Emotion-Assisted Image Aesthetics Recognition Algorithm

Aesthetics is a kind of advanced feeling after the interaction of a series of complex feelings such as vision,feeling,emotion and imagination.With the rapid development of information media,a large number of pictures are uploaded to various social media platforms every day,in which the aesthetic quality of images is uneven.In all kinds of poster advertising pictures,the massive design is also dazzling.When people look at pictures,they are more willing and interested to see pictures with high aesthetic quality.Therefore,how to design an efficient image aesthetics recognition algorithm is a research hotspot in the field of computer vision.Image aesthetics quality recognition from the early design image aesthetic manual features to the current automatic learning features based on deep learning.These studies ignored that the aesthetics of the image is a subjective task,and it is closely related to the feelings of human beings,so it is not comprehensive to only focus on the aesthetics of the image.In the field of neurological research,some researchers have pointed out that when making aesthetics judgments and triggering emotions,the areas of the brain that produce stress responses are the same.And intuitively,there is also a certain connection between image emotion and aesthetics.Therefore,in this way of thinking,the emotion-assisted image aesthetics recognition algorithm is designed in this paper.The main work of this paper is as follows:1.There is no dataset that contains both image aesthetic score and image emotion distribution in the existing dataset,so in order to complete the work of this paper,we label a unique dataset Flickr?EA,which contains both image 8 classification emotion distribution and image aesthetic score.This dataset can be used to complete the work of emotion-assisted image aesthetics recognition.And it can help this paper to explore the relationship between image emotion and image aesthetics.2.Three multi-task learning networks with progressive recognition performance are designed in this paper.They are emotion-assisted aesthetic recognition algorithm based on global image,emotion-assisted image aesthetic recognition algorithm based on image classification and image aesthetic recognition algorithm based on emotion aesthetics hierarchical feature fusion.These three methods have obvious advantages compared with the existing advanced image aesthetics recognition algorithms and convolution neural networks with excellent feature extraction ability.3.This paper explores the specific influence of emotion on image aesthetics recognition from two aspects.Firstly,in the Flickr?EA dataset,using three correlation coefficients,it is found that there is a certain relationship between the image aesthetic score in the dataset and different emotion distribution,which provides a certain theoretical feasibility for the experimental idea of this paper.In the second point of work,this paper also obtains a certain relationship between emotion and aesthetics in deep learning by inter-task correlation learning,which can help the image aesthetics recognition network to improve its performance.This verifies the practical feasibility of our experimental idea.