Cross dynamic range cross resolution Image Quality Assessment with Applications
In recent years, image and video signals have become an indispensable part of human life. There has been an increasing demand for high quality image and video products and services. Different types of distortions and contaminations can affect on perceived quality of images and videos. Therefore, objective image and video quality assessment tools play crucial roles in a wide range of applications throughout the field of image and video processing, including image and video acquisition, communication, interpolation, retrieval, and denoising. A number of objective image and video quality measures have been introduced in last decades such as mean square error (MSE), peak signal to noise ration (PSNR), and structural similarity index (SSIM). However, they cannot be applied to newly emerged challenges of visual signal quality assessment. For example, they are not applicable when the dynamic range or spatial resolution of images being compared is different from that of the corresponding reference images. In this work, we aim to tackle these two main problems in the field of image quality assessment.
Tone mapping operators (TMOs) that convert high dynamic range (HDR) to low dynamic range (LDR) images provide practically useful tools for the visualization of HDR images on standard LDR displays. Most tone-mapping operators have been designed in the absence of a well-established and subject-validated image quality assessment (IQA) model, without which fair comparisons and further improvement are difficult. We propose an objective quality assessment algorithm for tone-mapped images using HDR images as references by combining 1) a multi-scale signal fidelity measure based on a modified structural similarity (SSIM) index; and 2) a naturalness measure based on intensity statistics of natural images. To evaluate the proposed Tone-Mapped image Quality Index (TMQI), its performance in several applications and optimization problems is provided. Specifically, the main component of TMQI known as structural fidelity is modified and adopted to enhance the visualization of HDR medical images on standard displays. Moreover, a substantially different approach to design TMOs is presented, where instead of using any pre-defined systematic computational structure (such as image transformation or contrast/edge enhancement) for tone-mapping, we navigate in the space of all LDR images, searching for the image that maximizes structural fidelity or TMQI.
There has been an increasing number of image interpolation and image super-resolution (SR) algorithms proposed recently to create images with higher spatial resolution from low-resolution (LR) images. However, the evaluation of such SR and interpolation algorithms is cumbersome. Most existing image quality measures are not applicable because LR and resultant high resolution (HR) images have different spatial resolutions. We make one of the first attempts to develop objective quality assessment methods to compare LR and HR images. Our method adopts a framework based on natural scene statistics (NSS) where image quality degradation is gauged by the deviation of its statistical features from NSS models trained upon high quality natural images. In particular, we extract frequency energy falloff, dominant orientation and spatial continuity statistics from natural images and build statistical models to describe such statistics. These models are then used to measure statistical naturalness of interpolated images. We carried out subjective test to validate our approach, which also demonstrates promising results. The performance of the proposed measure is further evaluated when it applied to parameter tuning in image interpolation algorithms.