Research On Eyeglasses Removal For Frontal Face Recognition

Automatic face recognition has become an important topic of research due to its potential use in a wide range of applications, such as access control, and human-computer interaction. One important requirement for successful face recognition is robustness to variations arising from different lighting conditions, facial expressions, poses, scales, and occlusion by other objects. Among occluding objects, glasses are one of the most common occluding objects and they have a significant effect on the performance of face recognition systems. Therefore, it is of utmost importance to have a research on how to remove the glasses on facial images to improve face recognition systems' performance.This paper makes a study of the background of project and basic principle of AFR technology, introduces the usual algorithm for face recognition. Then we review the classical method proposed by Saito Y.etc, which using PCA to reconstruct face image with glassless. We also discuss this method's limitations on reconstruct natural looking glassless facial images: The representational power of the PCA depends on the training set, therefore, errors would spread out to the whole reconstructed image, resulting in the degradation of the quality with some remaining traces of glass frames.On the basis of above-mentioned study, this paper proposes an algorithm using PCA+ICA to reconstruct face image: this algorithm fully use the characteristic of ICA, which is adept in expression of local detail on the face, is a very good supplement and the improvement to PCA. With the PCA+ICA to reconstruction, it can generate more natural looking and more lifelike face image with glassless. Further more, we propose a new method to deal with face image: recursive reconstruction and error compensation, which could generate glassless image with few distort and no glass trace. Then, we can provide high grade image input for later AFR system. This article has also realized the algorithm simulation and the test on MATLAB, the test result data also proven this algorithm's validity.