Article Preview
Top1. Introduction
The value of an artistic painting depends on the associated style and the artist. For instance, the works of famous artists often have higher artistic and inheritance value. Due to the natural disasters, man-made disasters, etc., the destruction and disappearance of artistic painting accelerates. The digitalization of cultural heritage (Donghui 2019) is conducive to protect different kinds of cultural heritage and artistic painting works. During digitalization, the digital image recognition and classification (Grilli 2019) plays an important role in digital protection of cultural heritage, which is the foundation to storage, recovery, promote, and spread the cultural heritage and artistic painting. Thus, it has become an urgent problem to effectively organize and artistic style works and enable users to easily access these works.
In recent years, the artistic style classification (Chu 2018) has become a hot topic in the field of computer graphics. The convolutional neural networks (CNN) and scale-invariant feature transform (SIFT) are two methods which are used in artistic style classification. The artistic style recognition methods based on CNN or SIFT suffers the issue that the model is too complex and time-consuming. The accuracy and efficiency needs to be further improved. Additionally, the existing classification methods may be invalid when there are many types of artistic painting styles. This paper focuses on Chinese painting recognition (Zi 2019). The Chinese painting contains ink painting, pyrography, mural, and splash ink painting. Some samples are shown in Figure 1.
Figure 1. Some samples of Chinese paintings. (a) Ink painting; (b) Pyrography; (c) Mural; (d) Splash ink painting
The image information entropy represents the aggregation features of the grayscale and color distribution. It can reflect the local structure and color features of the image. These features can be used to determine the image quality (Ye 2018). In this paper, we propose a painting style recognition algorithm by using information entropy. According to the gray information entropy, the color distribution entropy, block entropy and contour entropy are proposed as the features to represent the painting and used to train an oracle for classifying the styles of Chinese painting. The color distribution entropy (Shen 2016) represents the color distribution feature of the image. The block entropy (Kumar 2016) reflects the local spatial distribution feature. The contour entropy relates with the edge and structure information of the image. The intelligent classification of artistic painting styles is helpful for users to analyze the style of artistic paintings and beneficial to protect the inheritance of artistic works, which has a wide range of applications, such as digital library, television and movie works, advertising, digital entertainment.
Compared with previous painting style recognition works, this paper also considers the unknown style in test paintings. The unknown style maybe contains few artistic and inheritance value. The framework to recognize the style of artistic paintings is summarized in Figure 2.
Figure 2. The architecture to recognize the style of artistic paintings
In Figure 2, we first collect the ink paintings, pyrography paintings, murals, and splash ink painting to construct training set and test set. The training set is used to learn an oracle which can automatically predict the future painting’s style. The test set is the one which is regarded as the future paintings to verify the effectiveness of the learnt oracle.
Second, the painting images are preprocessed by denoising, cropping, and normalization. By setting a mask operator, the noises in painting image are eliminated by Gaussian filter. When the background area accounts for most of the painting area, the painting area is cropped to maintain the main area. The painting images in training set and test set are normalized by resizing as .