Image segmentation, Grabcut, Maximum flow, minimum shear, Gaussian mixed models.
Image segmentation is a field Relevant research in image processing.
Many advanced algorithms have been developed to separate a region of interest in a color image of its background ( snakes , live-wire , among others). However, the results obtained are not satisfactory in many cases. More accurate methods are based on representing the image as a graph and separate it into two sub-graphs that represent the region of interest( foreground ) and the background ( background ). The GrabCut algorithm belongs to this category . In this work we present the functional theoretical data and detailed implementation of the Grabcut algorithm with some improvements not presented in its original version. In particular, the calculations of the N-Link , T Link and minimum cut-off were modified. These changes allow better results in the pixels of the border between the foreground and background ,as well as speed up the minimum cut algorithm. Our implementation shows good results for test images used.