Noise Robust edge detection based on wavelet transform and convolutional neural networks
Jayeon Yang, Fuping Wang, Jingjing Ren · 2023
The image edges is the most fundamental features and carries most of the information in the image. It is also one of the most important research topics in image processing. However, real images are generally contaminated by noise. Therefore, it is important to consider how to eliminate the interference caused by noise while ensuring the precision of the detected edges. This paper proposes a noise robust edge detection algorithm based on wavelet transform and convolutional neural network. Firstly, the noise in images is suppressed with wavelet transform and threshold denoising in the wavelet domain. Secondly, rich local features are further extracted suing residual dense modules to eliminate redundant noise-induced background clutter. Lastly, the edge detection module integrates multi-scale features in the hierarchical network and iteratively improves the output edge map through deep supervision. The experimental results on the BSDS500 dataset demonstrate the effectiveness of proposed method. Compared with the typical methods, The poposed method achieves higher quality edges under different noise levels and has strong generalization ability.