An Efficient Image Quality Assessment Guidance Method for Unmanned Aerial Vehicle

Abstract

More and more advanced unmanned aerial vehicles (UAVs) equipped with different kinds of sensors can acquire images of various scenes from tasks. Some of them have to assess the obtained images first and then decide the subsequent actions like humans. Accurate and fast image quality assessing capability is critical to UAV. One or more objective quality indexes are usually selected by UAV to assess all the whole image, which may lead to inefficient evaluation performance. In order to further link human cognition pattern with intelligent vision system and provide useful guidance to shorten the image quality assessment time for UAV, a new experimental method of subjective image assessment based on local image is proposed in this paper. 60 participants are invited to conduct subjective image quality assessment experiment, in which 15 original images including people, scenery and animals are distorted by four methods, i.e., Gaussian additive white noise, Gaussian blur, jpeg compression and jp2k compression. Moreover, a new local image segmentation method is designed to segment each image into 6 local areas. For the subjective scores, global-local correlation is analyzed by Spearman Rank Order Correlation Coefficient (SROCC). The experimental results show that the global subjective assessment has the strongest correlation with the local subjective assessment having the best image quality. Further analysis shows that the local images with the best quality often have sufficient color information and rich texture details. Assessing the local images instead of the global ones provides a shortcut to design objective evaluation algorithms, which is a practical guidance for UAV to perform efficient images quality assessment.