点击次数：

论文类型：期刊论文

发表时间：2016-01-08

发表刊物：NEUROCOMPUTING

收录刊物：EI、SCIE、Scopus

卷号：172

期号：,SI

页面范围：134-142

ISSN号：0925-2312

关键字：Face beautifier; Triangular mesh; Laplacian; Mobile devices; Social networking service

摘要：The pursuit of beautiful appearance for human greatly inspires the studies on digital beautifiers especially for the popularity of social networking services (SNS) in this booming era of mobile phones. Unfortunately, existing beautification techniques are not efficient enough for mobile devices that have limited computation and storage resources. In this paper, we propose a beautifier that is efficient on both space and time, and develop a mobile APP available online. We focus on the local beautification of facial chin that has a great effect on facial attractiveness, and simply use an averaged chin shape as the template. We leverage Laplacian as the constraint to remap the original shape to the target, and recursively apply the Laplacian constraint to the contour points from the coarse to fine scales. Thus, we only need to inverse a small matrix for the remap, which requires quite low expenses on space and time. Finally, we apply an efficient warping algorithm based on triangular meshes to generate a beautified face image. The performance of the algorithm is validated by a survey on 70 facial images since no objective metric for beauty exists. Twenty volunteers are invited to grade the results labeled as 'worse', 'unchanged' and 'better'. The results show that 81.50% of inputs are believed becoming more beautiful, and another 10.07% of them are considered unchanged. We also provide the comparisons on the beautification quality and time expenses with the algorithm based on radial basis functions (RBF). (C) 2015 Elsevier B.V. All rights reserved.