当前位置:主页 > 科技论文 > 软件论文 >

采用多重特征蒙板的人像皮肤美化技术

发布时间:2018-03-25 08:42

  本文选题:人像皮肤美化技术 切入点:多重特征蒙板优化技术 出处:《浙江大学学报(工学版)》2017年12期


【摘要】:为了解决现有人像皮肤美化技术中出现的细节模糊以及特征丢失问题,提出一种基于构建多重特征蒙板的人像皮肤美化技术.第一重蒙板为利用高反差保留和强光迭加构建的面部显著瑕疵特征蒙板,基于该蒙板可预先去除人像面部的显著瑕疵;第二重蒙板为利用肤色检测构建的人像背景特征蒙板;第三重蒙板为利用Face++算法构建的面部关键区域特征蒙板;第二、三重蒙板的结合使用可以解决人像的背景特征及面部关键区域特征由于后续滤波处理而变模糊的问题;第四重蒙板为利用高反差保留构建的面部立体视觉控制蒙板,该蒙板可在之前的皮肤美化处理基础上还原人像的面部立体视觉特征.实验结果表明,该技术使用较小的皮肤美化系数即可去除瑕疵,并能在皮肤美化系数较大的情况下保留人像关键区域以及立体细节等真实感信息特征,从而解决人像皮肤美化过程中细节清晰度与瑕疵去除程度无法兼得的矛盾,最终达到人像美化的目的.
[Abstract]:In order to solve the existing technology in skin beautification portrait details and fuzzy feature loss problem, put forward a kind of beautification technology to build multiple features like skin mask. Based on the first mask to face significant defect feature is constructed by high pass and high light superposition Mongolia plate, significant flaw of the mask can be pre removed based on the portrait face; for the use of double mask mask portrait background characteristics building skin detection; face mask third key regional features constructed by Face++ algorithm mask; with second, third mask use can solve the key features of regional background features and facial portraits due to subsequent filtering and fuzzy problems; fourth for stereo vision control facial mask keep using high contrast mask construction, the mask before the skin beautification processing based on the reduction of portrait of the face Department of stereo visual features. The experimental results show that this technique uses smaller coefficient to remove defective skin beautification, and can beautify the large coefficient of retention of key area and three-dimensional details of realistic portrait information characteristics in the skin, so as to solve the contradiction of skin beautification portrait cannot both details defect removal process, the final achieve the portrait beautification.

【作者单位】: 浙江大学流体动力与机电系统国家重点实验室;浙江大学国际设计研究院;
【基金】:国家自然科学基金资助项目(51275460,61472355)
【分类号】:TP391.41


本文编号:1662373

资料下载
论文发表

本文链接:https://www.wllwen.com/kejilunwen/ruanjiangongchenglunwen/1662373.html


Copyright(c)文论论文网All Rights Reserved | 网站地图 |

版权申明:资料由用户2a603***提供,本站仅收录摘要或目录,作者需要删除请E-mail邮箱bigeng88@qq.com