Behavior Of Mice Classification And Segmentation Research

Behavior analysis has been applied widely, such as man-machine interfaces,surveillance, image analysis, video coding and so on. In the field of behavior analysis,human behavior analysis has been done and achieved many outstanding achievements.Many aspects of animal behavior analysis will been researched and developed byresearchers. In recently Animal behavior analysis are more and more paid attention. In thisfield, mice are the most interested in animals. The mouse behavior analysis has been widelyused in genetics, psychology, neuroscience, pharmacology and other science fields. Inbehavioral analysis, according to the short limbs of mice, it is difficult to extract relevantinformation. The paper is applied characteristics that extract the mice contour to analysismice eight everyday behavior.Behavior of mice has been studied deeply in the thesis, and correspondingexperimental analysis systems are completed. Firstly, the theory that behavioral analysisprocess involved are elaborated. Definition the behavior of classification and segmentationare outlined, target detection method in pretreatment processes such as: TemporalDifference Method, Background Subtraction Method, Threshold Method, Optical Flow;Classification methods such as: Nearest neighbor and K-nearest Neighbor Method; Simpledistance calculated such as: Euclidean distance, Manhattan distance, Ming Crowes distance,log-covariance distance. Then the behavior of mice is classified. In process of theclassification, First mouse contour is extracted from each frame behavior video. Then thedistance of eight directions are calculated which are from contour center to edge andcalculated two main components of contour, using these values to form ten dimensionalvectors as feature vector, calculate the log-covariance distance. Then the mice videosequence is segmented, In process of the segmentation, histogram are drawn using ratio ofmain components of contour, calculate the histogram square distance, combining log-covariance distance and histogram square distance, mice behavior video sequence issegmented. Finally, the full text is summarized and pointed out innovation in mousebehavior in the text and analyzed the current problems in mice behavior analysis andpossible development direction. In this paper the innovation points: Lack of limbs information from the mice behavior,extract eight directions of mouse contour and two contour main components of contour,classify based on log-covariance matrix distance for mouse behavior. Using ten features ofinclude the part experimental mouse contour and whole, improve the accuracy of theclassification; Combining log-covariance distance and histogram square distance, micebehavior video sequence is segmented. The rate of behavioral segmentation frame errorjudged is reduced.