Geometric Distortion Correction for the Underwater Images

Non-metric cameras have been widely used in applications of obtaining geometric information of the underwater objects using either digital photogrammetric approaches or computer vision algorithms. All the underwater images exhibit significant geometric distortions caused by lens distortions and light refraction in underwater imaging, which must be geometrically corrected. In this paper, a geometric distortion correction method for the underwater images is proposed, which uses the sets of distortion parameters obtained through the iterative camera calibration to determine the position relationship between the original images and the final corrected images, and then the gray values of the final corrected images are directly resampled from the original images. The Go Pro Hero 5 Black calibration results show that the final accuracies are close to 0 pixel after three iterations; all the final distortion parameters calculated with the iterative calibration method are decreased after several iterations and can be ignored. By contrast, the original image was corrected well with the three sets of distortion parameters calculated with the iterative calibration method. An example shows that the successful generation of point clouds illustrates the effectiveness of the geometric correction. The proposed correcting method provides a technique not only to greatly reduce the distortion through applying a series of distortion parameters but also preserve the image quality through a smart resampling way.