Eyeglasses removal based on attributes detection and improved TV restoration model
- PDF / 3,975,987 Bytes
- 22 Pages / 439.642 x 666.49 pts Page_size
- 84 Downloads / 164 Views
Eyeglasses removal based on attributes detection and improved TV restoration model Minghua Zhao1,2
· Zhe Zhang1 · Xin Zhang1 · Lili Zhang1 · Bing Li1
Received: 11 February 2020 / Revised: 26 July 2020 / Accepted: 25 August 2020 / © Springer Science+Business Media, LLC, part of Springer Nature 2020
Abstract Eyeglasses are common occlusions on face images. The detection of eyeglasses attributes and eyeglasses removal are key factors for correct automatic face recognition. A method for eyeglasses removal based on attributes detection and improved Total Variation (TV) restoration model is proposed in this paper. First, existence of eyeglasses frames is determined based on the width-length ratio after location of the eyeglasses; second, color coefficient and skin likelihood ratio are defined and color information is determined; third, bright index is defined and the reflective areas are detected based on luminance information. Finally, for rimmed, colorless and non-reflective eyeglasses, influence function based on gray difference ratio is defined to improve TV restoration model for eyeglasses removal. Experimental results show that our proposed method can not only discriminate the existence of the frame, but also detect color information and reflective areas accurately. In addition, the eyeglasses removal effect is superior to the traditional methods. Keywords Eyeglasses attribute · Reflective area detection · Total variation model · Eyeglasses removal · Image analysis
1 Introduction Face recognition technology is an important biometric recognition technology. After decades of development, many face recognition technologies have achieved high accuracy in normal conditions and have been applied to different practical applications [9, 24, 29]. However, many factors such as illumination, face posture, expression and facial occlusions have limited face recognition performance, among which eyeglasses are the most common occlusion objects of face images [28, 37]. Therefore, how to effectively remove the eyeglasses from face images has become one of the urgent problems in face recognition Minghua Zhao
mh [email protected] 1
School of Computer Science and Engineering, Xi’an University of Technology, Xi’an, 710048, China
2
Shaanxi Key Laboratory of Network Computing and Security Technology, Xi’an, Shannxi, China
Multimedia Tools and Applications
technology [5, 16, 18, 20, 31]. Good eyeglasses removal effect will greatly improve the accuracy of face recognition. Eyeglasses consist of frames and lenses, both of which have great influence on face recognition rate. Therefore, the removal of eyeglasses should be based on eyeglasses location and eyeglasses attributes detection [3, 13, 14, 21]. The main methods of eyeglasses removal can be divided into two categories, one is based on Principal Component Analysis (PCA) and the other is based on image restoration. The main principle of the first category is to get the eigenvector matrix using several face images without eyeglasses, and then the eyeglasses removal results can be
Data Loading...
