Object Scaling System Based On Segmentaion And Inpainting

For a long time,image content editing has been a classic problem in the field of computer vision and an important application in people's production and life.Image object scaling task is a sub-task of image content editing.The operation object of this task is an target object in the image,whose shape will be changed.It has strong application requirements in the fields of general image post-processing and human body shape transformation.However,the current mainstream implementation scheme has the problem of background distortion,which will lead to the extremely unreal effect.In order to improve this problem,this paper proposes an object scaling scheme based on image segmentation and image inpainting.In this scheme,image segmentation is used to obtain the pixel range of the target object,and the scaling is only implemented within the effective range.The foreground and the background are separated,so the background distortion problem can be improved.Then the background will be completed through image inpainting.In the task of image inpainting,the present methods have such shortcomings as unstable effect and large parameters quantity.These shortcomings make it hard to deploy the models,which makes image inpainting a weak part in the system.Therefore,this paper focuses on the study of image inpainting and proposes inner constraints network.Different from the existing global constraints model,the method proposed in this paper uses inner constraint to supervise the model.Experiments are designed to prove that inner constraints model achieves the best performance on multiple data sets.Based on the inner constraints model,an automatic human body beautification system with separated foreground and background is implemented in this paper.This system aims at human body beautification tasks,including human body segmentation,human landmarks,image Warping,image inpainting and post-processing modules.Finally,the operation results of the system are presented.The results show that the object scaling system based on image segmentation and image in painting designed in this paper has successfully solved the background distortion problem,and the background of the output image have reasonable semantic information and real structural information.