Combined No-Reference Image Quality Metric for UAV Applications

Abstract

An expansion of the use of UAV images requires the improvement of methods and means for image analysis and processing in order to effectively solve various problems. Visual quality metrics play a key role in this sense, since their use allows determining the need in different image processing operations, their type and parameters, automating the entire process. The paper considers the problem of using no-reference visual quality metrics and test image databases to solve such problems. The effectiveness of more than 40 existing visual quality metrics for images with typical distortions for UAVs has been evaluated. The paper proposes a combined metric based on an artificial neural network to improve the accuracy of visual quality assessment and the possibility of its application in practice with sufficient efficiency.