Self-supervised endoscopic image key-points matching

Feature matching and finding correspondences between endoscopic images is a key step in many clinical applications such as patient follow-up generation of panoramic image from sequences for fast anomalies localization. Nonetheless, due to the high texture variability present images, development robust accurate feature becomes challenging task. Recently, deep learning techniques which deliver learned features extracted via convolutional neural networks (CNNs) have gained traction wide range computer vision tasks. However, they all follow supervised scheme where large amount annotated data required reach good performances, generally not always available medical databases. To overcome this limitation related labeled scarcity, self-supervised paradigm has recently shown great success number applications. This paper proposes novel approach based on techniques. When compared standard hand-crafted local descriptors, our method outperformed them terms precision recall. Furthermore, descriptor provides competitive performance comparison selection state-of-the-art methods score.