Image Alignment Based on Deep Learning to Extract Deep Feature Information from Images

Abstract

To overcome the limitations of traditional image alignment methods in capturing deep semantic features, a deep feature information image alignment network (DFA-Net) is proposed. This network aims to enhance image alignment performance through multi-level feature learning. DFA-Net is based on the deep residual architecture and introduces spatial pyramid pooling to achieve cross-scalar feature fusion, effectively enhancing the feature's adaptability to scale. A feature enhancement module based on the self-attention mechanism is designed, with key features that exhibit geometric invariance and high discriminative power, achieved through a dynamic weight allocation strategy. This improves the network's robustness to multimodal image deformation. Experiments on two public datasets, MSRS and RoadScene, show that the method performs well in terms of alignment accuracy, with the RMSE metrics being reduced by 0.661 and 0.473, and the SSIM, MI, and NCC improved by 0.155, 0.163, and 0.211; and 0.108, 0.226, and 0.114, respectively, compared with the benchmark model. The visualization results validate the significant improvement in the features' visual quality and confirm the method's advantages in terms of stability and discriminative properties of deep feature extraction.

Keywords: deep learning; feature extraction; image alignment; infrared and visible images.

Figure 1

DFA-Net network structure.

Figure 2

Structure of the deep feature information extraction network.

Figure 3

Structure of the spatial information fusion module.

Figure 4

Structure of the feature enhancement module.

Figure 5

Visualization of the ablation experiment results. Red boxes indicate people. Green boxes indicate objects such as streetlights, vehicles, etc.

Figure 6

Quantitative comparison with SOTA methods in the MSRS dataset.

Figure 7

Quantitative comparison with SOTA methods in the RoadScene dataset.

Figure 8

Image alignment visualization results on the MSRS dataset. Red boxes indicate people, green boxes indicate environmental references such as vehicles, streetlights, etc. (a–c) highlight the spatial alignment effect of pedestrian targets; (d–f) verify the accurate alignment of multi-level targets in composite scenes of people and backgrounds; and (g) shows the alignment effect of vehicle targets.

Figure 9

Image alignment visualization results on the RoadScene dataset. Red boxes indicate people, green boxes indicate vehicles, foliage, buildings and other objects. (a–c) highlight the registration between vehicles and backgrounds; (d–f) demonstrate the consistency of multi-object registration involving people, vehicles, and backgrounds; and (g) shows the registration results for people and backgrounds.