Details

Autor(en) / Beteiligte

• Wu, Heng
• Chen, Bingxin
• Guo, Zijie
• He, Chunhua
• Luo, Shaojuan

Titel

Mini-infrared thermal imaging system image denoising with multi-head feature fusion and detail enhancement network

Ist Teil von

• Optics and laser technology, 2024-12, Vol.179, p.111311, Article 111311

Ort / Verlag

Elsevier Ltd

Erscheinungsjahr

2024

Link zum Volltext

Quelle

Alma/SFX Local Collection

Beschreibungen/Notizen

• •An infrared image denoising method is proposed to remove noises from the infrared image captured by the mini-infrared thermal imaging system.•A multi-head feature fusion block composed of multi-scale and feature-fusion blocks can enhance the width and depth of the network and extract and fuse multi-scale features.•A detail enhancement network based on residual learning and multi-level local feature attention blocks (MLLFA) that can enhance the expression of the detail feature information. The mini-infrared thermal imaging system (MITIS) has been widely applied in outdoor photography, military observation, medical auxiliary diagnosis, and so on. However, due to the limitations of infrared sensor performance, imaging system size, and environment interference, the infrared images captured by the current MITIS generally encounter noise interference and low contrast issues. To solve these issues, we propose a multi-head feature fusion and detail enhancement network, namely MdNet, for complex features (texture and edges) retaining and denoising of infrared images obtained from the MITIS. Specifically, MdNet consists of a multi-head feature fusion block (MFFB), detail enhancement block (DEB), and reconstruction convolution. The MFFB is designed to extract and fuse multi-scale feature information, including multi-scale and feature fusion blocks. The DEB is developed to enhance the noise feature and reduce the loss of detailed information by three multi-level local feature attention blocks and residual learning. Quantitative and qualitative experimental results demonstrate that the MdNet achieves competitive performance in removing synthetic and natural noises compared with other state-of-the-art infrared image denoising methods and provides an effective solution for MITIS image denoising.