Abstract:

To elucidate how to optimize combination features and to design a classifier with high classification accuracy, a challenging problem, a method based on error rate of classification as standard to select combined feature was presented so as to raise the classification accuracy. First, four kinds of f
ractal dimensions are extracted as texture features. Then, various combination features are training samples of SVM. With combination feature with the lowest classification error rate as a vector to be applied to the classification, the classification accuracy of the classifier can be improved. A variety of patterns are generated by primitive gene and regenerative gene. The proposed method is simple and easy in operation that can be widely popularized. So it can lay the foundation for the combination of image features.

Key words: fractal dimension, Xinjiang folk art patterns, support vector machine(SVM), image classification