三维二值图像连通体标记处理的高速化研究

Three-Dimensional (3D) images have been widely used in fields of medical treatment, biology, industrial detection, and scientific computing. Connected component labeling (CCL for short) is one of the most important basic processes for recognizing objects in 3D binary images. Because numbers of voxels are large and connected relation between voxels are very complicated in 3D binary images, research on how to speed up processing of CCL in 3D binary images has important academic value and practical significance. In our proposed project, we will extend our successful experience and methods achieved in the research on high-speed CCL algorithms for Two-Dimensional (2D) binary images, and find the optimal checking order of 13 adjacency voxels to reduce the numbers of voxels to be checked in processing the current object voxel. Moreover, we will adopt suitable data structure and optimal combined order for multiple equivalent labels merging processing. Thus, we will propose a more efficient CCL algorithm than the existing algorithms for 3D binary images. According to the fact that objects in 3D binary images are solid bodies in most practical applications, we will propose a super-voxel-based fast CCL algorithm for 3D binary images, and a one-scan-based algorithm for calculating solid object’s volume of 3D binary images. By the achievements of our project, the efficiency of CCL processing for 3D binary images will be improved greatly. Our research will provide a stronger technical support for practical applications and make some academic contributions to pattern recognition, computer science and machine vision related to 3D images.

三维图像已被广泛用于医疗、生物、工业检测和科学计算等领域。连通体标记处理是三维二值图像物体识别不可或缺的基本处理之一。由于三维图像的体素数巨大、体素之间的连接关系极为复杂，三维二值图像连通体标记处理的高速化研究具有重要的学术价值和现实意义。本项目借鉴申请团队在二维二值图像连通体标记高速化研究中取得的成功经验和方法，并通过确立13个邻接体素的最佳检查顺序减少处理当前物体体素时邻接体素的检查次数，采用适用于多个等价标号合并处理的数据结构及合并顺序，提出比现有的三维二值图像连通体标记算法更有效的算法；根据现有实际应用中三维二值图像中物体多为实心体的现状，提出基于超级体素（段、基本立方块）的三维二值图像连通体标记的高速算法及基于一次扫描的体积计算算法。旨在实现三维二值图像连通体标记处理的高速化，为三维图像的实际应用提供技术支持，为推动三维图像的模式识别、计算机科学及机器视觉的学术发展做出贡献。

随着三维成像技术的快速发展以及三维图像包含信息的多元性等优势，三维图像已被广泛用于医疗、生物、工业检测和科学计算等领域。连通体标记处理是图像识别不可或缺的基本处理之一，由于三维图像的体素数巨大、体素之间的连接关系极为复杂，三维二值图像连通体标记处理的高速化研究具有重要的学术价值和现实意义。为此，本项目开展了连通体标记处理高速化问题的关键理论和方法的系统研究，通过利用体素间的邻接关系以及处理过的体素信息标记当前体素的原则，确立邻接体素的最佳检查顺序，减少了处理当前物体体素时邻接体素的检查次数，另外，采用适用于多个等价标号合并处理的数据结构及合并顺序，提出了比现有方法有效的连通体标记算法；提出了在进行连通体标记处理的同时提取各个物体的各种形状特征的快速算法；利用连通体的标记处理提出了纸病的快速检测、角点的快速检测以及纸张填料粒径分析等算法；在图像测试数据集上测试了所有所提算法的正确性和有效性。本项目的开展取得一系列研究成果，大大缩短了三维二值图像连通体标记处理所需的时间，为三维图像的实际应用提供技术支持，为推动三维图像的模式识别、计算机科学及机器视觉的学术发展做出了贡献。