Abstract: By researching into the shape from shading(SFS) problem and supporting vector machine(SVM), a shape recovery method from light and shade based on SVM is proposed. The theoretical basis and feasibility of this method are analyzed, and key problems such as the determination of SVM input parameters and the choosing and constructing of core function are discussed. Examples show that compared to traditional SFS method, the method can acquire directly the height of an object relatively accurately, is affected by the environment slightly and need very few imaging condition assumption. At last, the direction to improve the method is discussed.

Key words: Depth forecast, Shape from shading, Supporting vector machine Kernel function