Abstract: The key issues of machine vision system is the speed, accuracy and stability of image processing. To make the machine vision system of automatic screw-classifying devices to monitor the geometric parameters, the image edge and its detection stability are enhanced by the Curvelet transform and the speed of image processing is accelerated by the program optimized by MMX, SSE, and IPP functions. The application shows that the image processing speed of the system can reach 25 ms per frame with repeat accuracy of 8 μm. So the system can meet the requirements of stability and real-time monitoring geometric parameters.