Region of interest detection with iterative colour clustering and a global weighting scheme

Many modern surgical systems utilize advanced computer vision and image processing techniques for a vast variety of tasks, such as object recognition, segmentation, classification and 3D reconstruction. In many cases, the extraction of a specific region of interest (ROI) in the image is performed prior to further processing steps. This increases the computation speed and may improve the quality of the system´s output because misleading elements in the image are masked out. We present a novel approach that considers global colour distributions to extract one (or more) ROIs in a camera image stream. Our method is both fast and robust against illumination discontinuities, irregularities in the texturing of surfaces and false areas with very similar colour. The method is used for the detection of human skin in an operating room environment for a camera based navigation system.