Image Aesthetic Assessment Based on Attention Mechanisms and Holistic Nested Edge Detection

Aesthetic assessment of images has attracted a great deal of research interests as one of the hot spots in the field of computer vision. This paper proposed a new method for classifying image quality from an aesthetic perspective, by classifying images into two categories: high aesthetic quality and low aesthetic quality. Since the local features and edge features of images affect detail and holistic perception of aesthetic, we propose an image aesthetic assessment model based on attention mechanism and holistic nested edge detection. The model consists of two parallel deep convolutional neural networks named upper and lower branches. The upper branch uses an attention mechanism for extracting local features of RGB images and the lower branch is used to extract edge features of images. Experimental results indicate that the proposed method can predict image aesthetic with the accuracies of 88.13% and 94.10% in AVA dataset and CUHKPQ dataset respectively. Compared with the other methods, our approach takes into account both the details and the holistic information of images and performs more efficiently in assessing visual aesthetic of images.