Pavement cracks as a common highway damage, the complex road condition and the underdeveloped image acquisition software and hardware technology make the achieved pavement images with high noise, it brings difficult image processing and low crack detection. Therefore, it is significant to study a crack detection system processing high noisy images. This thesis focuses on image processing of vehicle-mounted pavement crack detection system, aiming to deal with high noisy images. Doing research in terms of specific system image acquisition, pre-processing, grayscale and color images morphological operations, etc., and to complete the image of binary and determine the optimal threshold to achieve the MATLAB image processing. Experiments show that this method can be a reality.