基于梯度向量流的图像骨架提取方法-系统理论专业论文.docxVIP

基于梯 基于梯度向量流的图像骨架提取方法 摘 要 随着现代科技的发展，图像资源飞速增长，基于图像分析与理解技术的图像目标检 测，是人工智能、模式识别和图像处理等相关领域的关键问题。图像的形状特征是最常 用的图像目标特征，它包含了很多视觉信息。骨架是一种保持拓扑结构的形状特征，具 有直观的结果，且易于相似性计算，是目标识别的有效工具。本文在寻求一种抗噪性强、 保持骨架拓扑性与连续性的骨架提取方法。 本文分析和总结了目前国内外骨架提取的相关研究进展，以及五种典型的骨架提取 方法。这些骨架提取方法都普遍存在骨架不连续性和不能保持拓扑性等缺点。本文取神 经元图像作为实验对象，提出了基于梯度向量流的骨架提取方法，使其保持拓扑性与连 续性。 首先，本文分析了梯度向量流的相关特性，提出了几种关键点：吸引点、排斥点、 鞍点与高曲率点。本文认为这些关键点就是神经元的骨架点。为了连接这些骨架点，完 成连续的骨架，本文提出了一种基于边界权值的最小生成树。其次，由于同一个神经元 在不同时刻所表现的形态不一样，但是其骨架的主体部分基本不变，本文提出了图形态 学腐蚀算子，用于进一步调整神经元骨架结构和删除小刺和非脊柱的分支。 最后，对不同背景与不同结构的神经元提取骨架，实验表明，该算法对大幅度变异 的神经元与复杂背景下的神经元提取出良好的神经元骨架。说明该算法具有良好的稳定 性、抗噪性和保持拓扑结构性质。 关键词：图像骨架；梯度向量流；最小生成树；图形态学；骨架树 Abstract With the development of modern technology,image resources are increasing dramatically. Image object detection based on image analysis and understanding is the key issue in artificial intelligence, pattern recognition, image processing and other areas. The feature of the image shape is the most common feature of the image. It contains a lot of visual information. And the skeleton, as a kind of the shape feature keeping the topology structure and the visual result, is a powerful tool to object recognition, because it is easy to calculate the similarity of objects. This paper aims to find a method to extract skeletons that has strong noise immunity and maintains the skeleton topology structure and the continuity. This paper presents the lasted researches in skeleton extraction at home and abroad. And then five typical extraction methods are described and analyzed, with which some common problems existing in the current skeleton extraction algorithms are summarized. The discontinuity and not keeping the topological structure are ubiquity disadvantages. This paper takes neurons image as experimental objects, and proposes a kind of skeleton extraction method based on gradient vector flow, which effectively maintains the skeleton topological structure and the continuity. Firstly, this paper analyzes the related characteristics of gradient vector f