Automatic quantitative assessment of vascular damage in mice's dorsal skin-fold window chamber (DSWC) plays an important role in personalized and precise vascular targeted photodynamic therapy (V-PDT). The narrow-band light blood vessel image of DSWC, however, will produce nonrigid deformations in varying degrees, which is caused by breathing, heartbeat, and involuntary movement of mice during V-PDT, and this leads to quantification errors of vascular damage. A judgment hybrid registration method combining the advantages of feature-based and gray-scale registration methods was proposed. The slight and severe deformations were first distinguished using the mean square difference (MSD) of registration groups. For slight deformation, the differential homeomorphism Demons (Log-Demons) algorithm was directly adopted for image registration. For severe deformation, the speeded up robust features thin-plate splines (SURF-TPS) algorithm was used for global registration, and then the Log-Demons algorithm was used to realize local registration. The experimental results show that the proposed algorithm has better registration accuracy and higher processing efficiency than the SURF-TPS, Log-Demons, and SURF-TPS+Log-Demons algorithm. Furthermore, this algorithm could effectively correct the inaccuracy assessment of vascular injury caused by severe deformation.