摘要
提出一种通用的融合基于可变形模型及区域特征统计的自动图像分割方案.像素相似性定义综合考虑灰度、梯度、色彩及纹理等多种图像因素,采用梯度矢量流Snakes算法自动进行目标匹配.为使分割结果更趋合理,利用相似单元分解技术对属于同一目标的区域进行自动合并,而对同一区域中属于不同目标的区域进行自动分裂,实现对未知目标个数及目标位置的输入图像的自动分割.对人工合成的低信噪比图像及自然图像的实验表明,本文方法性能良好,工作稳定,具有较强的处理边缘不连续及进入凹形边缘能力.
A general purpose automatic image segmentation scheme that integrates deformable models and region statistics was proposed. By using a combination of image cues including intensity, gradient, color, and texture rather a single source of information such as gradient magnitude, and gradient vector flow(GVF) Snakes technology, this scheme is able to automatically segment objects of an unknown number and unknown locations in images. For reasonable segmentation results, affine cell decomposition (ACD) technology was employed to automatically merge models corresponding to the same object, while split models corresponding to different objects. Experiments on synthetic images with low signal to noise ratio images and nature images show good performmance and robust of the approach, especially it’s capable to capture the discontinuous boundary and move snakes into boundary concavities.
出处
《红外与毫米波学报》
SCIE
EI
CAS
CSCD
北大核心
2000年第1期33-37,共5页
Journal of Infrared and Millimeter Waves
基金
自然科学基金!(编号 69772 0 0 2 )
关键词
可变形模型
像素相似性统计
自动图像分割
deformable model, active contour, gradient vector flow, pixel affinity statistic
