Rice paper feature analysis based on texture parameter statistics of multispectral imaging

Abstract

As one of important carriers of the traditional Chinese painting, rice paper has attracted wide attention. Current studies of rice paper, which have described some of rice paper's features, however, are confined to empirical macroscopic features and mechanical properties analysis. Any of the methods mentioned above cannot characterize the structural features of rice paper accurately and quantitatively, and cannot distinguish between different kinds of rice paper either. To solve these problems, we propose a novel approach for rice paper feature analysis based on texture parameter statistics of multispectral images in this paper. The multispectral imaging system is applied to obtain rice paper's spectral images under different band channels. And then texture parameter statistics are used to form a feature vector which is able to digitalize rice paper's feature. To evaluate the accuracy of the feature vectors, they are entered into the support vector machine(SVM) classifier for rice paper classification. Results show that under 550nm spectral band which is just the center of visible spectrum, rice paper's differentiation feature is pronounced most, and under that band the average accuracy is 86%. It means that application of multispectral imaging and texture analysis can describe the rice paper's feature with high accuracy.