Image Segmentation Using Color Information and Its Application in Colonscopy

An image segmentation algorithm, based on human colour perception, has been designed and implemented. An image in the RGB space is obtained through a conventional frame grabber. It is transformed into a perceptual colour space (HSI) and a histogram is constructed to estimate the size of any required feature. A regular decomposition of the image is then made, in which each node contains statistical information about the colour attributes of the pixels in the corresponding region. The best node is selected as a seed, and a merging process then obtains the boundary of the region. The algorithm has been tested on the identification of fluid in colon images observed through a conventional endoscope. In the intended application the segmentation will be part of a control system, and will enable the instrument to suck fluid out of the human colon automatically. Preliminary results suggest that a fast and accurate segmentation can be obtained using simple preceptual colour criteria. Real time performance could be achieved by implementing the algorithm on a small pyramid architecture.