Evaluating grade of tone mapped high dynamic range images

摘要

To display high dynamic range (HDR) pictures by traditional displays, Tone mapping operators (TMOs) intent to compact HDR pictures to low dynamic range (LDR) pictures. Conventional image similarity assessment like MSE, structural similarity (SSIM) and PSNR cannot be directly applied to analyze the alteration between pair of images with distinct dynamic ranges. This paper calculates the grade of tone mapped images by using structural fidelity along with statistical naturalness. Determining the image grade has essential significance for many image processing tools and applications. The algorithm should automatically assess the grade of images and it should perform closely with the quality of human perception. Subjective evaluation is the straight and genuine way to assess tone mapped pictures however it is costly and highly time absorbing, making them difficult to be set into automatic optimization algorithms.