Neural learning-based image quality metric without reference

In this paper, we propose a new framework to optimize the utilization of the image quality estimation without reference. This framework is based on two principal steps. Features are first extracted from the image to characterize each considered degradation type. From this modeling step, a No Reference Image Quality Metric (NR-IQM) per degradation type is obtained. In the second stage, outputs of the previous model are combined to achieve a unique index. The modeling and combination step are here realized using an Artificial Neural Networks (ANN). Our method is compared to some recent methods. The obtained results show the relevance of the proposed framework.