DR IQA Database V2

In practical media distribution systems, visual content usually undergoes multiple stages of quality degradation along the delivery chain, but the pristine source content is rarely available at most quality monitoring points along the chain to serve as a reference for quality assessment. As a result, full-reference (FR) and reduced-reference (RR) image quality assessment (IQA) methods are generally infeasible. Although no-reference (NR) methods are readily applicable, their performance is often not reliable. On the other hand, intermediate references of degraded quality are often available, e.g., at the input of video transcoders, but how to make the best use of them in proper ways has not been deeply investigated.

This database is associated with a research project whose main goal is to make one of the first attempts to establish a new IQA paradigm named degraded-reference IQA (DR IQA). We initiate work on DR IQA by restricting ourselves to a two-stage distortion pipeline. Most IQA research projects rely on the availability of appropriate quality-annotated datasets. However, we find that only a few small-scale subject-rated datasets of multiply distorted images exist at the moment. These datasets contain a few hundreds of images and include the LIVE Multiply Distorted (LIVE MD), Multiply Distorted IVL (MD IVL), and LIVE Wild Compressed (LIVE WCmp) databases. Such small-scale data is not only insufficient to develop robust machine learning based IQA models, it is also not enough to perform multiple distortions behavior analysis, i.e., to study how multiple distortions behave in conjunction with each other when impacting visual content simultaneously. Surprisingly, such detailed analysis is lacking even for the case of two simultaneous distortions.

We address the above-mentioned and other issues in our research project titled Degraded Reference Image Quality Assessment. As part of this project, we address the scarcity of data by constructing two large-scale datasets called DR IQA database Version 1 (V1) and DR IQA database Version 2 (V2). Each of these datasets contains 34 pristine reference (PR) images, 1,122 singly distorted degraded reference (DR) images, and 31,790 multiply distorted final distorted (FD) images, making them the largest datasets constructed in this particular area of IQA to-date. These datasets formed the basis of multiple distortion behavior analysis and DR IQA model development conducted in the above-mentioned project. We hope that the IQA research community will find them useful. Here we are releasing DR IQA database V2, while DR IQA database V1 has been separately released, also on IEEE DataPort. If you use this database in your research then please cite the following paper (Details about the DR IQA project can also be found in this paper):

S. Athar and Z. Wang, "Degraded Reference Image Quality Assessment," Accepted for publication in IEEE Transactions on Image Processing, 2022.