摘要
提出一种基于深度学习目标分割的动态场景高动态范围渲染HDR(high dynamic range)影像生成方法,结合行人目标匹配与光流映射变化的信息对场景中行人目标运动状态进行判断,生成用于后期融合的动态场景掩膜。HDR静态场景使用加权融合的方法生成,HDR动态场景则使用低曝光影像中该区域曝光质量较好的一张填充,最后利用泊松融合算法平滑静态区域与动态区域之间的过渡部分,得到一张"无鬼影"、影像亮度均衡并且各处细节清晰的HDR影像。算法的创新点是基于目标分类,比传统的基于像素的算法更具有可控性。
In this paper, we proposed a HDR image generation method for dynamic scenes based on deep learning object segmentation. This method combined with pedestrian matching and optical-flow mapping to judge the motion state of each object, and then generated a dynamic-scene mask for later fusion. We generated HDRI of static-scene by weighted fusion, while generated HDRI of dynamicscene by LDRI with the best well-exposure of the region. Finally, we utilized Poisson image fusion algorithm to smooth the transition zone, and obtained a "ghost-free" HDR image with balanced luminance and more details. Compared with the existing algorithms, the innovation of this method is based on the object classification and more controllable than the traditional pixel-level algorithms.
引文
[1]Mitsunaga T,Nayar S K.Radiometric Self Calibration[C].IEEE Computer Society Conference on Computer Vision and Pattern Recognition,1999
[2]Grossberg M D,Nayar S K.Determining the Camera Response from Images:What is Knowable[J].IEEE Transactions on Pattern Analysis and Machine Intelligence,2003,25(11):1 455-1 467
[3]Zhang W,Cham W K.Gradient-directed Composition of Multiexposure Images[C].Computer Vision and Pattern Recognition,IEEE,2010:530-536
[4]Hu J,Gallo O,Pulli K,et al.HDR Deghosting:How to Deal with Saturation[C].IEEE Conference on Computer Vision and Pattern Recognition(CVPR),2013,9(4):1 163-1 170
[5]Dai J,He K,Sun J.Instance-Aware Semantic Segmentation Via Multi-task Network Cascades[C].Computer Vision and Pattern Recognition,IEEE,2016:3 150-3 158
[6]Dosovitskiy A,Fischery P,Ilg E,et al.FlowNet:Learning Optical Flow with Convolutional Networks[C].IEEE International Conference on Computer Vision,2015:2 758-2 766
[7]Farenzena M,Bazzani L,Perina A,et al.Person re-Identification by Symmetry-driven Accumulation of local Features[C].Computer Vision and Pattern Recognition.IEEE,2010,23(3):2 360-2 367
[8]Gangnet M,Blake A.Poisson Image Editing[C].Acm Siggraph.ACM,2003:313-318
[9]张建桥,王长元.基于泊松方程的数字图像无缝拼合[J].现代电子技术,2010(17):139-141
[10]Mertens T,Kautz J,Van Reeth F.Exposure Fusion:A Simple and Practical Alternative to High Dynamic Range Photography[J].Computer Graphics Forum,2010,28(1):161-171