DCD-Net: Image Deraining with Delta Convolution and Joint Calibration Attention

Rain streaks typically result in pronounced visual degradation and foreground occlusions, impeding the progress of visual tasks in outdoor scenarios. Presently, predominant methodologies for image deraining rely on Convolutional Neural Networks (CNNs). However, CNNs tend to focus excessively on low-level visual features, exhibiting limited capability in capturing high-dimensional, global features. They lack targeted attention to channel information and spatial details. Consequently, this paper propose the Delta-Calibration Derain Network (DCD-Net). This model introduces a sequential delta convolutional layer (DCL) structure to expansively augment the feature acquisition range. Furthermore, this study pioneers the Joint Calibration Attention (JCA) module to precisely capture channel and spatial feature information, thereby elevating network performance. Experimental findings substantiate that the proposed methodology outperforms conventional CNN-based deraining models, demonstrating superior visual performance.