基于各向异性自适应高斯加权方向窗的非局部三维Otsu图像门限分割

Non-local Three-dimensional Otsu Image Thresholding Segmentation Based on Anisotropic Adaptive Gaussian Weighted Window

针对传统3维Otsu(3D-Otsu)门限分割方法中的滤噪性能和小目标保持性能的不足,该文提出一种基于各向异性自适应高斯加权方向窗的3D-Otsu门限分割的新方法。新方法改进了3D-Otsu的邻域窗口设置方法,采用中心点的局部特征来自适应地确定邻域各向异性高斯加权方向窗口的尺寸、尺度和滤波方向。然后,提出非局部多方向相似度测量来更有效地捕捉图像中的模式冗余。最终,结合像素点灰度值、加权均值、加权中值构建3维直方图,并基于最大类间方差计算门限矢量进行分割。实验结果表明:与目前广泛使用的2维Otsu,2维最大熵以及传统3维Otsu方法相比,新方法有着更好的门限分割效果,并具有更好的滤噪性能和小目标保持性能。

Because of the shortage of noise removal and small target preservation for the conventional threedimensional Otsu （3D-Otsu） method, a new method based on adaptive Gaussian weighted directional window is proposed. The new method improves the window setting method of the 3D-Otsu. The window size, scale and filtering direction are adaptively determined by the local characters. Then, based on the proposed non-local multiple directions similarity measurement, the pattern redundancy in the image can be captured effectively. Finally, the 3D histogram is constructed based on the gray value, weighted mean value and weighted median value, and the threshold vector is computed by the maximum between-class variance method to segment the image. Compared with the commonly-used 2D Otsu method, 2D max-entropy method and 3D-Otsu method, the proposed method has better segmentation performance, with better performance for noise removal and small target preservation.