| Image processing and pattern recognition is interdisciplinary with bright prospect. Now it has been widely applied in artificial intelligence, biomedicine, satellite remote sensing, industrial monitoring and robot control etc, and quickly permeates agricultural scientific research and production technology. In this paper, the two main research contents are just the application of image processing and pattern recognition theory:one is the application of facial expression recognition in artificial intelligence, and the other is digital linear assessment of cow type in agricultural production fields.The expression is the external performance and the main carrier of our emotion, so facial expression analysis and identification is conducive to the realization of artificial intelligence and has the vital significance in artificial emotional development.According to the facial recognition step, the algorithms of facial expression recognition system mainly contain image preprocessing algorithms, feature extraction algorithms and classification algorithms. Firstly, all kinds of facial recognition algorithms are reviewed in this paper, and then some classical algorithms are introduced, the algorithms of feature extraction are mainly discussed at last.This paper strives to find suitable for facial recognition feature extraction algorithm on the base of various existing algorithms. A simple and effective feature extraction algorithm is put forward which is based on Contourlet Transform and FLD. This method firstly uses Contourlet transform to preprocessing image. Contourlet transform is a new kind of Multiscale Geometric Analysis method, it possess wavelet multi-resolution properties and characteristics of localization, also have strong more selectivity and the direction of anisotropic. After Contourlet transform, low frequency of the image reflect the general direction of expression, high-frequency subbands embody the outline of expression, texture. Here, we combine the low frequency and part of high frequency as the whole feature, so the data is compressed and also the essential characteristics of expression are embodied. Then Fisher's linear Discriminant method (FLD) is used to extract features and finally k-neighbor is used to classify them. Experimental results show that this method has faster speed and higher rate of expression than the classical FLD method of feature extraction. Meanwhile, the process is simple and easy to realize.In addition, some research on digital linear assessment of cow type has been done in this paper. Linear assessment of cow type is important in cow breeding. Digital image processing has been widely used in automatic control, pattern recognition, artificial intelligence, etc, but little application is in cow breeding. In digital field, it's just at the beginning. On the basis of the mature theory and algorithms I have mastered and the characteristics of cow type, this paper tries to apply image processing to linear assessment of cow type. Some research has been done on the choice of feature site, choice of shooting position and angle, initial establishment of cow type image library and treatment of holstein. Now preliminary research achievements are obtained.Digital linear assessment of cow type can improve efficiency and accuracy of linear evaluation, increase the dairy scale, standardization and automation, so the development of cow industry which is the most saving in resource utilization is promoted. It becomes a mainstream direction of dairy development. |
