Design And Implementation Of Smart Body Beautifica-Tion APP Based On Prior Knowledge And Deep Features

In recent years,the facial beauty function on smart phones has been favored by users.The major beauty apps have taken advantage of this to get a large number of users and create huge commercial value.Then,with the rapid development of deep learning,many advanced algorithms in the field of computer vision were applied to beautify people body in pictures with well performance.Smart body beautification has become a selling point of these beauty apps.In order to get more inexpensive and more natural results for body beautification,this thesis designs and implements a set of body beautification solutions based on prior knowledge and deep features,and finally develops a new smart body beautification app.There are two key issues that the smart body beautification apps needs to solve:Firstly,how to accurately obtain the body structure information of the characters in the picture,which is different from the solution of beauty apps mentioned before,using the private large-scale human contour key point data sets.This thesis proposes a smart body beautification solution based on human pose estimation and portrait segmentation.The solution fully utilizes the existing publicly available official data sets without additional large-scale annotation,and can also accurately obtain the body structure information of the characters in the picture.The cost is low.Secondly,how to achieve real and natural character body deformation.Based on the MRSL non-rigid image deformation algorithm,this thesis uses the character keypoints information and portrait mask information to realize an intelligent body deformation method,which has a faster processing speed and have achieved the effect for body shape deformation.In addition,in order to adapt to the computing power of mobile devices,the human pose estimation model and portrait segmentation model are simply lightened to increase the inference speed.At the same time,some mobile deep learning frameworks are used to accelerate model inference by GPU.Finally,it basically meets the requirements of the low response delay on the app,and improves the user experience of the app.Finally,we test the developed smart body beautification app in practical scenarios.The experimental results show that our built app has achieved a good beautification effect of the characters,without destroying the authenticity of the picture.The processing speed of the app is quick enough for the normal use,practical value and the application prospects are high.