A visual inspection system for small modulus gears was established to realize their industry measurement. A de-noise algorithm based on scale-correlation and direction-correlation in Curvelet transform was proposed according to the de-noising demands in system calibration. First, the theory and process of the Curvelet transform were introduced. Then, according to the gear symmetry itself, a new combined algorithm for image de-noising based on Curvelet scale-correlation and direction-correlation was proposed, the corresponding parameters were calculated, and the processing results of the algorithm were compared under different noises. Finally, the algorithm was compared with 3 de-noise methods (Gaussian filter,Wavelet De-noise and Threshold De-noise). Experimental results indicate that the Peak Signal to Noise radio(PSNR) from the combined algorithm proposed is 6 db, 3 db and 8 db better than those of Gaussian filter, Wavelet De-noise and Threshold De-noise, respectively. It maintains good image edges and satisfies the image quality requirements of camera self-calibration in visual inspection systems for small modulus gears.