Abstract: Aiming at solving the problem of reducing the depth extraction error of smooth foreground in defocus image, this work propose an algorithm to generate the depth map with a single 2D image based on multi-scale wavelet, which can do depth correction by pixel classification techniques and is suitable for both defocus and wide angle images.Firstly, a wavelet analysis method is used to extract depth maps from a single image at multiple scales.Secondly, an adaptive pixel classification method is proposed to do depth correction pixel by pixel according to the variation between scale and depth.Thirdly, the depth map is optimized regionally using region growing integrate with edge segmentation techniques.In order to accelerate the depth calculation, a fast zerocount method and a multi-scale segment method are presented, which can meet the requirements of real-time video processing.Experiments demonstrate that the depth maps generated by our algorithm are not only visually correct but also regionally consistent in both foreground and background.