Research on Image Quality Evaluation Method of Depth Learning Model Based on Core Concept

Abstract

In view of the fact that the existing image quality evaluation methods are generally artificial design features, it is difficult to automatically and effectively extract image features that conform to the human visual system. Inspired by human visual characteristics, a new full reference image quality evaluation method based on depth learning model based on core concepts is proposed. Firstly, depth learning algorithm is used to extract multi-layer features from reference images and distorted images respectively. Then, the local similarity of the feature map of the reference image and the distorted image in each layer is calculated as the local quality description of the corresponding depth. Finally, the local quality of all layers is synthesized to obtain the overall quality score of the image. On the basis of the pre-training model, the depth model network is fine-tuned by using the image visual evaluation data set to obtain a depth model for evaluation. The standard experiment shows that fine-tuning training of each pre-training model on the standard data set achieves good classification results. Experiments show that the designed depth learning model based on core concepts is superior to the existing full reference image quality evaluation methods, and its prediction results have good accuracy and consistency with subjective quality evaluation.