Abstract: As the raw data of daytime star gazing, there are single-point-noise, stripe noise and some other noise in infrared star map, which severely affect the sequential star-point extraction and centroid localization. According to the star map of field star gazing, methods of star map denoising and star-point position accuracy analysis were studied in the paper. An algorithm for single-point-noise detecting and compensating and a colum mean offset correction (CMOC) algorithm for stripe denoising were proposed. Then, a complex filtering method of Wiener 2 and Top-hat was used to denoise the star map furtherly. At last, a method of adding ideal simulated star-point to observation star map was used to calculate the starpoint centroid position accuracy. Experimental results indicate that the preprocessing method could denoise various kinds of noise in infrared star map, and the peak signal-to-noise ratio of star point target could be raised from 4.7 to 31.3, and the standard error of star-point localization of simulated 2 Mv star could achieve 1/30 pixel, which satisfies the accuracy of engineering application.