Aesthetic Evaluation of Images Based on Deep and Shallow Feature Fusion Network Models

Abstract

In the image aesthetics evaluation, the last layer of the deep network is usually utilized for classification of high-level semantic features, and there is a problem of poor classification performance using this traditional deep learning method. To address this problem, this paper proposes an image aesthetic evaluation based on a deep and shallow feature fusion network model. Improved image aesthetic classification methods allow the extraction of deep semantic aesthetic features and shallow detailed aesthetic features. Experimental validation is carried out based on the AVA dataset, and the results show that the network model has a higher classification accuracy than extracting only high-level features.