改进拉普拉斯金字塔模型的高动态图像色调映射方法

Improved Laplacian Model for Tone Mapping of HDR Image

高动态图像的色调映射是图像处理的研究热点,用于解决常规色调范围显示设备无法有效显示高动态图像的问题.针对当前色调映射不能精准地显示出图像内容(如丢失细节、模糊弱边缘)的问题,提出一种改进拉普拉斯金字塔模型的高动态图像色调映射方法.首先基于滤波和下采样处理计算出对输入图像进行了简单去噪的高斯金字塔;然后通过约束强边缘数量分离出高斯金字塔中每层图像的强?弱边缘和细节;再针对每层图像构造其分层映射函数,通过增强细节、弱边缘的对比度以及保持强边缘对每层图像进行逐点映射,并据此计算描述每层对比度变化的拉普拉斯金字塔;最后根据拉普拉斯金字塔自底向上重构出目标图像.与已有的多种色调映射方法相比,根据客观评价和主观评价的结果表明,文中方法能增强细节和弱边缘、避免强边缘处梯度反转,更清晰、准确地显示出高动态图像内容.

The tone mapping of HDR （high dynamic range） images is one of the hot topics in image processing, which is used to handle the problem that regular equipments cannot efficiently display HDR images. However, current tone mapping methods cannot accurately display image content such as losing details and blurring weak edges. This paper proposes a new tone mapping method by improving the Laplacian pyramid model. Firstly, the Gaussian pyramid of the input image is computed by using down-sample and filter to reduce noise. each Gaussian pyramid layer are separated via Then, the strong edges, weak edges, and details of restricting the pixel numbers of strong edges. Then different mapping function is defined for each Gaussian pyramid layer, and is used to implement pixel- wise mapping on each layer to enhance the comparison of details and weak edges and preserving strong edges. Based on the mapped Gaussian pyramid, the corresponding Laplacian pyramid is computed to record the comparison changes of each layer. Finally, the result image is constructed from bottom to top based on Laplacian pyramid. Compared with existing tone mapping methods, based on the subjective and objective assessments, our method can display HDR images more clearly and accurately by enhancing details, weak edges and avoiding gradient reverse.