CAJ | 학술논문

树木图像分割是将树木与其周围景物分离的技术，是虚拟现实和计算机仿真等学科在林业应用的核心技术，也是机器视觉领域的重要研究方向，拓宽了计算机技术在林业中的应用。本项研究基于树木图像形状复杂的特点，设计并实现了一种结合 C-V 模型水平集及形态学处理的彩色树木图像分割算法。运用改进的最小化能量函数作为水平集的演化曲线，可以更加自然地改变曲线拓扑结构，对含有分裂、合并、形成尖角等复杂形状的目标对象分割更为有效。如果再结合形态学后处理算法，将初次分割图像中非目标区的细密纹理和噪声剔除，可以快速准确地得到全局最优的图像分割效果。同时进行了与基于梯度变换的改进分水岭树木图像分割和基于灰度-梯度最大熵的树木图像分割算法的对比试验，试验表明，结合 C-V 模型水平集和形态学算法对树木图像分割效果更为有效。
수목도상분할시장수목여기주위경물분리적기술，시허의현실화계산궤방진등학과재임업응용적핵심기술，야시궤기시각영역적중요연구방향，탁관료계산궤기술재임업중적응용。본항연구기우수목도상형상복잡적특점，설계병실현료일충결합 C-V 모형수평집급형태학처리적채색수목도상분할산법。운용개진적최소화능량함수작위수평집적연화곡선，가이경가자연지개변곡선탁복결구，대함유분렬、합병、형성첨각등복잡형상적목표대상분할경위유효。여과재결합형태학후처리산법，장초차분할도상중비목표구적세밀문리화조성척제，가이쾌속준학지득도전국최우적도상분할효과。동시진행료여기우제도변환적개진분수령수목도상분할화기우회도-제도최대적적수목도상분할산법적대비시험，시험표명，결합 C-V 모형수평집화형태학산법대수목도상분할효과경위유효。
The trees image segmentation is a technology separating trees from its surrounding landscape, which is the core technology in virtual reality and computer simulation of forestry application and also is one of the focus areas in machine vision to provide basic data and technical support for the application of computer technology in forestry.Base on characteristics of complex shapes of the trees image,this paper designed and implemented a color image segmentation based on a set level of C-V model and later morpho-logical processing operation. The improved minimization of energy function was made as the level set evo-lution curve,because it could naturally change the evolution curve,and also more effectively segment com-plex shapes with parts of fission,mergence and sharp corner.Then combined with the later morphological processing operation,which could clear the non-target parts such as texture and noise from the initial im-age segmentation,finally the global optimal image segmentation could be obtained fast and accurately.At the same time,we compared the color trees image segmentation,combined with a set with level of C-V model and the later morphological processing operation,with the improved watershed trees image segmen-tation based on gradient transform and the maximum entropy trees image segmentation based on gray gra-dient.The results showed that the method combined a set level of C-V model and the later morphological processing operation was more effective.