基于多模型融合的木材表面缺陷图像快速识别

Fast Recognition for Wood Surface Defect Image Based on the Multi-model Fusion

根据木材缺陷图像识别技术的现状,针对适合识别木材各种表面缺陷图像的算法,对现有典型模型法进行图像识别方法的分析,提出了一种基于多个模型融合的木材表面缺陷图像快速识别算法。首先,在C-V模型中引入加权总变分能量(VTg(u)),使得二者分别能够与GAC模型连接,实现了在同一全局最小化框架下统一2种模型;然后采用全变分范数对偶化方法对模型进行了快速求解;最后给出了模型的数值化实现算法。结果表明:该算法不依赖初始轮廓线的选择,能够比较快速、准确地识别出木材的节子、孔洞、腐朽、空心等缺陷和单板多节子缺陷图像。

With the algorithms for all kinds of wood surface defect images , we analyzed image recognition method of the existing typical models, and put forward a fast recognition algorithm for wood defect images based on the multi -model fusion.First, the weighted total variation energy VTg（u） was introduced in ROF model and C-V model, so that two models could, re-spectively, connect with GAC model.Therefore, we achieved the unity of the three models in the same global minimization framework.And then, we used a dual formulation of TV norm to realize the fast minimization process .Finally, we provided the numerical algorithm of the model .Our algorithm does not depend on the choice of the initial contour , and it can quickly identify the outline of a variety of wood surface defects including the knots , holes, rot,hollow and veneer defect images with multiple knots .