Real-time Portrait Segmentation Technology And It's Application Among Virtual Background Photography

Self-photography under virtual background is the application of augment reality technology in the amusement photography field. In virtual background self-photography system, the software of image analysis and processing in virtual background is the kernel of this system. The software mainly includes processing of video capture,portrait extraction,face color correction,image fusion, etc. By now, domestic and abroad virtual background photography is sill stayed at stage of static photography and off-line synthesize. The study of full-blown dynamic real-time virtual background photography system is a blank branch.To solve the technology difficulty of the virtual background real-time photography, this article analyzes the relationship between portrait quality and display speed scientifically, integrates the former research result, and utilize the high-performance DaHeng CG300 capture card, then import the quantum genetic algorithm to the portrait division method. At the same time we apply diverse super image real-time processing technology so that the performance of dynamic fusion display is improved much well. At last, with this technology, we successfully satisfied the requirement of virtual background real-time photography in lots of actual testing Furthermore, to acquire the content effect of imaging, and to correct the face color cast, this article firstly analyzes the color perception principle according to color vision theory and photometry theory, then provide a keystone of light arrangement in small photography room, and realize light arrangement equably. The second part of this article concludes and summarizes diversified face detection methods, and studies the face detection technology aiming at blue background, then, ground on that, brings forward a way of face color correction based on lamp-house color temperature estimation.This article provides a great deal of experimental picture and simulation effect. All of these can be used to testify the validity of corresponded algorithms, and they also can use for references for the subsequent research.