Research And Application Of Five-Color Theme Harmony Prediction Algorithm

In the field of art,color matching is widely used in a variety of artistic design,such as image,poster,clothing and interior design.Among them,harmonious color matching is often a decisive factor for the popularity of a design.Although artists have never stopped the research of color harmony,the current color harmony theory only provides a limited number of harmonious color combinations,and the public without knowledge of color harmony cannot quickly select harmonious color combinations and apply them to appropriate scenes.In recent years,with the continuous improvement of computer performance and the rapid development of the field of computer vision,researchers began to use computer vision methods to deal with outstanding problems in other disciplines.On the one hand,people set up large-scale websites for users to share harmonious color matching.Users can refer to published color matching schemes,or upload harmonious color matching conforming to their own aesthetics.On the other hand,people collect data from these websites to build datasets.On this basis,they use machine learning methods to build models to predict the degree of color harmony,quantifying the fuzzy concept of color harmony into an intuitive standard.However,the existing methods have two main problems.First,the color features from the dataset do not reflect users’ preference for different colors;Second,the influence of different users’ aesthetic preferences on color harmony should be considered in the model.In view of the above problems,this paper begins to study the harmony algorithm of the five-color themes.Our main contributions are summarized as follows:In this paper,we propose a color harmony evaluation method based on color-pair to accurately evaluate color theme harmony.First,we use the method of two-layer maximum likelihood estimation(MLE)to predict the color harmony by statistical modeling the paired color preference of the existing dataset.Then,the preliminary prediction results were refined by back-propagation neural network(BPNN),and various color features were extracted from different color spaces to obtain accurate harmony estimation.In order to consider the influence of different users’ aesthetic preferences on color harmony,this paper proposes a color harmony estimation model based on crowdsourcing.By establishing a probability model between the hidden layer and the output layer of BPNN,the distribution parameters of different users’ aesthetic cognition were obtained,and the harmony estimation was obtained by applying the model to the training of BPNN.Through this color harmony estimation model,we try to reveal the relationship between different users’ aesthetic cognition and color harmony.In this paper,we hope to promote the development of related research fields through the research on the harmony degree algorithm of five-color color themes,such as the rapid assessment of the color harmony degree of an image,and the one-click color change in scenes such as clothing,home,3D model,etc.,according to the different needs of users.