Fisheye Image Distortion Correction Based on Deep Learning Regression Model

Compared to pinhole cameras, fisheye lenses offer a much wider field of view but also introduce significant image distortion, which poses challenges for both human observation and computer image processing. Therefore, it is necessary to process the fisheye image with large distortion. In this paper, we propose a VGG16-based deep neural network to remove single fisheye image distortion. Compared with the traditional method of template extraction and neural network classification, we use the regression model to calculate the parameters of distorted images, which can obtain better accuracy and produce less error on a large number of images. The experimental results show that the proposed method performs better in the image distortion correction task, particularly in correcting fisheye image distortions with varying degrees of distortion. The dataset used in our experiments was synthesized using CitySpace[1]