Towards a benchmarking framework for quality-optimized endoscopic video stabilization

In invasive diagnostics and therapy, especially in microsurgery, there is a trend towards miniaturization of medical devices to keep operative invasion at a minimal level. This affects endoscopic systems in particular: Endoscopic video recordings or live-feeds during operation become more prone to interference by shaking due to continuous miniaturization of the endoscopes, with increasing difficulties in interpretation by a surgeon. Hence research on video stabilization algorithms must extend to the medical field of endoscopy. Numerous video stabilization algorithms have been developed, dealing with the three steps of stabilization, namely motion estimation (ME), motion compensation (MC) and image composition (IC). Yet only little research has been made on procedures to compare different video stabilization algorithms - even if limited to quality/accuracy of the stabilization algorithm and leaving out real-time performance and computational costs. Our goal is to develop a combined and unified scalar measure of only the quality of all video stabilization algorithms, i.e. one that does not consider computational complexity. This measure is composed of relevant procedures discussed and introduced in this paper, which are weighted according to their importance. These procedures are compared, a measure for benchmarking is developed and these measures are applied to selected original and stabilized videos from Bronchoscopy and Rhinoscopy.