Abstract: To study the 3D reconstruction method of shape from shading，a new method is proposed combining the wavelet transformation with the traditional minimization algorithm.While an image is wavelet transformed，the frequency and the directional characteristics are different in different sub-frequency bands，so different reconstruction algorithms are chosen in different sub-frequency bands.The natural stripe phase is extracted directly in low frequency bands，and the traditional minimization algorithm is used in the high frequency bands.The ideal conditions supposed in the traditional shape from shading methods are no more necessary in the low frequency region，and the detail part restructuring quality of the minimization algorithm in the high frequency region is maintained in the proposed method.This method is applied to reconstruct several emulation images.The experimental results show that the error of this novel method is less than that of traditional minimization algorithm.

Key words: shape from shading, wavelet transformation, frequency characteristic, stripe phase

摘要： 提出了一种结合小波变换的从明暗恢复三维形状的最小化方法，该算法利用图像小波变换各个子频段的不同频率特性和方向特性，分别采用不同的算法重构。在图像被分解后的低频区域采用小波直接提取自然条纹相位，高频区域使用最小化方法重构。这种方法在低频区域避免了SFS最小化方法的假设条件，而高频区域又发挥了它对细节部分重构的优势。实验结果表明该算法比单独使用最小化方法的误差要小。

关键词: 从明暗恢复形状, 小波变换, 频率特性, 条纹相位