With the study object of tooth shaped chain board, a precise measurement technology for small parts was proposed using image processing technologies combined with least square support vector for regression (LSSVR). Electric coupling apparatus（CCD）was adopted as the sensor to collect images, single pixel edge information of the undeterminate part was obtained after image preprocessing such as image denoising and image binaryzation, etc. Afterwards, on the basis of structure feature analysis of the part, candidate regions were searched automatically where the coordinates of the points were taken out as training point set to train LSSVR. Then the candidate geometry parameters, such as the included angle between every two lines, diameters of the round holes as well as errors of the two round holes were gained through experiment results. The last two geometry parameters with millimeter unit were obtained through unit transmitting from pixel to millimeter by camera calibration, and the precision can attain the micron level. The experiment showed that the proposed solution has a high convergence speed as along with the high precision and excellent stability.