An Automatic Image Stitching Method for Infrared Image Series
Huang Zhi-jian, Bingwei Hui, Sun Shujin · 2021 International Conference on Control, Automation and Information Sciences (ICCAIS) · 2021
Infrared images have a limited field of view in general, and it is difficult to monitor the entire scene with only a single frame. To generate a wide field of view, image stitching is the necessary tactic. However, in real infrared images stitching tasks, high overlap ratio between the adjacent frames in the image series often leads to high computation cost and accumulative errors. Therefore, manually selecting the subset frames is sometimes inevitable. In order to tackle the drawbacks, an automatic method for infrared images series is proposed. Firstly, dominant frames subset is selected based on cosine similarity between images, which could be used as the substitute for all the image series. Secondly, feature extraction and image registration is performed on the image subset. And finally, several image blending methods are utilized to obtain the seamless panoramic result. Experiments carried out on real infrared dataset demonstrated that the proposed method could effectively select the dominant frames of the images series and significantly reduce the time cost to automatically generate the panoramic result.