The feature extraction is crucial for fault diagnosis,although the existing feature extraction methods in time domain,frequency domain and time-frequency domain are effective,it is also necessary to find new methods in other domain.On the basis of the characteristic of snow images generated by analyzing SDP（Symmetrized Dot Pattern）method,a feature extraction method based on image processing are put forward.Firstly,the original vibration signals are transformed into snow images in polar coordinate by SDP method.Then,the shape features of the snow images in different fault pattern of rolling bearing are extracted by using image processing technology,and the characteristic parameters are analyzed.Finally,BP network are created to realize fault pattern recognition.The experimental results show that this method is effective.