Through the analysis of confocal microscope in the imaging process caused by the position, such as optical hardware deviation converge and pinhole position deviation occurring in image distortion phenomenon, a position correction function into interpolation algorithm was proposed for nonlinear distortion image correction and rehabilitation. The convolution neural network based on machine learning technology was applied to improve the quality of image position correction after degradation when training a single image. The five layers of convolution and down sampling to join pooling layer were employed to reduce the order of magnitude in network parameters. The results show that the pooling layer can improve the operation speed significantly and improve the sharpness of the image.