Study on a high-level processing for large size image using a effective geometrical graphic data structure

提出并研究了一种新型的基于软件的图像处理技术，它可以实现非常高效和智能的图像处理，特别是对各种类型的图形和彩色图像等大尺寸(或大体积)图像数据。在项目的第一阶段，通过对该课题组于1983年提出的一种多维数据结构BD树的设计扩展，引入了一种适用于2种三维图形基元(线、面、体等)的高效通用数据结构BD树(Graphic)。通过使用各种类型的真实地图数据进行计算机模拟，从内存使用和检索效率的角度揭示了这种图形数据结构的有效性。在下一阶段，提出并评估了基于软件的大图像处理的一般框架。大图像数据(通常由游程长度方法编码的8000x6000像素)首先被卷积…通过我们独特的方法更多地转化为适当的图形基元，例如轮廓向量，然后将其以名称的方式插入到BD树(图形)中，为以后的处理做准备。所有需要的图像处理功能都可以通过对代表原始图像的图形基元进行高效和智能的几何运算来执行。原始图形基元的抽象层次通过核心向量层、更高的图形基元层(如圆)和最终的语义识别层逐步提高。最后，提出了一种新的基于扩展三维BD树的颜色选择方法。通过从日本学术标准图像库SIDBA的1600万幅彩色图像中选择16色256彩色图像的实验，证明了该方法优于其他已有方法。较少

A new type of software-based image processing technique is proposed and investigated in this paper, which can realize very high efficiency and intelligence especially for large size (or large volume)image data such as various types of drawings and color images.In the first stage of the project, an efficient general purpose data structure suitable for 2 of 3 dimensional graphical primitives (lines, areas, volumes etc.),named the BD tree(graphic),has been introduced by devised extension of a multi-dimensional data structure BD tree which was proposed uniquely by this research group in 1983. Effectiveness of this graphical data structure has been revealed from the viewpoint of memory usage and retrieval efficiency through many computer simulations using a various types of real map data.In the next stage, a general framework for software-based large image processing has been proposed and evaluated. Large image data, typically 8000 by 6000 pixels coded by run-length method, are first conver … More ted into appropriate graphical primitives such as contour vectors by our unique method, which are then inserted bynamically to the BD tree(graphics) as preparation for hereafter processing. All the required image processing functions can be performed by a efficient and intelligent geometric operations on the graphical primitives representing the original image. Abstraction level for original graphical primitives are gradually enhanced through core vector level, higher graphical primitive level such as circle, and finally semantically recognized level. A drawing processor, the AI-MUDAMS is implemented using portable language C as a concrete embodiment of this general frame-work.In final, a new color selection technique using a similar framework using the extended threedimensional BD tree has been also proposed. Many experiments, including 16-color 256-dolor selection from 16 million color images in the Japanese Academic standard image database, SIDBA, reveals the superiority of the proposed method to other existing methods. Less