人脸图像个体特征统计与价值评估

Facial Image Rendering Individual Features Oriented to Statistics and Value Evaluation

法庭科学人脸比对结果是一类重要的证据,针对人脸比对所用的特征,本文关注其中人脸个体特征(色素、痣等)在比对中的价值评估。通过对公开人脸图像数据集中的个体特征进行统计分析,计算相同和不同身份(ID)条件下人脸图像上个体特征出现概率、出现在特定位置的概率等,最终通过似然比来衡量个体特征匹配的价值。结果表明在开展实验的人脸数据分辨率条件下,单个人脸个体特征匹配相关的似然比值能够达到10的4次方数量级。人脸个体特征在比对中具有重要的价值。在实验条件下,两个个体特征相关的似然比能够超过10亿量级,能够支撑同一认定的肯定结论。

Facial comparison is often required and carried out for forensic purpose.Such an operation relies on the morphological analysis and evaluation into facial features.Therefore,there is the necessity to deploy in-depth researches on facial features,especially the individual ones.An evaluation approach was tried in this paper of focusing on the individual features(pigments,moles,and other specifics)in facial images.Through statistical analysis of the individual features from the facial image dataset of open source,calculations were performed about the probabilities of individual features appearing on facial images and/or even at some specific sites under the same identity and/or different ones.Subsequently,the individual features were meausred of their matching values by likelihood ratio manipulation:the probability ratio of individual features appearing in a specific location under the above set two conditions.With the facial data resolution in this experiment,the likelihood ratio was able to reach the level of 104 in relation to a single facial individual feature being matched,and even exceed 1 billion with matching two facial individual features–a digital magnitude supporting the confirmative conclusion of same source/identity.The facial images adopted in this paper hold a resolution of 178×218,the size of the facial area is accordingly about 120×150.The resolution utilized with this research is believed to be ideal for facial image comparison to match the appearing area according to the individual features.If resolution-higher facial images are available,the results rendered from this research may serve as a lower threshold.The facial images used for this research are from public dataset with different imaging conditions,displaying the face images of different race,gender and age characteristics among the number of IDs(identities)of 10177.Although the public dataset tested with this research attains such a scale,a question is still open to answer about whether the distribution of its contained facial individual features and rendered matching can represent a wider range of facial data.