Application And Research Of The Classification In Art Style Based On Cross Contrast Neural Network

Art style is an important part of art works,which will help us to have a better grasp of the meaning of art works.Oil painting is a main stream in the history of Western art.With the development of digital technology,many museums are building digital oil painting library,which has created a new digital sharing era.For art works,its artistic style is a relatively complex description,so how to correctly identify the artistic style of painting is the key to building a large art database.Many researchers in computers science have devoted considerable efforts to classifying the styles of art works and using scientific classification as a tool to achieve artistic quantification.Difficulties in the study of art style classification are as follows.The first is the definition of art style is subjective and there is no specific quantitative index for artistic style distinction,which makes it difficult to screen features in the classification task.The second is that the number of art works is limited which is not suitable for classifying based on a large sample of deep learning models.In view of the above problems,this paper proposes a Cross Contrast Neural Network(CCNN)model,which performs well in oil painting artist classification tasks.The main points of this paper are divided into the following sectionsFirstly,research backgrounds and significance of art style classification are introduced.Then we import the research status of style classification respectively.The research methods and progress of music style classification,calligraphy style classification and oil painting style classification are introduced respectively.Secondly,a multi-classification method of CCNN is proposed,which can measure the similarity and achieve the multi-classification task of the art style.This is the main part of this paper.We used the Wiki Art Library to build a data set of oil paintings,including more than 2,000 works painted by 20 artists from 13 styles.CCNN achieved good performance on the oil painting dataset(85.75%in the artist classification),far exceeding the traditional deep learning network such as Resnet.Finally,we further explores the migration characteristics and application scenarios of CCNN.Based on the experimental results of various data sets(CMU-PIE,IDRiD),it is found that CCNN is more suitable for multiple classification tasks for small dataset images with obvious style features.Furthermore,this paper also visualizes the filter of CCNN to understand model better.Our study provides a new approach to art quantification.