Eyeglass Removal And Region Recovery In Face Image

Because of wide applications in real world, the face recognition technology has received long-term attention. Although some improvements have been made in this area, face recognition in complicated environment remains an open problem, where pose variation, lighting condition variation, expression variation and complicated background exits. This paper mainly focuses on eyeglass detection, removal and region recovery. Because the dark-color thick-frame eyeglass has very contrastive color to face region and strong structure and texture character, it would cause performance loss in face recognition. Therefore it is necessary to remove eyeglass before face recognition. The main contribution of this paper is as follows:1. We propose the procedure of eyeglass detection, localization and removal to address the problem of dark-color thick-frame eyeglasses.2. Three different methods are proposed to solve the eyeglass detection problem, including multiple color gradient feature+SVM classifier, single gradient feature+linear classifier and HOG feature+SVM classifier. Experiments show the good results in detection rate.3. For eyeglass localization and removal, we propose the method of skeleton extraction with GVF Snake model and multi-image patch match method for eyeglass removal via neighborhood similarity, which yields good results.4. For the uncertainty problem in eye localization, a new eyeglass localization algorithm based on color and gradient information is proposed. Combining with recursive PCA reconstruction and multi-image patch match method, the proposed method yields good recovering results.