， 2009， 硕士

【摘要】 画像识别是人脸识别的一个全新分支。在刑侦破案或反恐追逃等应用领域,犯罪嫌疑人的照片往往无法直接获取,通常情况下会通过目击者和画家的合作得到其模拟画像,然后在已有的照片数据库中进行检索,进而实现犯罪嫌疑人身份的确认和识别。由于画像与照片的产生机理和信息表达方式不同,两者之间存在较大的几何形变及纹理和灰度差异,若直接采用画像与照片之间的匹配程度进行识别势必产生很大误差。因此,如何减小二者的差异成为画像-照片识别研究的重点和难点,从而画像生成成为人脸画像-照片识别中的关键技术。本文首先深入研究了图像类推的算法,并在此基础上提出基于图像类推的模拟画像的生成方法,由于传统的图像类推方法在合成模拟画像时在合成时间和合成效果上都不能满足实验的要求,针对上述问题本文对传统的图像类推算法进行了改进,在实验中本文分别从分块类推、参数选择、像素的搜索邻域与像素特征向量的选取等几个方面对传统的图像类推进行改进,使得它能更好的利用于模拟画像的生成,最终实验得到了比较满意的结果。最后本文采用了基于小波加DCT变换的人脸模拟画像的特征提取方法,它通过对经过小波变换的人脸模拟画像的低频子图进行DCT变换,然后利用最近邻法进行分类检索,在本文中93幅人脸模拟画像数据库中进行实验。实验表明这样的方法能够获得比直接应用DCT变换提取特征更好的识别率。更多还原

【Abstract】 Sketch recognition is a new branch of face recognition. In applications of case-solving and suspect-searching, to deal with the problem that the photo of a suspect is unavailable in most case, a simulated sketch is generated by cooperation of artists and eyewitnesses as a substitute, so that the content-based image retrieval for identification is conducted in the existing photo database. Due to large differences between a photo-sketch pair caused by different generation mechanism and information expression manner, it is impossible to recognize by simply matching the photo-sketch pair. As a result, how to reduce the difference becomes important for sketch-photo recognition, and sketch synthesis becomes the key technique of recognition.Firstly the paper studies the algorithms of image analogies and presents a face sketch synthesis method based on image analogies. Because the traditional image analogy method lost too much time and the results of synthesized sketch is not ideal. So a novel image analogy algorithm is proposed in this paper. To improve the synthesis performance, some novel idea is introduced on sub-block dividing image analogy, chosen of image analogy parameters, neighborhood of the pixel, and the eigenvector of the pixel in the experiment. Experimental results show that the novel algorithm can achieve good results in face sketch synthesis.Finally, a face sketch recognition method based on wavelet transform and discrete c osine transform is presented. First the original image was decomposed by wavelet transf orm. Then, the DCT method can be directly used to the low frequent sub-image obtain ed form on the previous step. In the end, the classical K-Nearest Neighbor method is p erformed on the reductions for the pattern classification. Experiments over a data set co ntaining 93 people show that the above mentioned algorithm performs higher accuracy o f recognition than discrete cosine transform method.更多还原