Research On Image Aesthetic Quality Evaluation Method Based On Semantic Perception And Layout Perceptio

In recent years,image aesthetic quality assessment has attracted considerable attention due to the massive growth of digital images in social platforms and the Internet.The image aesthetic quality assessment aims to measure "image beauty",and its research goal is to design an algorithm to simulate human aesthetic perception of images and then make an objective evaluation of the aesthetic quality of images,which is consistent with human subjective evaluation of image aesthetic.Image aesthetic quality assessment is a key research direction in the field of computer vision.It has a wide range of application scenarios in the industrial field,and has very important research significance and practical value.This paper mainly studies the general aesthetic quality assessment of natural images,aiming to achieve accurate aesthetic quality prediction by deep learning methods.Two methods are proposed respectively from the perspective of image semantic information and layout information:(1)Automatically assessing aesthetic quality of an image is a challenging task,because the image aesthetic is affected by many various factors,and the criteria for judging the aesthetics of images with different semantic information are different.To this end,an image aesthetic quality assessment model based on Semantic-Aware and attentional Multiscale feature fusion,SAM-CNN,is proposed.The model can fuse intermediate features of different layers in the CNN to obtain a more comprehensive and accurate aesthetic expression,under the joint supervision of image aesthetic quality assessment task and semantic classification task in a multi-task learning manner.Besides,through an attention mechanism,it uses the abstract semantic information with a large receptive field contained in the deep features to guide the model to pay attention to the key parts of the features to be fused,so as to improve the effect of feature fusion and the performance of the model.Experimental results on the AVA dataset and Photo.net dataset demonstrate the effectiveness and superiority of the proposed model.(2)The layout information of images is significant for assessing image aesthetic quality.Relevant psychological researches have confirmed that there is a strong correlation between the layout of images and the perceived image quality.However,the characteristic of conventional convolution operation is locality,and due to the inherent limitations of receptive field,it lacks effective perception of the overall layout.To this end,an aesthetic quality assessment model based on graph convolution Layout-Aware and Recurrent Connection feedback,LARC-CNN model,is proposed.The model utilizes graph convolution method to reason the dependency between local areas,so as to perceive the overall layout of images.In the process of modeling the overall layout,aspect ratio is utilized to restore the original spatial composition information of images to achieve more accurate layout learning.At the same time,a recurrent connection feedback structure is constructed to fundamentally improves the layout perception process of graph convolution.Experimental results on the AVA dataset and AADB dataset demonstrate the effectiveness and superiority of the proposed model.