Study Of Eye Location And Eye State Recognition Algorithms With Glasses

Eyes stand for various information what could be processed into more advanced information of human being. The location and states of eyes is the premise to get this information. When wearing glasses, the framework changes the feature of the area around eyes and the glasses reduce the image quality. So the performance of the most common eye location algorithms and eye states recognition algorithms emerge in a significant decline.A morphology-based glasses detection algorithm is proposed is this paper to detect whether a face image wearing glasses. At first,the Gaussian filter is used to reduce the noise of the input face image. Then the area above the nose is selected according to the face orientation. An edge detection and morphological processing is adopted in this area to determine whether the target wearing glasses. Experimental results show that the proposed glasses detection algorithm can achieve an accuracy rate of 95.91%, which can effectively detect whether wearing glasses.To improve the accuracy when wearing glasses, an eye location classifier with glasses is trained and then an eye location framework based on glasses detection is proposed. At first, the new input face image is detected whether wearing glasses or not. When wearing glasses, the eye location classifier and eye state recognition classifier which are trained with eye samples with glasses are adopted. Otherwise the eye location classifier and eye state recognition classifier which are trained with eye samples without glasses are adopted. Experimental results show that the eye location algorithm in the framework proposed in this paper can achieve an accuracy rate more than 90% and a low error rate.An eyes state recognition algorithm based on feature fusion is proposed in this paper to overcome the effects of head movement and complex illumination. At first, eye samples are selected in eye database with and without glasses respectively to train different eye states classifiers. Then the Pseudo-Zernike moment features and Gabor features of these samples were extracted. Principal Component Analysis(PCA) is adopted to reduce the high dimension of Gabor feature. At last, fuse the Gabor features and Pseudo-Zernike moments feature in series and enter the fusion feature into the Support Vector Machine(SVM) to train eye states recognition classifiers. Experimental results show that the proposed fusion feature can achieve an accuracy rate of 98.95% when wearing glasses and an accuracy rate of 96.75% when without glasses, which can meet the application requirements.