A grayscale weight with window algorithm for infrared and visible image registration

The registration of Infrared (IR) and visible images is an important prerequisite for image processing tasks such as image fusion, target detection and tracking, and remote sensing. The registration task of the IR and visible images usually involves two problems: (1) extracting consistent features from multi-sensor images is difficult and (2) similarity measurement methods such as normalized mutual information (NMI) algorithms are prone to falling into local extremities. To solve these complications, this study proposes a grayscale weight with window algorithm (GWW) to extract common strong edge features from IR and visible images, reduce the joint entropy values and local extreme values of NMI, and improve the performance of NMI to calculate IR and visible images for registration. Qualitative and quantitative experiments demonstrate that the GWW can effectively extract the common features of IR and visible image pairs, improve the performance of the surface peak, increase the ratio of primary and secondary peaks, and effectively reduce the local extremum. The performance of NMI combined with the GWW algorithm is better than the traditional MI, NMI, and ECC, and has better matching accuracy and higher matching probability. The registration of IR and visible images can be fully realized by NMI combined with the GWW algorithm.