Improved interpolation with sub-pixel relocation method for strong barrel distortion

Abstract

In this paper, we propose an improved interpolation method with sub-pixel relocation method to correct fisheye images with strong barrel distortion. The proposed method takes as inputs the fisheye image and the projection model, and outputs the corrected perspective image. In the method, we project the fisheye image pixels to perspective image plane using the backward projection. The projected pixels are treated as scatter data, and their location is refined by the sub-pixel relocation method. The method is designed to aim at the sub-pixel bias phenomenon of the projected pixels. The sub-pixel bias phenomenon is that there is a sub-pixel bias between the projected pixel center point and the center point of the projected pixel. This phenomenon is caused by the uneven distortion of the fisheye image plane, and it will affect the accuracy of the interpolation method. The sub-pixel relocation method relocates the projected pixel centers to the centers of the projected pixels. Then the relocated points are meshed by the Voronoi diagram method. The mesh is used to select the neighbors of the interpolation points as natural neighbor selection method. The weights of the neighbors are calculated using the inverse distance method. The indexes of the neighbors and their weights are stored in a look-up-table structure for reuse. For each fisheye image, the proposed method can quickly find the neighbors of the interpolation points and the weight of each neighbor through the look-up-table, and the perspective image can be calculated in real-time. Test results illustrate that the proposed method achieves the best accuracy among the tested methods for fisheye images with strong barrel distortion.