An Image Deblurring Method Using Improved U-Net Model

Abstract

Deblurring methods in dynamic scenes are a challenging problem. Recently, significant progress has been made for image deblurring methods based on deep learning. However, these methods usually stack ordinary convolutional layers or increase convolution kernel size, resulting in limited receptive fields, an unsatisfying deblurring effect, and a heavy computational burden. Therefore, we propose an improved U-Net (U-shaped Convolutional Neural Network) model to restore the blurred images. We first design the model structure, which mainly includes depth-wise separable convolution, residual depth-wise separable convolution, wavelet transform, inverse wavelet transform, and a DMRFC (dense multireceptive field channel) module. Next, a depth-wise separable convolution is designed, which reduces model calculations and the number of parameters when compared with the standard convolution. A residual depth-wise separable convolution is designed, which allows for propagation of detailed information from different layers when compared with standard convolution and a standard residual block. The wavelet transform realizes downsampling by separating the contextual and texture information of the image. It also reduces model training difficulty. The inverse wavelet transform realizes upsampling, which reduces the loss of image information. Finally, by combining an extensional receptive field and channel attention mechanism, a DMRFC module is proposed to extract detailed image information, which further improves the reconstructed image quality via inverse wavelet transform. Experiments on the public dataset GOPRO show that the image deblurring method in this paper has higher-quality visual effects, while the PSNR (peak signal-to-noise ratio) rises to 30.83 and SSIM (structural similarity) rises to 0.948. Fewer model parameters and a shorter recovery time are needed, which provides a more lightweight image deblurring method.

CNB Net: A Two-Stage Approach for Effective Image Deblurring

“…Their approach, utilizing a residual module, a cascade cross-attention module, and a two-scale discriminator module, enhances detail processing. Lian et al [3] employ a U-Net-based [4] method incorporating attention mechanisms and depth-wise separable convolutions, focusing mainly on local details. Cui et al [5] propose a novel dual-domain attention mechanism, combining spatial and frequency attention modules, thus addressing both local and frequency-dependent aspects of images.…”

“…Traditional deep CNNs often struggle with capturing global information due to their limited receptive fields, as highlighted by Chen et al [13]. To address this, some researchers, like Lian et al [3], recommend using convolution with a larger receptive field for better global information comprehension, thereby enhancing deblurring effectiveness. Additionally, attention mechanisms, as proposed by Cui et al [5], have been integrated to more precisely focus on critical image areas for detailed information capture.…”

“…Zhang et al's [2] method for single-stage image motion deblurring excels in extracting local feature information, yet it somewhat lacks in addressing global contextual relationships. Lian et al's [3] U-Netbased [4] image deblurring method, enhanced with an attention mechanism, focuses more on local details. Similarly, Cui et al [5] introduce a dual-domain attention and self-attention model for image deblurring, which primarily learns from local regions while reducing computational demands.…”

Onboard Processing of Drone Imagery for Military Vehicles Classification Using Enhanced YOLOv5

“…Lian et al [7] explained wavelet transformations, inverse wavelet transforms, residual depth-wise separable convolution and a DMRFC (dense multi 7 receptive field channel) module. A convolution which is depth-wise separable is created.…”

WIG-Net: Wavelet-Based Defocus Deblurring with IFA and GCN

“…Chen et al [10] embedded smooth dilated convolution to the network, while keeping the number of parameters in receptive fields constant to improve network performance. Zuozheng Lian et al [11] introduced an enhanced U-Net, incorporating depth-wise separable convolutions, residual depth-wise separable convolutions, and wavelet transform. This approach enables the extraction of finer image details while simultaneously reducing computational complexity.…”