基于卷积神经网络的LDR图像重建HDR图像的方法研究

目的为了在高动态范围成像技术中更简单、有效地拓展图像的动态范围,提出一种基于低动态(LDR)图像重建高动态(HDR)图像的算法。方法基于扩张卷积层的卷积深度神经网络模型,提出一种根据相同场景中各种照明与曝光的LDR图像组来建立HDR图像新模型的图像融合算法。结果通过所提出的LDR和不同比特深度HDR映射关系,采用链式结构完成了从LDR图像到HDR图像的重建。结论通过拟建的HDRI模型,拓宽了图像的动态范围,并提高了物理光信息恢复能力。与传统算法相比,该研究所提出的方法能减少运算量,能较好地还原高动态范围场景。

Reconstruction Method of HDR Image Based on Convolutional Neural Network for LDR Image

The work aims to propose an algorithm for reconstructing high dynamic range (HDR) images based on low dynamic range (LDR) images, in order to expand the dynamic range of images more easily and effectively in high dynamic range imaging technology (HDRI). Based on the convolutional deep neural network model of expanded convolutional layer, an image fusion algorithm based on various illumination and exposed LDR image groups in the same scene was proposed to establish a new model of HDR image. Through the proposed mapping relationship between LDR and HDR with different bit depths, the chain structure was used to complete the reconstruction from LDR image to HDR image. Through the proposed HDRI model, the dynamic range of the image is broadened and the physical light information recovery capability is improved. Compared with the traditional algorithm, the method proposed in this study can reduce the amount of computation and better restore the scene with high dynamic range.