均值漂移（MeanShift）算法是一种高效的非参数、迭代模式搜索方法。它完全依靠特征空间中样本点进行分析，不需要聚类数目等先验知识，也不需假设样本的分布状况，近年来广泛应用于图像提取、图像滤波以及视频识别、提取等计算机视觉领域，并取得了比较满意的效果。本课题主要研究均值漂移在原木检测系统中的应用，实现原木端面自动识别、提取，原木端面识别是原木材积自动检测中的关键步骤，是实现原木检尺自动化的基础工作。

本文首先运用Mean Shift算法实现对图像的平滑，并运行Meanshift实现对原木图像的初次粗聚类提取，将图像提取成若干类别，由于Meanshift参数的选择会对提取结果产生影响，设计算法实现Meanshift相关参数根据实际情况自动选择；然后区域合并策略在图像提取中起着非常关键的作用，为了实现对初次粗聚类结果进一步的细提取，研究区域相似性度量指标，并选择或构建相应的相似性度量准则，根据区域合并策略、区域停止准则，实现相似初次粗聚类区域的合并，完成图像提取，从而实现原木横断面的自动提取。

 

关键词：目标识别、提取，Meanshift，原木端面

 

    Mean shift (MeanShift) algorithm is a highly effective non-parametric iterative pattern search methods. It relies entirely on the sample point in feature space analysis, does not require a priori knowledge of the number of clusters, etc., without also assumed that the distribution of the sample, in recent years, widely used in image segmentation, image filtering and video recognition, computer vision and other segmentation, and achieved satisfactory results. This thesis mainly study the mean shift in timber inspection system applications, to achieve the end face of timber automatic identification, segmentation, Log Cross recognition is the original wood product automatically detects the key step is to achieve the automation logs gauging the basic work.

    Firstly, using Mean Shift algorithm for image smoothing, and run Meanshift achieve initial rough clustering logs image segmentation, the image is divided into several categories, because of what the results would be split Meanshift parameters affect the design algorithm Meanshift Related parameters are automatically selected according to the actual situation; and regional consolidation strategy plays a key role in image segmentation, in order to achieve the initial coarse clustering results further fine segmentation, regional similarity metric, and select or construct appropriate similarity measurement criteria, according to the regional consolidation strategy, regional stopping criteria, achieve a similar initial rough clustering region merging, to complete the image segmentation, enabling the automatic extraction of logs cross section.

 

Keywords: recognition, segmentation, Meanshift, Log Cross