Research And Application Of Active Contour Model In Tree Image Segmentation

The research of forestry automation plays an important role in the modern science, this paper mainly research the segmentation problems of log end and plant leaves in complex natural background. These images with complex natural background and highly similar characters between background and foreground are difficulty to segment with traditional segmentation algorithm. The active contour model based on level-set has many 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 widely used in medical image segmentation, three-dimensional reconstruction and video tracking and so on. This method is used to solve the problem of segmentation for the complex forestry image in this paper.It is easily to be interfered by the background and noisy when the traditional active contour model segments image in complex natural background because of the lack of effective constraints of the evolution of contour line, which results in over segmentation. This paper improved the amount of energy model according to the characteristics of image segmentation based on the traditional active counter model as well as the introduction of shape constraint. This paper formatted some of the new active counter model algorithm through the research and improvement, which can get good effect for the complex forestry image. The main contents of this paper are as follows:(1) This paper established an improved CV-LIF active contour model based on the circular restriction. According to the segmentation problem of log end image in natural background, this paper improved the traditional C-V model, combined with the characteristics of log end image, which fitted the internal evolution curve with gradient and the LIF (local image fitting) model with the integration into the local image as well as the addition of circle prior, and all of that can constrict the evolution of contour line combined the global energy and local energy. To segment the single and multiple log end images respectively, and the results show that this method can do the segmentation of log end image efficiently.(2) Plant leaf image is difficult to obtain ideal segmentation effect whit the traditional active counter model without constraints because of the complex background that has high similar with the goal and the abundant gradient information in the image. According to these problems above, this paper introduced the polygon model to simulate the shape of the leaves combined with the biology classification knowledge of the shape of leaves, which on one hand it made the shape of the leaves obtained as the initial counter of active contour model, and on the other hand it can be the shape constraint to contour line. At the same time, this algorithm improved the gray gradient of traditional model to the color gradient of HIS space, which enhanced the ability of segmentation of colorful leaf images. Experiments show that this method can constraint the evolution of contour line efficiently and can avoid the interference of background, which obtained the high segmentation accuracy.