Contour extraction process is often disturbed by complex backgrounds or foregrounds in the machine vision system. A new contour extraction algorithm is proposed based on wavelet transform and dynamic contour compensation. The wavelet transform method is used to extract the edge of image, and the dynamic compensation is used to restore the contour which is affected by the complex dark areas such as bar code or black icon. The algorithm is applied in a sausage machine vision inspection system, and it is proved effective in eliminating the contour depression and retaining the actual defect features.