论文部分内容阅读
由D.Donoho等人提出的能有效捕获图象的几何结构,但基于变换的多尺度分割算法在楔形方向的选择上需要计算所有分解楔形系数,且没有利用上层分解的结果,计算量特别大。从图像的几何结构出发,在图象四分树的基础上加以楔形区域分割,对于矩形区域的楔形方向选取上建立了多分辨分析算法,在上层分割的基础上,只需计算八个方向的Wedgelet,而不是所有的方向,既避免了窗口初始化,降低了分割过程特征抽取的复杂性,减少了迭代次数。经试验比较,该方法优于同类方法。
The geometry proposed by D. Donoho et al. Can effectively capture the image. However, the transform-based multiscale segmentation algorithm needs to compute all the decomposed wedge coefficients in wedge direction selection and does not make use of the result of upper decomposition, . Starting from the geometric structure of the image, the wedge-shaped area is divided based on the image quaternion, and the multi-resolution analysis algorithm is established for the wedge-shaped direction selection of the rectangular area. On the basis of the upper layer segmentation, only eight directions Wedgelet, not all directions, not only avoid the window initialization, reduce the complexity of feature extraction in the segmentation process, reduce the number of iterations. The experimental comparison, the method is better than similar methods.