The Study Of Feature Extraction And Classification Algorithm In Traditional Chinese Paintings

Image semantic classification is an important and challenging task in the field of semantic-based image retrieval. The image contains abundant semantics, it is necessary to classify these images according to semantics which expressed. The traditional technology of image classification mostly retrieves images by analyzing the similarity of image visual features, however, it neglects the impact and function of image semantic, which can not meet users' real needs very well, so classifying images rationally using image semantic will greatly improve the performance of semantic-based image retrieval systems.With the improvement of living standards, people pay attention to material civilization at the same time pay more attention to spiritual civilization, the appreciation of Traditional Chinese painting(TCP) will be more and more demand, the previous mark for keywords search on the way can no longer meet the people's needs, people pay more attention to identify a certain type of calligraphy and painting works, such as search flower-and-bird or landscapes paintings. So, semantic-based TCP image classifications have great practical value. In this article, the classification is based on the five different categories(flower-and-bird paintings/landscapes/figure paintings/pommel horse paintings/bamboo paintings) which TCP have.In this paper, TCP images are data source, some important technologies and algorithms of TCP images semantic classification system are studied deeply. "TCP images semantic classification system" is designed and developed, which implements the TCP images classifying based on semantic initially, we also make some meaningful discussions on how to choose TCP by its semantic. In this paper, image semantic model is introduced firstly, which is the abstraction of image's entire semantic representation and process procedure, and the theoretical basis for the topic is also provided. Secondly, the extraction methods of visual image features such as color, texture and shape etc are expounded. When features of TCP images are being extracted, integrating color and shape of TCP can express semantic commendably, so, in this paper, a method that fusing color and shape features is proposed, studied the color and shape of the feature extraction algorithm, fusioned the color and the shape of target, constructed a new feature vector, we establish the relationship between multi low-level features and high-level semantic of TCP images, Support Vector Machine is used in the Semantic Classification, the experiment results show that this method extracted the stability feature vector, it can reach the higher classification accuracy. And by a large number of social surveys and documentation refer, established the mapping relationship between low-level feature and high semantic of TCP, that is the mapping relationship between the fusion feature about color and shape and TCP semantic. That is, the relationship between TCP's color-shape synergetic features and its semantic. That lays the foundation for the achievement of TCP images semantic classification.Support vector machine (SVM) technology has a unique advantage in solving nonlinear, small sample issues, so this paper choose the Support Vector Machine as the classification algorithm, the experiment achieved the desired results. SVM classification function like a form of support vector machines, output is a linear combination of intermediate nodes, each intermediate node corresponds a plot of input samples and a support vector. This paper implements the issues of TCP five-classifying using multiple classifiers.Through researching of image feature extraction and classification algorithm, a system of TCP images classification based on semantic is exploited, according to TCP images' color and shape feature values; we input these values into the inputting layer of support vector machine classifier, train and test classification of the support vector machine, which achieves a better result. The system's practicality is also validated.The whole paper's works proved that it is feasible to classify based on image's semantic, and it can greatly improve the accuracy of image classification using multi-feature integration and improved classification algorithm. The issue has important theoretical and practical significance.