Research On The Machine Learning Method Of Image Aesthetics Evaluation Of Natural Scenes

The rapid development of computer vision and artificial intelligence,the derivation of digital devices,and the gradual improvement of people's living standards have spawned a large number of digital images,but there are differences in the sense of beauty between a large number of images.Through the effective evaluation of the image aesthetics,it can help people to select suitable and effective images from a large number of images.At the same time,it has a positive effect on the editing industry and home users to manage images.Based on the machine learning method,this paper starts with the two classification methods of image aesthetics and score regression,and proposes a corresponding model solution,which enables the computer to imitate human aesthetics and effectively evaluate different types of images.For the two-category evaluation method of image aesthetics,in order to solve the problem of redundant features,poor validity,limited generalization ability,and complex algorithm implementation,this paper proposes a multi-model feature-based feature optimization.A classifier evaluation method.This method extracts three types of multi-type aesthetic features,and then combines the order of feature importance of random forests with the complexity of feature dimensions and the effect of individual features classification,so that significant features are selected and generalization ability is obtained.Strong,stable,and low-dimensional best features.Based on the optimal characteristics,two classifiers,SVM and Random Forest,were used to realize the two-class evaluation of image aesthetics.The experimental results show that this method achieves higher classification accuracy for different types of natural scene images and images of the same type with different aesthetics.The model method in this paper has a good classification effect.For the regression evaluation method of image arts,in order to solve the manual visual features can not fully capture the different visual aesthetics of the image,ignore the effective combination of different types of images and features,as well as learning and optimization of the performance of the returner.This paper presents an evaluation method based on multi-angle features and a combination of optimal features and regression.This method extracts four aesthetic features of different perspectives,and then combines the high-level aesthetic features of different types of images with the optimal features.Secondly,SVR and three ensemble learning regressors are used to select the best regressor,and the best regression is achieved.Stepper parameters are optimized so as to realize the regression evaluation of different natural scene images.Experiments show that the four different types of features extracted by this method have good complementarity,and the prediction results of different types of natural images are in line with human subjective aesthetic evaluation,and have good aesthetic prediction performance.