The Bayer GWR algorithm is proposed to obtain the high-quality color image for embedded machine vision system. The method changes the original GWR input parameters according to the structure of Bayer CFA. Using tricolor’s each mean value to replace their sum value in the GW assumes of GWR, and the each mean value of the near white point to replace their maximum in the Retinex assumes of GWR. Test shows: the method can effectively reduce the color temperature with a low data computation and process the 640?480 original image data in the smart camera with 600MHz within 34 ms. It can satisfy the practical application of practical and real-time requirements.