In the three-dimensional scanning technique based on image parallax,in order to calculate the shape of objects,it is required to match two points at different positions in images polluted by noises.An improved method for image matching was proposed to locate the object point and its corresponding image point in a mirror.By calculating the difference between the gray values of the pixel and its neighborhood,with a variable weight,by the algorithm a multidimensional discriminant vector was constructed and the angle between two discriminant vectors determines whether a pixel in object space matches another pixel in mirror space.The experiment results show that,compared with the traditional SAD,NSAD,SSD and NSSD,the algorithm is better to deal with the image matching problem involving the attenuation,fuzzy and chromatic aberration,and it is capable of improving the geometric shape precision of the reconstructed object.