Proposed in this paper is a new dynamic gesture recognition framework to solve time-scale variability and complexity of dynamic gestures. This framework constitutes a dynamic gesture contour image by using the gesture contour in time series ， and calculates the mean and variance images of gray-scale images in different time scales. Furthermore ， these mean and variance images are organized into a dynamic gesture contour model library.On this basis ， dynamic gestures are recognized by using correlated information method and improved dynamic time warping method. Experimental results show that the proposed dynamic gesture contour model is of strong robustness to the dynamic gestures in various time scales ， and that the improved dynamic time warping method helps obtain recognition rate higher than that of the traditional method.