一种基于GVF变形模板的椭圆提取算法

给出一种结合梯度矢量流（GVF）及椭圆变形模板的椭圆提取新算法。普通的动态轮廓算法仅提供局部的连续与正则性约束，虽然使得曲线可以灵活适应各种不同的边缘形状，但也导致轮廓强烈敏感于图像噪声及邻近边缘点，引入全局模型是改善性能的关键。给出一种新的基于椭圆形状约束的变形模板算法，其能量最小化过程也直接在椭圆的参数空间中进行，从而保证了提取结果一定是椭圆；算法与GVF相结合，大大扩展了边缘能量的捕获范围，而不致造成边缘模糊；此外，本算法容许同时提取多个椭圆，且可利用各个椭圆参数及其相关性（例如同心椭圆）等先验知识，从而实现快速、准确、鲁棒的椭圆提取，仿真结果与实际图像应用验证了算法的有效性。

Ellipse Extraction Algorithms Based on GVF Deformable Templates

A new ellipse extraction algorithm based on the ellipse-special deformable templates and gradient vector flow (GVF) was proposed. In general, the general active contour models is constrained barely by local features such as continuity and regularity, which made the curve moving flexibly and being capable of capturing various edge shapes. Nevertheless, this character consequently causes the contour to be highly sensitive to noise and nearby edges. Introducing global model is in evidence the critical procedure to improve it. A novel ellipse-special deformable templates technique was illustrated, which brought in a strict global constraint on the contour. The energy minimization is also accomplished directly in the space of parameters used to describe the ellipse, thereby ensures the solutions to be ellipses. With the introducing of GVF force field, the capture range of edge energy increases remarkably without blurring the edge. Besides, this approach allows the synchronous extraction of multi-ellipses and the priori knowledge about the parameters of each ellipse as well as the correlated features between them (such as concentric ellipses etc) can also be integrated with easily, such that a fast, accurate, robust ellipse extraction technique is acquired. Experiments with simulated data as well as real image are presented to validate the algorithms.