Application of the Non-Reference Metrics for the Quality Estimation of the Current Images of the Stationary Ground Multi-object Scenes. Part 2

Abstract

Quality estimation of the received images is one of the most important operations of image processing in technical vision. This operation allows us to take decisions about a possibility of the image processing and choose the procedures for this processing. A preliminary image quality estimate makes it possible to reduce significantly the computing resources for solving of the problems of objects detection, recognition and selection. This article focuses on estimation of sensitivity of the non-reference metrics (represented in Part1) to various factors: noises, smoothing and different object compositions. Furthermore, this article focuses on the correlation relationships between the metrics. The main task of this article is simplification of the procedure of a real time image quality estimation and improvement of its effectiveness by removing the related metrics and selecting the most adequate and efficient metrics. Sets of images were created for solving of this task. Those sets contained images with different kinds of scenes and object compositions. Images were received at different times of a day and under different weather conditions. Besides the received images were distorted by different kinds of noises and smoothing. Sensitivity estimation was measured by the method of analysis of variance and Kruskal- Wallis test. The correlation analysis of the relationships between the metrics was done by means of Pearson correlation coefficient and Spearman's rank correlation coefficient. The analysis of the sensitivity estimation results for the presented set of metrics revealed that it was appropriate to use a limited number of metrics. Further results of the correlation analysis revealed that there was a strong linear relationship for a number of metrics. Thus, the obtained results made it possible to select the most sensitive metric with the least computational costs for a preliminary estimation of the quality image in the vision systems.