Research On Active Contour Model And Its Application To Some Complex Cell Image Segmentation

The analysis of cell image becomes more and more important in recent scientificresearch. In this paper, the problem of segmentation for complex plankton, protoplastand plant cell images is discussed. It is difficult to segment these cells using thetraditional image segmentation methods because of their blurred edge andcomplicated grey-level distribution. The active contour model based on level-set hasmany advantages which can be showed as follows: It is not sensitive to noise,numerically stable and is easy to join the prior knowledge. This method is widelyused in medical image segmentation, three-dimensional reconstruction and videotracking and so on. This method is used to solve the problem of segmentation for thecomplex cell image in this paper.The traditional active contour model is time-consuming and can’t obtain goodsegmentation effect for the cell images in this paper. On the basis of traditional activecontour model and fast active contour model, some improved fast active contourmodels are proposed and can get good effect for the complex cell image in this paper.The main contents of this paper are as follows:(1) A novel histogram-based fast active contour model is proposed. Thetraditional active contour model based on probability estimation is discussed then theidea of histogram is introduced in the fast active contour model. The old speedfunctions based on threshold and gradient are transformed into a novel speed functionbased on histogram. Experimental results show that the new method can segment theedge-blurred and complicated plankton cell quickly and accurately.(2) A new circle dependent fast active contour model is proposed. The protoplastcell image has the features that the grey level of background region and part region ofobject is very close as well as that the edge of the object is blurred. Therefore, it isdifficult to obtain good effect using the traditional segmentation methods. The circularinformation is joined into the fast active contour model because the protoplast cell isround, and the improved method can get fine segmentation result for a singleprotoplast cell. On the basis of this method, to solve the problem of segmentation formultiple protoplast cells, the fast active contour model composed of one level-setfunction is improved as the one composed of multiple level-set functions. Firstly, theone level-set model is used to do pre-segmentation. Secondly, the one level-set regionis split into multiple regions. Lastly, the circle dependent fast active contour model is applied to accurately segment every cell. The method of eight directions chain codetracking and the random Hough transformation are used to split the single regionrespectively for the multiple cells that are separated and overlapped from each other.Experimental results demonstrate that the proposed method is able to segment bothseparated and overlapped multiple protoplast cells accurately and quickly.(3) The color difference in HSI color space is introduced into fast active contourmodel and then a novel fast active contour model based on color difference isproposed. Unlike the traditional fast active contour model, the new model is driven bycolor information which includes the grey intensity, hue and saturation, so it is able toget better segmentation result for color image. Experimental results indicate that theproposed model can get good segmentation result for the color plant cell images inthis paper.