Super-Resolution Techniques Evaluated in the Context of HD Endoscopic Imaging

In this work we evaluate different reconstruction-based super-resolution algorithms in order to enhance the spatial resolution of endoscopic images acquired with an HD endoscope. The main question we try to answer in this work is whether it is feasible to employ super-resolution algorithms in order to increase the resolution of our HD images to reveal new details which may have got lost due to the limited endoscope magnification inherent to the HD endoscope used (e.g. mucosal structures). For this purpose we first briefly explain the principles of super-resolution algorithms. We then describe the evaluated algorithms in more detail, which is followed by experiments using different super-resolution algorithms. We show that for artificially generated sequences and non-endoscopic real-world sequences the SR algorithms are indeed able to deliver a higher quality as compared to a simple bicubic interpolation. In case of endoscopic sequences, however, there is no real improvement observable. But we also show that the quality measures used very frequently show a rather high disagreement upon an improvement.