In order to solve the problem that for defocus binary coding technique, the higher harmonic components in the fringe pattern cause measurement errors, a method of generating a binary image by threshold optimization dithering algorithm is proposed. By introducing the parameter γ to dynamically adjust the threshold, and considering the influence of different pixel gray values of the image on the dithering algorithm, the sinusoidality of the dithered image stripe after defocusing is improved. Simulation and experiments show that using the threshold optimization dithering algorithm, the fringe after defocusing has better sinusoidality when the parameter γ is 50. Compared with the traditional Floyd-Steinberg dithering algorithm, the root mean square error can be reduced by 0.013 7 rad when the fringe period is 40 pixel. The imaging is more clear for the sag of the model, and the image quality of the surface reduction of the object is significantly improved.