Comparison of different registration algorithms for infrared polarization image

The interpretation of infrared polarization image is contingent upon the accuracy of image registration. Given the inherent limitations of infrared polarization imaging, include low resolution and blurred texture details, the image registration must be robust and precise. This paper presents the results of feature point extraction and image matching experiments conducted with infrared intensity images acquired by infrared polarization imaging system with three different polarized directions: 0°, 120°, and 240°. The performances of the scale invariant feature transform(SIFT), speedup robust features (SURF) and oriented fast and rotated brief (ORB) algorithms are compared with respect to computational speed and accuracy. The results demonstrate that, in the case of large-target infrared polarization scenes, the SIFT algorithm is able to extract the greatest number of feature points and has the highest matching accuracy in comparison with the SURF and ORB algorithm. Furthermore, the ORB algorithm is the fastest in terms of computational speed, with a speed that is more than 1.5 times that of the other two algorithms. In the case of small-target infrared polarization scenes, the registration accuracy and speed of the three algorithms are comparable. The SIFT algorithm is therefore more suitable for rapid registration processing of infrared polarization of small-target scenes. It is necessary to select the most appropriate registration algorithm in accordance with the specific requirements.