Experimental Mice Body Features Image Recognition Algorithms

The observation, recording and statistical information of rat's spontaneous activity are very important for biomedical research. The traditional study of animal's spontaneous activity is mainly depended on the visual observation and qualitative evaluation, but there is also a lot of inconvenience.In recent years, with the development of video and computer image processing technology, automatic identification of rat's spontaneous activity has been greatly improved. However rat's body posture is diverse and complex, the method also has many deficiencies, especially, posture recognition effect is not satisfactory.This paper, focuses on the image processing algorithm on rat's body feature recognition, respectively designs the top shooting and side shooting video monitoring devices to collect rat's activity digital image;Usually, there are uneven illuminations in rat's activity place. In this case, the regular gray scale and two values methods are not perfect.This paper uses a series of open and closing operations to eliminate this effect.The results show that, the improved algorithm can effectively eliminate the adverse effects of rat's experimental image.This paper uses three kinds of filtering methods to process image, and compares the processing effect; it finds that the median filtering is not suitable for the picture that contains boundary nodes. In addition, linear smoothing filtering appears many vague details, wiener filtering is the best, and more suitable for rat's activity image recognition.Paper also uses three edge detection methods that based on the gradient to extract the rat's edge, it shows that there are serious shortage. But using object boundary tracking method for edge extraction in rat, rat's tail is fine and continuous. Thus, compared with the previous three methods, this method is better for rat's image recognition.When the image acquisition method is top shooting, we can use rat's centre to track related motor position information, research and design corner detection method to extract the rat's head, tail root feature information, and their location. Based on the relative position parameters of head, tail root, the center of gravity, we can make certain posture recognition.Since the top shooting method to obtain the information is limited, it is not conducive to identify the rat's complex posture, this paper combined the side and top shooting to recognize body feature, especially this method is good to distinguish the squat and erected postures. In side shooting algorithm, we identify the body mainly by the rat's body direction, the basic idea is to remove the tail of rat, then rat is covered by the minimum bounding rectangle, and rat's special posture is determined by the minimum distance of the vertex of the rectangle in vertical direction.Combining with the two shooting methods, it can collect the rat's activities within larger areas, not only recognize some simple activities, but also detect rat in some complex posture, and enrich the rat's body posture.The results of this study can provide help for improving the spontaneous activities of rat's computer image recognition system, improve the recognition rate and effect, it also provides effective information for biomedical research.