Shadow Detection and Effect Evaluation on a Face Image Using Automatic Threshold Selection
Trần Anh Tuấn, Mingyu Song, Jin Young Kim · 한국정보기술학회논문지 · 2011
In this paper, we propose a novel method for detecting shadow regions from a single face image thereby calculating their effect value on the human face. From an input image, we first make the nonlinear color transformation and use some bounding rules to segment the image into all face candidate regions. Secondly, after applying some techniques to reduce noise, we construct the only skin map image by subtracting the eye-mouth map image from the face map image. Unlike previous shadow detection algorithms, we introduce a novel and simple one to extract shadow regions from face image by utilizing the within class variance in Otsus method. The automatic threshold selection, which is the selection for the optimum value in within class variance, will help us to extract all shadow regions from a human face. Finally, based on the extracted shadow region, we make some evaluation about the strength effect of the shadow on one face. The experimental results demonstrate that, for different lighting variations in indoors and outdoors, our approach also gives a promising result with all extracted shadows accurately and robustly.