Quasi-Quantitative Evaluation of Overexposure in the Facial Image for Supporting Portrait Selection

Abstract

In the selection of portraits for ID and a profile, it is useful to retrieve high quality facial images from a lot of portraits. In addition, overexposure may give a negative impact on us because overexposed facial parts look unnaturally white; therefore the recognition for qualitative meaning of overexposure is important for retrieval of high quality facial images. To get qualitative meaning by a image analysis method that can evaluate images quantitatively, it needs a quasi-quantitative evaluation method that can use an index of quantitative evaluation having qualitative meaning. In this paper, we investigated the index for quasi-quantitative estimation of overexposure in the facial image on the basis of the heuristic assumptions, and defined the index as the overexposure intensity (OI). We developed an algorithm with Support Vector Machine (SVM) for judging the OI. The results of experiment conducted with 11 evaluators for analyzing images suggest that the proposed method can retrieve a facial image for ID and the profile with an F-measure of 96.2%.