Research on Dual-Channel Image Denoising Algorithm Based on Deformable Convolution
<p>Yuan Chen<sup>1</sup>, Yun Yang<sup>1</sup>, Palizhati·Wusiman<sup>1</sup>, Yanan Wu<sup>2</sup>, Hong Yang<sup>3</sup></p> · Academic Journal of Computing & Information Science · 2024
In response to the issues of excessive smoothing and detail loss in existing image denoising algorithms, a dual-channel image denoising algorithm based on deformable convolution is proposed. Firstly, a noise estimation network is used to obtain a noise level map, enhancing the detail preservation capability of the edge feature extraction block enhancement network. Then, efficient channel attention is combined to focus on key channel features, effectively capturing noise characteristics. Finally, the deformable convolution with deformable learning kernels is integrated into the dual-channel denoising network to extract offset pixels of feature mappings, avoiding excessive image smoothing. Experimental results show that on the Set12 dataset with a noise level of 25, compared with commonly used algorithms, the proposed algorithm improves the average peak signal-to-noise ratio and structural similarity by 0.08dB to 0.62dB and 0.001 to 0.02, respectively, while removing Gaussian noise and preserving image details as much as possible.