基于权重最小二乘结构边缘保持色调映射算法

Edge-preserving tone-mapping algorithm based on weighted least square framework

高动态范围图像由于其动态范围超过普通显示器的动态范围,所以无法正常显示,从而需要研究在保留高动态范围图像对比度、细节信息以及色彩信息的情况下压缩高动态范围图像的动态范围以适应低动态范围显示器进行显示的色调映射算法。本文提出基于权重最小二乘结构的边缘保持图像平滑色调映射算法。首先,建立基于权重最小二乘的边缘保持图像平滑滤波算子;然后,将输入的高动态范围图像转换至NTSC空间分离亮度信息和颜色信息,利用该滤波算子对亮度信息进行多级分层,获得基本层以及多级细节层信息;最后,对基本层进行动态范围压缩,利用压缩后的基本层结合多级细节层信息并转换回RGB空间获得输出的低动态范围图像。文中通过实验采集的多曝光图像序列利用Debevec和Malik提出高动态范围图像融合算法获得拍摄场景的高动态范围图像,采用本文提出算法对高动态范围图像进行色调映射处理获得较为理想的保留图像有效信息的低动态范围图像,从而验证了文中提出算法的有效性。

Because the dynamic range of high dynamic range image exceeds the dynamic range of the display device,the display device cannot display images normally. Tone-mapping algorithms can retain the detail information and color information while compress the dynamic range of high dynamic image,which makes sure that the output low dynamic range image can display normally in the common displayer. The edge-preserving image smoothing tone-mapping algorithm based on weighted least square framework was proposed. Firstly,edge-preserving image smoothing filter operator based on weighted least square was established; secondly,the input high dynamic range image was changed from RGB space to NTSC space to separate the brightness information and color information. The brightness information of high dynamic range image was decomposed by filter operator,and then the base layer and alternative multi-level detail layers were obtained; finally,the base layer was compressed,and the output low dynamic range image was acquired by combining the compressed base layer and multi-scale detail layers and the low dynamic range image was changed to RGB space. For aquiring the exposure image sequences,high dynamic range image was gotten by high dynamic range image fusion algorithm proposed by Debevec and Malik. Using the proposed algorithm,the tone-mapping processing is carried out for high dynamic range image,and then the ideal low dynamic range image with effective information are obtained,which verifies the effectiveness of this algorithm.