Abstract
Next-generation display technologies provide significantly improved dynamic range over conventional display devices. In the long run, advanced CCD or CMOS sensor technologies and data formats will provide high dynamic range (HDR) content for these display devices. Despite the increasing availability of HDR content, legacy low dynamic range (LDR) images and videos represent the majority of content in the near term. Recently more and more researches work on reproducing real-world appearance images through LDR images using inverse tone mapping methods. Therefore, evaluation metrics to qualify the performance of inverse tone mapping operators (iTMO) are needed to understand the effects of important features such as nonlinearity. Most evaluation requires a reference ground truth to compare with the generated HDR images. However, a reference HDR image may not be available or hard to verify as a ground truth. Dynamic range independent quality assessment metric is proposed to measure visual distortion based on the detection and classification of visible change on an image pair. However, complex contrast detection predictor and careful calibration are needed. In this work, we propose a new quality assessment metric to detect the visual distortion based on the probability of various contrast change using JND model on the iTMO curves directly. We also apply our quality assessment scheme on image structures by comparing HDR images generated by various iTMO with the original LDR images. Various contrast changes include loss of original visible contrast, enhancement of original invisible contrast, and inverse of original contrast. Results of our metric using the iTMO and image structure matches with that of the dynamic range independent assessment method. Our metric does not require complex calibration and the computation is simple. This method also help to understand the effect of various parameters on the iTMO curves and analyze which method is suitable for a given image. In addition, we also propose a blind image quality assessment that measures test images without information from reference images. The blind image estimation method contains attributes of contrast, brightness, and colorfulness. The analysis results of contrast and brightness matches with that of the JND-based contrast analysis and the characteristics of iTMO curves. Compared with previous approaches, our evaluation metric provides valuable quality assessment results consistent with human perception and the computation complexity is much lower.