Aperture-mask-render-based defocus blur generation method for calibration plate image

The accuracy of camera calibration is the key to the accuracy of visual positioning and measurement. However, when calibrating a shallow depth of field camera, the calibration image is prone to local defocus, which affects the calibration accuracy. Recently, deep learning network shows great prospects in deblurring, but the network training requires accurate training data composed of pairs of full focus and defocus images, which is difficult to collect in practical application. According to the regular arrangement of calibration patterns, we propose the aperture-mask-render-based defocus blur generation method for calibration plate image, which combines the correlation between physical lens aperture and defocus blur to generate high-resolution and realistic local defocus effect of calibration plate. Through the proposed method, we can adjust the blur area and focus to generate a large group of realistic defocus and full focus image pairs of calibration plate, providing datasets for network training. Experiments show that compared with the previous methods, this method is closer to the real image data and can improve the camera calibration.