Improvement On Gabor Filter And Its Application On Identification For Cross-section Images Of Cane Plants

The traditional Chinese medical science is treasure of Chinese culture, and itmade a great contribution for the prosperity of China. Moreover, until now it still isthe important part of Chinese medical and health services. And the quality of theChinese medicinal crop has an effect on the curative effect of the traditional Chinesemedical. Therefore doing research on the classification and identification of theChinese herbal medicine has important significance.The traditional identification methods of Chinese herbal medicine mainly dependon the inspection by human beings and subjective experience. However, the methodshave high labor intensive and low automation, and lack of scientific accuratequantitative evaluation. Therefore they have not only low efficiency but also higherror rate. So based on image recognition technology, this paper have done researchon the identification techniques of Chinese herbal medicine in order to make thecomputer classification technology partly replace the traditional artificial methods.But there are many kinds of the Chinese herbal medicine, so the classification of allkinds of the medicine does not make sense. So this paper chooses cane plants, theimportant kind of Chinese herbal medicine, as the object of study. The content of thepaper is as follows:1. Through the research and analysis of the cross-section image features of thecane plants , this paper proposes that take color feature and texture feature as the toolof characterization and classification.2. According to the color characteristics of the images, we take the colorhistogram as the color feature extraction method. And the experiments show that themethod do be effective.3. We have studied the cane plants’texture feature extraction method based onGabor filter. Actually, we propose an improved Gabor filter algorithm and study thePseudo-Zernike moment and its fast algorithm. Based on previous work, we propose anovel texture feature extraction method based on Gabor filter and Pseudo-Zernike.And the experiments show that the method does be effective, which lay thefoundation for the later identification work.4. We study and create an identification system based on support vector machine for the cross-section images of the cane plants. We study the kernel functionsand parameters of the support vector machine, and finally do comparativeexperiments that take different functions and parameters. And the experiments showthat the method does be effective, and the method we take improve the classificationaccuracy greatly of the cross-section images of the cane plant.At last, we have taken large number of experiments and analysis, which all provethat our methods are feasible and effective.