The accuracy of image detection and classification will be reduced by image nonlinear distortion, hence distortion correction is an important preprocessing step in image processing. This paper proposes a non-metric distortion correction method based on adaptive corner detection to better solve the image distortion problem. First, the distorted checkerboard image is used as the research object, and the sub-pixel corner coordinates obtained by adaptive corner detection are used as input information. The distortion loss function is established based on the invariance of straight-line projection and the weight coefficient based on the radius of curvature is set to adjust the distortion of each curve. Then, the nonlinear optimization method is used to calibrate the distortion coefficient and the distortion center to realize the distortion correction. Two sets of experiments are designed: standard image simulation experiment and real image experiment. At the noise level σ=1 in standard image simulation experiment, the relative error of distortion parameter k₁ is 0.733%, and the relative error of the distortion coefficient is 4.933%, which is close to the true value. In the real image experiment, the average straightness of the method in this paper is only 0.5152 pixel, which shows that the correction accuracy is high, and the average operation time is about 8.4 s, which shows that the correction efficiency is better. Experimental results show that the method is feasible, accurate and efficient.