Phase unwrapping for a noisy image suffers from many singular points. Singularity-spreading methods are useful for the noisy image to regularize the singularity. However, the methods have a drawback of distorting phase distribution in a regular area that contains no singular points. When the singular points are confined in some local areas, the regular region is not distorted. This paper proposes a new phase unwrapping algorithm that uses a localized compensator obtained by clustering and by solving Poisson's equation for the localized areas. The numerical results demonstrate that the proposed method can improve the accuracy compared with other singularity-spreading methods.