Real-Time Restoration Algorithm for Motion Blurred Images under Deep Learning
Hui Liu, Haipeng Zuo · 2024
Due to high complexity and computational costs, traditional methods are difficult to achieve real-time restoration of motion blurred images. This article introduces a real-time restoration algorithm for motion blurred images based on deep learning, which is suitable for rapidly changing application scenarios. This article adopts an improved U-Net architecture and makes specific adjustments to motion blur, using the VGG network (Visual Geometry Group) Network as the feature extractor. SENet (Squeeze-and-Excitation Networks) can be introduced to enhance the sensitivity of the network to key features. During the network training process, the Adam optimizer can be used to improve training efficiency and algorithm stability. The research results indicate that the method achieves a PSNR (Peak Signal to Noise Ratio) of 43 dB in image restoration with zoom motion blur, which is approximately $43.3 \%$ higher than the GAN (Generative Adversarial Network) model. The SSIM (Structural Similarity Index Measure) under extreme blur 2 reaches 0.65, with a processing time of 0.63 seconds.