Deep Learning Based Image Aesthetic Quality Assessment

Image Aesthetic Quality Assessment aims to assess a given image from the view of aesthetic perception.The study of algorithm in Image Aesthetic Quality Assessment has provided much support for image visual experience based applications,such as image retrieval,image beautification,and social media and so on and has been a hot topic in computer vision.The study of algorithm in Image Aesthetic Quality Assessment mainly focuses two parts: the feature engineering and assessment model engineering.The study of feature engineering aims to design some better aesthetic feature improving the ability of aesthetic representation.The study of model engineering aims to design better assessment model improving the assessment ability.This paper introduces some works we have conducted in the study of the assessment manner,the study of convergence rate of training for assessment model.We list the works as follows:First,we have studied the image aesthetic quality assessment via label distribution learning.Generally,one single label and numerical aesthetic quality value are adopted for assessment.However,due to the complexity and subjectivity of this problem,one single indicator is not sufficient and objective.In order to solve the problem of the conservative,we proposed one novel assessment manner:the assessment manner of label distribution.The experiments conducted on standard dataset indicated that the proposed assessment manner has the characteristics of objective,strong fault tolerant ability.Secondly,we have studied the image aesthetic quality assessment using transfer learning.In order to make up the problem of small scale of image visual evaluation data set and to speed up the convergence,we fine tune some pre-trained deep learning model on aesthetic assessment specifically data to obtain the deep learning model for image aesthetic quality assessment.Experiments have proven that the proposed method is effective.The above works have covered some score issues in the study of image aesthetic quality assessment.Our works improved the ability of aesthetic representation in aesthetic features,the accuracy of assessment model and the efficiency of algorithm.The works in this paper have made some contribution to the study of image aesthetic quality assessment.