Improved C-V Model For The Leaf Margin Image Segmentation

Leaf margin contains a lot of important information of plant characteristics, and the accurate extraction of leaf margin is helpful for the quantitative analysis of plant growth which has important application value in 3D recognition of leaves and machine recognition. Image segmentation can extract the effective margin information, which is helpful to identify features. This paper mainly does the following research on the problem of leaf margin segmentation:This paper introduces the theory of curve evolution, level set modeling and its numerical solution. Then presents a geometric active contour model based on level set and Mumford-Shah model, namely C-V model (Chan-Vese model), which is used to capture the edge of the target, for image segmentation. Advantages of C-V model are introduced such as it is a gradient-independent algorithm, it is suitable for margin extraction as its noise robustness, simple modeling, independent initial direction and wild available range. However, there are defects such as C-V model can't properly extract contours from given images in homogeneous areas, and with the increase in the number of iterations, it often needs to re-initialize the signed distance function (SDF) which increases evolving time of C-V model.This paper makes improvements on the two defects:(1) The fast C-V model based on local statistical information has been designed. First, increase the use of local information, as combine the local information and global information effectively, so in the rich detail area C-V model can converge to the correct target edge. Then use an internal energy functional term to describe the degree of deviation from the SDF. By increasing the iteration times of solving level set to ensure that the level set won't deviate seriously from the SDF and, to a certain extent, saving evolution time in this way. In short by combining the local statistical information and constraints of energy function to get the details better and faster convergence speed as well. Experiments has been tested to verify the practicality of the improved model. (2)The fast segmentation algorithm of curvature-independent direction based on C-V model has been designed. First, it combines an edge function which is curvature-independent directional to remedy the defect of evolving without using edge information effectively. Then uses mean curvature motion to minimize the first-order length energy function, to overcome the defect that the effect of active contours is not so good when the inside and outside of target edge is homogeneous region. And then the internal energy functional item of the energy function is increased to simplify the model when it needs to re-initialize the signed distance function, improving processing speed in this way. The improved model is applied to the leaf margin extraction, by comparative analysis of experiments to verify the improved model can be more effective and practical.