The work aims to solve the color correction problem of multi-light scene images. First of all, mesh generation and key-point sampling were used to divide the multi-light image into a number of regions. Then, the light source of the divided region was estimated with the color constancy algorithm of a single light source. The contribution of each region made to the scene light source color was integrated as the composite light source color to be the approximate estimate of the multi-light scene. Finally, the correction was applied with the diagonal model. Such method was compared with the estimation independently made based on the single light source color constancy algorithm. The effect of multiple light sources on the image could be weakened by dividing the multi-light image into several regions. Compared with the estimation independently made based on the single light source color constancy algorithm, the correction effect of the image was significant. Through the locally estimated multi-light source image, the proposed method can effectively solve the problem of color constancy for multi-light images of outdoor scenes.