| In recent years, with the rapid development of Chinese digital painting images, how to establish the digital museum and manage the digital library becomes the urgent task. The technologies to process the Chinese painting image become the key problem. So the low-level feature extraction, data compression, semantic annotation, retrieval and automatic classification attract researchers’ attention. There still exist some difficulties. First, it is difficulty to just use the features of low-level global visual to classify the Chinese painting images because of the "semantic-gap". Second, it is different with the natural scene images. The Chinese paintings’ characteristic is "pursuit the inner spirit of the object", this will limit the low-level visual features’ performance. This paper is mainly to research the Chinese painting images’ classification algorithm. The innovation and principle of this algorithm as follows:1. This paper proposes an algorithm to extract the salient region in the image which is based on the low-rank matrix factorization theory. The principle of the algorithm starting from the low-rank matrix factorization theory, divide the semantic content of the image into salient-region and unsalient-region. The matrix of salient part due to its high content of redundancy, so it corresponds to a low-rank structure, while the salient target has one or more of the features with high difference and therefore can be map to a sparse composition. At last, the salient region is extract from the images, and this will provide an effective representation model for semantic annotation. The comparison results with seven other kind of algorithm also present in this paper. The results of this algorithm applied in the MIT, Bruce and MSRA’s dataset, the algorithm performance in low-entropy image is better. The proposed algorithm is superior to other methods and accordance to the attention of human visual.2. In this paper, a semantic classification algorithm is proposed based on the bag-of-word model. The principle of the algorithm is:firstly use the simple space grid layout to divide the input image to regular area, extract the Scale Invariant features Transform (SIFT) in each of the component, fuse the SIFT descriptor in each channel to Color-SIFT descriptor, and this can describe the region-shape feature in each of the sub-region area in Chinese painting. Secondly, introduce the Bag-of-words (BOW) which is used in natural image field to semantic representation of the Chinese painting images. This paper proposed a simple and effective mechanism of visual attention computational model to analyze the salient region of images based on the bottom-up mechanism. In next section, this algorithm integrated into the supervised learning strategy, then statistical frequency of visual words in each of the semantic category and weighting the visual words semantic, build a visual word frequency histogram. Finally, using support vector machine classifier to form the algorithm which is based on the bag-of-words model. And the validation of this algorithm is applied on the images dataset which we mentioned in the paper. The results of experimental shows that the overall performance of our algorithm’s accuracy reaches74.4%.3In this paper, we also proposed an algorithm of Chinese painting image classification based on the structured information. This is because that Chinese painting has rich structure information. The principle is:firstly, decompose the Chinese painting image into four parts:main body, inscription, liubai and seal, then based on the characteristics of each part extract the color and texture feature, finally integrate the four parts’ feature into multi-tasking joint sparse representation model and use the feature to classify the images. The experiments show that the proposed algorithm can effectively decompose the Chinese painting images, and the classification strategy based on multi-tasking joint performance better than the global-based classification method. |
PDF Full Text Request