Channel Attention Network for Wireless Capsule Endoscopy Image Super-Resolution

Wireless Capsule Endoscopy (WCE) is a technology used for examination of Gastrointestinal (GI) tract. WCE is comparatively pain-free process to examine the parts of GI tract including internal walls of small intestine, stomach and esophagus. WCE uses a capsule with camera installed on board to record video while traveling through GI tract. The acquired data is transmitted wirelessly to a device outside the body. Despite numerous benefits of WCE technology, the size and battery capacity limit the quality of acquired images. One such qualitative degradation of the spatial resolution of the images which is coarser depending on the frame rate of acquired video. An effective and economic way to enhance the spatial resolution of recorded samples is by employing software-driven algorithms referred to as “Super-Resolution (SR)”. Recently, Deep learning-based approaches have been used in medical-domain due to their potential to obtain qualitative High-Resolution (HR) images without the cost of additional scans. This paper presents an approach for SR of WCE images with upscaling factor \(\times 4\) using deep neural network architecture which consists of a dense design of convolutional layers along with Channel Attention (CA) module to extract high-frequency details from Low-Resolution (LR) WCE images. The approach is validated on a derivative dataset of original Kvasir dataset consisting of 10, 000 samples and shows considerable improvement over the other state-of-the-art methods. Additionally, the perceptual quantitative assessment demonstrates the effectiveness of the proposed method over the others along with many distortion metrics.