人脸检测建模源照片筛选方法

A method of selecting source data of pictures in the modeling of face detection

安防系统输入数据质量各异,部分质量差的照片会引起误报率升高,如何将这部分数据优化排除,筛选用于人脸建模的安防系统数据源是本文的目的。采用量化照片评价指标的方法,通过对照片中人脸部分的清晰度、对比度等质量因素进行评价,将其量化并排序从而得到数据源质量序列。在软件工程化实践中对随机抽取的100张照片质量评价验证,其评价结果准确率可达95%。以文中所提到的方法对采集数据优化后,其建模后的数据能有效地降低比对过程中误报率,提高安防系统识别的准确率。

The quality of input data is quite different,some poor quality pictures will increase the rate of false alarm in the safety defense system. The purpose of this paper is to optimize these data and choose the source data for modeling the face detection system. Using the method of quantifying the picture evaluation indexes to measure the pictures with the factors influencing quality of face,such as resolution,contrast ratio,etc.,then quantify and sequence these factors,and thus derive the quality sequence of data source. In the software engineering practice,100 images were selected randomly to evaluate their quality,proving that the accuracy of evaluation reaches 95%. The data collected by the method mentioned above were optimized and then used for modeling,proving that the false alarm in the process of comparison can be reduced effectively and the accuracy of identification of safety defense system increased.