Automatic Face Replacement In Uneven Lighting Situation

Automatic face replacement belongs to the field of image synthesis.It replaces the identity features of the face in the image while keeping other features in the image unchanged.It is widely used in the fields of AI security and privacy protection,film and television production,virtual experience,etc.It is a hot and frontier direction of research at home and abroad.This method first captures the face to be replaced in the image,then performs identity replacement on the face,and finally merges the face replacement image into the source image.When actually performing face replacement operations,it is often necessary to face complex imaging conditions,in which uneven lighting is a very common and complex scene.Uneven lighting will not only affect the performance of face capture,but also make the fused face image not real enough.Therefore,the fusion degree of face replacement in real scenes is generally not high,and often requires manual debugging.In this paper,we will study and implement the automatic face replacement algorithm under uneven lighting situation,and carry out the following work around face replacement,face capture,and face fusion:1)Investigate and sort out the main research results in the field of face replacement,divide the face replacement methods from a new perspective,and analyze and compare the advantages and disadvantages of various methods and applicable scenarios.Especially for the face replacement method based on deep learning,this paper conducted visual quality comparison and result comparison analysis.The field of face replacement is still in rapid development,this paper provides a reference method overview,hoping to help later researchers.2)Aiming at the difficulty of capturing faces in uneven lighting images,a robust face capturing system is implemented,which expands the traditional face capturing algorithm based on ideal lighting assumptions under uneven lighting situation.As an important pre-step for face replacement,face capture lacks a complete and good-performance implementation.For uneven lighting situation,this paper combines the image lighting preprocessing algorithm in the traditional face capture framework to achieve a real-time face capture system with good results,which can help any scene that needs to capture faces.3)Aiming at the unsatisfactory fusion of the face replacement image with uneven lighting and the background,a face replacement method based on the key points of the face is proposed.This method simplifies the illumination of the face replacement image and can obtain high-quality face replacement results.High-quality face replacement needs to meet the actual nature of the imaging level and the accurate identity replacement of the semantic level,and many existing methods only focus on one aspect.The face replacement method based on the key points of the face helps to improve the image quality from the above two aspects and performs well in uneven lighting situation.The above work has been verified by experiments.In uneven lighting situation,the face replacement system can capture faces robustly from the image,synthesize real facial images,and can accurately replace facial identities and retain non-identity features in facial images.