应用小波去噪提取人脸光照不变量.docxVIP

应用小波去噪提取人脸光照不变量 任昆 REN Kun 住房和城乡建设局档案馆,湖南省湘潭市411101 Archives of Housing, Town and Country Construction Bureau, Xiangtan City, Hunan Province 411101, China REN Kun. Using Wavelet De-noising To Extract Face Illumination Invariant. Computer Engineering and Applications Abstract: Because of the performances of face recognition system can sharply decrease due to the influence of outdoor illuminating environment condition changes; therefore, we need to draw the face illumination invariant that is not affected by the illumination to recognize faces. The core of illumination invariant drawing is to do de-noising of the face image in logarithmic domain. The effect of dc-noising can directly affect the quality of illumination invariant. Regarding the problem of de-noising, it puts forward a new wavelet de-noising model. Firstly, it reflects the image to the logarithmic domain, and does two-dimension wavelet decomposition in the logarithmic domain, then gets four sub-graphs, including LL, LH, HL and HH. Secondly, it separately calculates LH^ HL^ HH sub-graphfs every weighted average value of all coefficient values in neighborhood of a centre coefficient which is belong to the sub-graphs; The probability of being noise coefficients is large when the weighted average value smaller than a preset thresholds , otherwise the possibility is smalle匚 Finally, keep the sub-graph LL unchanged; multiply the wavelet coefficient which has large possibility of being the noise coefficient in sub-graphs LH, HL and HH with a auto adaptive constriction factor a. As proved by the experiments, using this wavelet de-noising method to get the Illumination Invariant to do the face recognition, the recognition rate is better than other methods. Key words： illumination; wavelet de-noising; weighted average value; illumination invariant; lace recognition 摘要：针对人脸识别系统的性能会随室外光照环境条件变化的影响急剧下降，因此需要提取 不受光线影响的人脸光照不变量，用于识别人脸。光照不变量的提取的核心就是在图像对数 域去噪，去噪效果的好坏直接影响捉取的光照不变量的质量。针对去噪问题，提出了-?种新 的小波去噪模型。首先，将图像映射到对数域，在对数域进