Semiautomatic hysteroscopic video summarization

Hysteroscopic videos have been used to measure the uterus appearance, and are of fundamental importance for gynecologists. Unfortunately, such videos contain lots of information, and only a reduced number of frames are actually useful for diagnosis purposes (e.g. analysis of the spatial distribution of the glandular openings). We propose a new method for hysteroscopic video summarization, with the objective of reducing the number of frames necessary for the video contents representation. Our method is based on an extension of known statistical principles, with advantages over previously published works. Our method is adaptive, in the sense that it minimizes the need of parameter adjustments. Also, it is robust to changing illumination conditions, and to artifacts, that are common in such videos. Experiments with our method have indicated compatibility with the visual analysis results, as carried out by specialists.