细粒度背景下强对比度纹理无监督提取-计算机辅助设计与图形学学报.PDF

第27 卷 第4 期 计算机辅助设计与图形学学报 Vol. 27 No.4 2015 年4 月 Journal of Computer-Aided Design Computer Graphics Apr. 2015 细粒度背景下强对比度纹理无监督提取 马 爽, 许 刚 (华北电力大学电气与电子工程学院 北京 102206) (lanxi1279@126.com) 摘 要: 针对相似细粒度背景下容易出现强对比度纹理误提取的问题, 提出一种两通道纹理图像无监督提取算法. 通过对图像水平、垂直方向梯度场的非线性扩散, 在不改变空间目标边界位置的前提下获取主纹理的边缘结构与区 域灰度特征; 同时建立包含调整项与模糊因子的两通道纹理提取主动轮廓模型, 以具有较大差异的特征为主导项驱 动曲线演化, 并采用水平集方法实现对强对比度纹理的无监督提取. 实验结果表明, 该算法对多种自然纹理的提取 具有较高的准确性和计算效率. 关键词：强对比度纹理; 两通道特征; 灰度调整项; 边缘模糊因子; 主动轮廓模型 中图法分类号：TP391 Unsupervised Extraction of High Contrast Texture from Fine-Grained Backgrounds Ma Shuang and Xu Gang (School of Electrical and Electronic Engineering, North China Electric Power University, Beijing 102206) Abstract: To reduce common inaccuracies in the segmentation of high-contrast textures from fine-grained back- grounds, a two-channel unsupervised approach for texture extraction was proposed in this paper. Without chang- ing the locations of target boundaries, edges and regional gray-scale features of high-contrast texture were ac- quired through the nonlinear diffusion of pixel intensities, horizontal gradients and vertical gradients of the image. In addition, a two-channel active contour model that contains fuzzy information and adjustment factor was estab- lished. In this model, the feature with larger differences between textures was taken as the leading term to drive the evolution of segmentation curves. Moreover, the level-set method was employed to achieve unsupervised ex- traction. Experimental results on various textures demonstrated the improvements in terms of segmentation per- formance and computational efficiency. Key words: high contrast textures; two-channel features; gray-scale adjustment term; edge fuzzy factor; active contour model ,