Research On Image Aesthetic Quality Evaluation Method Based On Deep Learning

Image aesthetic quality evaluation is to make the computer imitate human's aesthetic process and give the aesthetic quality evaluation of the image to be evaluated,such as giving a simple and rough evaluation result — the level of the image aesthetic quality is good or bad,or give a detailed specific Evaluation result — the probability distribution of image aesthetic quality score or image aesthetic quality score,etc.Image aesthetic quality evaluation is widely used in image search,automatic photography,and photo optimization.The use of image aesthetic quality evaluation methods can further improve service quality and efficiency.In this paper,two methods of image aesthetic quality evaluation are proposed by studying the existing methods of image aesthetic quality evaluation combined with deep learning techniques.The main contents are as follows:(1)Combining the capsule network and the self-attention mechanism,an image aesthetic quality evaluation method based on the capsule network is proposed.The selfattention mechanism is used to solve the problem that the convolutional network cannot extract the long-distance feature dependence in the shallow network.Routing mechanism to update some model parameters.The structure of its network model is explained,and the same type of image test is carried out with the neural image assessment(NIMA)model proposed by Google to achieve the same aesthetic quality evaluation effect.The accuracy rate of the binary classification test on the large dataset of aesthetic visual analysis(AVA dataset)is 72.4%.(2)By studying the development history of image aesthetic quality evaluation technology,it is found that the existing image aesthetic quality evaluation is based on the idea of first extracting the features of the image to be evaluated,and then fitting the extracted features or other operations to output the final evaluation result.To achieve.This article proposes an image aesthetic quality evaluation method based on similarity comparison.The aesthetic quality score of an image is based on similarity comparison,that is,the aesthetic quality score of the image to be evaluated is based on the similarity or difference between the image to be evaluated and the most beautiful image corresponding to the image to be evaluated gives an aesthetic quality score.The applicable scenarios,algorithm principles,and specific implementation steps of the image aesthetic quality evaluation algorithm based on similarity comparison are explained and compared with the existing image aesthetic quality evaluation methods.Experimental verification shows that this method has a fast running speed,the image background effect is small,and the image score difference is obvious and so on.