基于方差因子的图像偏色检测研究

Research on Color Cast of Image Based on Variance by Matlab

随着公安工作中视频与图像的广泛应用,准确判断视频与图像中犯罪嫌疑人的衣着颜色等特点十分重要,因此对于视频与图像中产生的图像偏色需要进行偏色检测与矫正。本文利用统计学的方法研究图像的偏色检测,统计了SONY-NEX-5R相机在室外晴天自然光、室外阴天自然光、室内晴天自然光、室内阴天自然光以及室内灯光五种场景和多个白平衡模式下的图像偏色因子以及偏色因子之间的方差值。通过比较各种场景和白平衡模式下的偏色因子及其方差值,基于任何场景下自动白平衡模式拍摄的图像不偏色的原则,得出了图像的三个偏色因子之间差值小于0.1且方差值小于0.01时,图像不偏色的结论。该研究为公安视频侦查工作中准确判断犯罪嫌疑人的衣着特点提供了依据。

With the extensive application of video and image in public security, the accurate judgment of the color characteristics of a suspect＇s clothes in video or image is very important for crime investigation. Therefore, the detection of color cast and its correction are often needed. This paper adopted statistical methods and Matlab to study the color cast and the variances of pictures taken with Sony-Nex-5R camera at white balance mode under five scenarios： clear or cloudy weather with natural light at outdoors and indoors, and indoor illumination with lamp. Based on the fact that few color casts occur when auto white balance being set, it is concluded that there is no color cast when the differences among the three factors （kr, kg, kb, originating from and representing the color cast of the RGB of an image） are less than 0.1 and meantime the corresponding variance （var） is less than 0.01. This study may lay a theoretical foundation on judging the color characteristics of a suspect＇s clothes shown in a video or image.