No Reference metrics are based only in the analyzed image. They are capable of assessing the degree of distortion, without any knowledge of the original image, or the streaming/ codec parameters.
To create a set of distortion assessment metrics, we had to familiarize ourselves with the properties of the artefacts. To achieve this, we studied different publications and we analysed the properties of distorted images in MATLAB, like the relationship between the blocks, the spectrum, edge detection, histograms… Knowing this properties allowed us to evaluate every frame we received and to evaluate it in the way we need to detect each distortion.
The next step we took was to verify the behavior of the implemented metrics. In that way, we analysed the different results that the metrics returned when they were tried with images that suffered different distortions – the problem we found here was that the presence of an artefact in the image influences metrics of the other artefacts.
In this stage of the design, the metrics are subjected to test. These tests consist of the comparison between the results given by the implemented metrics and the results coming from subjective quality reviews from some different human testers (Subjective Video Quality).
The last stage is the implementation of metrics as a computer software (language C++). The open source OpenCV library is used here to optimize operations. The problems with which we meet here are the differences between MATLAB functions and libraries and C++ tools.