图像美学质量评价综述

Overview of Image Aesthetic Assessment

近年来,由于高美学质量图像的需求越来越多,图像美学质量评价成为热门研究课题之一。深度学习的发展极大推动了图像美学质量评价的发展。图像美学质量评价常用的数据集有Aesthetic Visual Analysis(AVA)数据集、Aesthetics and Attributes DataBase(AADB)数据集和CUHK-Photo Quality(CUHK-PQ)数据集。图像美学质量评价的研究方法主要分为两个类型,基于手工设计美学特征的方法与基于深度学习的图像美学质量评价方法。基于手工设计美学特征的方法是根据摄影规则或图像内容人工设计美学特征,再利用机器学习方法对图像进行高低质量分类。基于深度学习的图像美学质量评价方法是用卷积神经网络自动提取图像的美学特征,并对图像进行高低质量分类。图像美学质量评价用准确率(Acc)、均方误差(Mse)、平均绝对误差(Mae)、中位数绝对误差(Med)来度量模型的有效性。

In recent years,due to the increasing demand of high aesthetic quality images,the image aesthetic assessment has become one of the hot research topics.The development of deep learning has greatly promoted the development of image aesthetic assessment theory.The commonly used databases for image aesthetic assessment include Aesthetic Visual Analysis(AVA)database,Aesthetic and Attributes DataBase(AADB)and CUHK-Photo Quality(CUHK-PQ)database.The research method of image aesthetic assessment is mainly divided into two stages,the method based on the aesthetic features of manual design and the method based on deep learning.The method based on manual design aesthetic features is to manually design aesthetic features according to photography rules or image content,and then machine learning is used to classify images of high and low quality.The method of image aesthetic assessment based on deep learning uses convolutional neural network to automatically extract the aesthetic features of the image,and classify the high and low quality of the image.Image aesthetic assessment uses accuracy(Acc),mean square error(Mse),mean absolute error(Mae),median absolute error(Med)to measure the effectiveness of the model.