A mouse's scratching behavior is one of the reaction effected by a new drug, which is rapid movement of its hind leg around its head. Currently, the number of mouse scratching behavior is counted by manual human's decision, and conventional automatic detections used a high-speed camera. Improving the accuracy of the detection, the automation for counting the scratching behavior is necessary, and the evaluation must be quantitative. We propose the detection of the mouse's scratching behavior based on the shape and the motion feature using a wide-used camera. We apply Improved Dense Trajectories to the classifier to identify the scratching behavior. The classifier could detect the scratching around 90 % accuracy. Based on the length of scratching, we could decrease the ratio of the detection failure to half using a median processing.