Image aesthetics quality evaluation method based on multi-domain knowledge driving

Abstract

The invention relates to an image aesthetics quality evaluation method based on multi-domain knowledge driving, and the method comprises the steps: S1, designing a dense connection network as a backbone network, and extracting the aesthetics features of images; S2, designing a semi-supervised learning algorithm, learning style features from labeled and unlabeled images and extracting the style features of the images; S3, training a scene semantic classification model and a sentiment classification model by using the scene semantic classification data set and the sentiment classification data set, and extracting semantic features and sentiment features of the images; and S4, performing feature screening and fusion on the extracted features by using a gradient boosting algorithm XGBoost, andtraining an SVM classification model and an SVR regression model to predict the aesthetic quality of the image. According to the invention, the aesthetic quality prediction precision can be significantly improved.

Description

Technical field [0001] The present invention relates to the field of image processing and computer vision, in particular to a method for image quality evaluation aesthetics based multi-domain knowledge driven. Background technique [0002] Since the visual aesthetics is a subjective attribute, often it involves emotional and personal taste, influenced by the rules of photography, image content and image style, which makes automatic evaluation of the aesthetic image quality is a very subjective task. After the introduction of deep learning model in the evaluation of aesthetic quality, can learn to be more effective aesthetic characteristics, in order to further improve the performance, scholars have begun using the expertise of a variety of aesthetic and closely related fields to assist in the evaluation of beauty, Mai such as the introduction of semantic classification information as a network branch assistant aesthetic evaluation, summed up the seven kinds of impact on the image...

AI Technical Summary

Problems solved by technology

For the fusion of multi-domain features, the traditional method is to use dot product, merging, custom fully connected layer, etc. However, since the obtained multi-domain features do not contribute equally to the aesthetic feeling, the features are further processed according to the importance of the features. processing needs to be considered

Images