Summary: | 碩士 === 國立成功大學 === 電機工程學系專班 === 100 === Personal identity system has a variety of researches and developments, while face recognition is one of the best ways to achieve the most natural, convenient and non-contact method. Face recognition accuracy is easily affected by the tilted face or reflective and shadow from eyeglasses on facial features. In this thesis, an integrated affine transformation and eyeglasses reduction method have been proposed, where affine transformation method makes tilted face into normal one and eyeglasses reduction method is realized by eyeglasses symmetrical features and image morphology operations. In recognition stage, sobel edge detection algorithm is first adopted to calculate the edge of the face image, and then figure out orthogonal projection coefficient of facial feature and its similarity measure by using Euclidean distance.
ORL (Olivetti Research Lab.) database is utilized to verify the proposed method. Face image normalization by affine transformation and eyeglasses reduction is effective in improving the error of face recognition that caused by the tilted face or worn eyeglasses. Good results are presented without sample training or manually creating eyeglasses frame models.
|