In order to modify the luminance heterogeneity of a highly reflective surface image, restrain the phenomenon of local high luminance and to preserve enough original image information,a preprocessing method for highly reflective surface defect image is investigated and the optical characteristics of a highly reflective surface are analyzed by using Partial Differential Equation(PDE) theory and homomorphic filtering algorithms. A homomorphic filtering algorithm based on the PDE is proposed which uses the heat conduction equation filtering operator to process homomorphic decomposition images. The comentropy is induced to evaluate the processing effect in different dimension parameters, then the dimension parameter of heat conduction homomorphic filtering is determined. The algorithm is contrasted with the background subtraction algorithm and wavelet homostasis filtering algorithm,and the experiment indicates that the image processed by the proposed algorithm has the gray-level values distributed uniformly between 1～6 and a high comentropy value above 91%. This method modifies the image luminance heterogeneity, improves image quality and preserves sufficient original image information.