Deep Portrait Art Illumination Transfer

In photography,the source of light is very important.The combined effect of illumination is like the contrast between light and dark.As long as the suitable illumination is taken,the beautiful image can be generated.The illumination plays an important role for an image by shaping the form,space,and sense of direction.The illumination is used as material basis to produce environment atmosphere,express feelings,and add new artistic glamour to picture.The illumination effect of face images is a hot topic recently,which is widely used in film and television production,image enhancement and so on.However,it is hard to master the control technique of illumination,even the professional photographer also needs years of research.Due to the complexity and trivialness of using the professional image editing tools,an art illumination transfer algorithm of portrait using convolution neural networks is proposed in this paper.The art illumination transfer of portrait is divided into three steps: extract feature points,and aligning the face based on the feature points.Then,extract the mask of the illumination by using the shadow detection algorithm.Finally,transfer the illumination.The improvement is mainly made for transferring illumination in this paper.First,in the process of transfer,the skin color in the target image is different from that in the reference image,so the skin color of the target image is influenced in a certain degree by the reference image,which leads to the change of the skin color in the target image.In order to solve this problem,a stylized algorithm is proposed based on luminance channel in this paper.The lightness layer is decomposed with the color layer by the algorithm,and only the illumination is transferred in the luminance channel.Thus,the interference of the reference image color can be avoided,and the skin color of the target image can be retained better.Second,in the process of transfer,the illumination is light after transfer for the small art illumination such as loop lighting.Although the shades can be adjusted by increasing the style weight,the detail preserving of the target image is influenced by the method.In order to solve this problem,a weighted space control algorithm based on semantic segmentation is proposed to adjust the shades of the shadow area in this paper.The weight is added to the style loss function in the shadow channel solely,which can adjust the shades of the shadow and avoid impacting the detail preservingFinally,the experiments of transfer are conducted based on four kinds of classic artistic illumination,which verifies the performance of the algorithm in this paper.Furthermore,the experiment results are compared with those by other algorithms,and the advantages and shortcomings of the algorithm are analyzed based on user quality of service.