Attentive Convolution for Efficient Image Denoising

Denoising, a process that helps to remove corrupted pixels from images which are introduced during capture. A novel denoising method, Attention Convolution Block (ACB) is introduced in this paper to mitigate the issue like removing noise and preserving image quality effectively. Attention Convolution Blocks (ACB) have been devised to enhance feature points in the image. ACB integrates the Convolutional Block Attention Module (CBAM), combine both types of response that is channel and spatial into unified network and proved useful in improving the ability to deal with visual features by analyzing them as patterns that resemble blocks at varying levels. Each AC-Block compromises a Convolutional Block followed by an Attention Convolutional Module, which was designed using feature squeezing techniques to identify more complex structure exists within or between topological neighbors. Convolution blocks a chain of convolution layers with Batch Normalization (BN), which are fine-tuned for stability across different applications. CBAM introduces attention maps which specialize on different discriminative parts of the feature maps, enabling more flexible and interpretable visualization other than region-level information for easy use in the pixel-wise downstream tasks. The proposed architecture is intended to maximize feature extraction and processing at each level, significantly improving the performance of the model. Qualitative and quantitative tests clearly states that the proposed architecture outperforms existing approaches in a variety of applications, proving its effectiveness and robustness in removing high noise. The evaluation of proposed method is validated by the Peak Signal-to-Noise Ratio (PSNR) and the Structural Similarity Index Metrics (SSIM).