Web Assembled Benchmark for Image Visual Quality Assesment, Prediction and Improvement
Rostyslav Tsekhmystro, Viacheslav Oliinyk, Галина Проскура, Oleksii S. Rubel · 2020 IEEE 15th International Conference on Advanced Trends in Radioelectronics, Telecommunications and Computer Engineering (TCSET) · 2020
Visual quality of an image is important factor at all stages of its processing. Images taken by modern devices, such as, digital cameras or smartphones can be distorted by the noise. Currently various effective denoising methods are available to suppress noise of different types. A purpose of applying these techniques is to improve image visual quality in terms of human perception and effective automatic post-processing. This paper considers the following stages of image processing under noisy conditions: initial and intermediate image quality assessment (estimation) during treatment using metrics; prediction of such estimates without a priori information and reference data; and reasonable denoising. As the result, a benchmark is developed to deploy all aforementioned processing stages to a web platform and provide their functionality regardless of user device type and operating system. The proposed web-application allows a user to employ images represented in different format; set the noise model and its characteristics, get image quality assessment parameters; tune denoising methods and obtain a rationale of its applying. In addition, a rationale for filtering running predicted values of visual quality metrics and their improvement after denoising without its applying are provided using neural networks. The benchmark is implemented using WebAssembly for all calculations and evaluations and TensorFlow.js to run neural networks.