摘要： 针对传统水平集算法只采用图像沿x、y两个方向的导数来作为图像梯度的缺点，提出了一种新的基于多方向的水平集研究算法。对图像进行两层方向可调金字塔分解，将图像分解成不同尺度、不同方向的子图像，并将水平集停止项中不同尺度的图像梯度由原来的两方向扩展为四方向，采用插值法将边缘提取结果从粗尺度投影到细尺度上。实验结果表明，该方法优于传统水平集变换法。

关键词: 水平集, 方向可控滤波器, 边缘提取, 多尺度

Abstract: A new level set way based on multi-direction gradient is proposed, in order to solve the problem of the level set gradient that is got only through the derivative of x and y direction. The two-layer steerable pyramid is used to get some different scale and directional images. And the different scale image gradient of the level set stop item adapts the two-direction gradient modulus in the old model to the four-direction gradient modulus. The interpolation method is used to pass the result from the coarser scale to the finer one. The method is compared with the original level set method. And the experimental results show that the developed algorithm is verified better than the traditional one.

Key words: Level set, Steerable filter, Edge detection, Multiresolution