.png){kind=icon}
.png){kind=icon}
Return
Deep multi-modal convolutional transform learning network
.png){kind=icon}
PRE
AI
DOI：10.1016/j.neucom.2025.132182.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
• Novel analytical framework: We propose a Deep Multi-Modal Convolutional Transform Learning (DMCTL) network that replaces synthetic dictionary learning with an analytical paradigm, eliminating computationally intensive matrix decomposition. This innovation achieves a speedup in image denoising while maintaining superior PSNR and faster training convergence compared to traditional methods. • Multi-scale feature superiority: The integration of convolutional operators, multi-resolution analysis, and adaptive sparse constraints enhances cross-modal feature extraction, yielding an improvement in feature discrimination metrics. The framework demonstrates state-of-the-art performance in multi-exposure fusion, flash-guided denoising, and multi-modal medical image fusion (CT-MRI). • Extensible multi-modal architecture: DMCTL establishes a plug-and-play platform for integrating advanced technologies (e.g., vision transformers, optimization modules).
Journal
IF:
6.5
Papers:
2.5W
Citations:
6.5W