The Extraction Of Sheep Based On An Improved Non-interactive Grab Cut Algorithm

We have to consider the over-grazing pasture system when we construct the digital system, and the object of discussion is how to achieve balance between livestock, thus contributing to ecologically sustainable development. It is in this context, the paper work is done, some will be taken to the sheep pasture in the picture as a research vehicle, the paper addresses how many sheep in the flock, with emphasis on non-interactive extraction algorithm.Interactive image segmentation technology has been one of the field of computer vision and pattern recognition research.with many algorithms have been proposed, the field becomes more mature, have changed the result of segmentation be more precise. Image segmentation according to different classification criteria classification results will be different, which Shortest Path, Graph Cut algorithm, Intelligent Scissors and Level Sets algorithm is commonly used algorithms. In order to achieve non-interactive image segmentation, the need for a way to mark the image instead of the user's strategy, this interactive technology is to use the edge of Africa to achieve matching. It is the first use of the specific thinking of Hausdorff from the algorithm to the template image and the original image edge matching, and the result of the match started as a Grab Cut Algorithm for the initial conditions, the use of Grab Cut algorithm mentioned in the GMM model image RGB color model, then use the iterative formula for the minimum energy way, the algorithm of the energy equation is based on the Graph Cut algorithm to improve from the energy equation can also be called the partition cost function, by definition, for all the division to meet the hard constraints, of course, the smaller the resulting split costs will be more accurate segmentation results, the hard constraints mentioned here refers to already know the background or foreground region are some of pixels, the pixels in the image the user prior on the calibration point or region. Grab Cut algorithm was selected and Hausdorff distance from the matching algorithm to generate the prospects for a non-interactive extraction algorithm because Grab Cut algorithm not only considers the image boundary properties were also considered regional attributes, you can use the formula that add texture as the edge; The Hausdorff distance part of the matching algorithms from the Hausdorff distance algorithms under occlusion for the serious prospect of extracting very good results. Through experiment,the improved non-interactive Grab Cut algorithm achieves the desired results, as long as the use of the edge of the matching template set sufficient, even if the existence of occlusion detection accuracy rate can reach 80% or more.In this paper the edge templates generated by way of artificially, and its advantage is the generation of the template set is very convenient and fast, the disadvantage is that if the foreground objects to be changed, we need to re-generate the template set of man-made. In this region, foreign scholars had tried to have the Grab cut algorithm in combination with the sample classification which made Obj Cut algorithm. Compared to the paper, Obj Cut has the advantage of generating the edge templates collection different ways, Obj Cut algorithm to the sample quantity and quality of pictures very high demand, and generate the template collection requires a lot of computation, does not meet the requirements of real-time foreground extraction.