In order to solve the problem of mutual restriction between high spatial resolution and large field of view of industrial cameras prominently protruded in the visual measurement of finepitch gear, this paper studies on the subpixel image registration method for finepitch gear that has strong foregroundbackground contrast, insufficient local contrast and large area of the same color under rimming light condition. Aiming at the problem of less feature points and low correct matching rate in the image registration process based on feature points in finepitch gear image registration, the registration process suitable for finepitch gear image is described in this paper, the global histogram equalization is introduced to improve image contrast, enrich grey level and enhance the feature information in dark part of the image, so that the number of feature points and correct matching rate are improved. In order to overcome the deficiency of incorrect threshold setting, parameter adjustment difficulty and model nonuniqueness of traditional matching pair purification algorithm, combining with 2σ criterion, the global selfadaptive parameter matching point pair purification algorithm is put forward, which not only ensures the uniqueness of the purification result, but also adjust the judgment threshold adaptively to avoid artificial error and ensure the repeatability of result. Experiment results show that the number of feature points and correct matching rate are greatly increased when the image is preprocessing with the method described in this paper. Besides, the registration result measured by the average of translation amount purified with the method described in this paper outperforms that of the traditional method in correctness and precision. Aiming at the finepitch gear image under rimming light condition, the registration precision is better than 0.083 pixel, which has great practical application value for the visual measurement of finepitch gear.