基于纹理提取和优化的图像矢量化方法
时间:2018-12-10 21:04 来源:毕业论文 作者:毕业论文 点击:次
| 摘要矢量图即为运用几何特性来绘制的图像，以点、线或一些关键点的颜色等存储其图像信息。相较于一系列的色彩点阵组成的图案的位图，矢量图具有文件小、可无限缩放而不出现马赛克或失真、易于程式化与几何编辑等，并利于基于矢量图进行3D重构或动画编辑。在 CAD/CAM 软件应用、3D 打印技术、医疗、动漫及航空航天等领域具有明确的实用意义,而在各应用场合通常可能只有位图格式的资料。鉴此，本文旨在保持位图特征及优点前提下，进行基于位图图像的矢量化方法的讨论。 然而对纹理稍复杂的位图图像，我们的数值实验表明直接采用传统的图像矢量化手段结合颜色渲染系统所得结果并不理想。 因此，本文从位图图像特征着手，首先通过分析比较各种纹理提取算法，选取全变差方法完成对纹理图像的预处理，将原始位图处理为主结构部分和纹理部分，进而对这两部分依次进行处理。其次，对于主结构部分的矢量化以及颜色渲染，采用扩散曲线的方法进行矢量化处理。分割预处理图像后，将其转换为灰度图，通过 Canny 边缘检测提取图像的边缘信息以及细节信息，将这些曲线信息进行 K-Means 聚类以及 B 样条拟合，提取关键点的像素值。基于矢量化信息的基础，通过求解 Poisson 方程的扩散曲线方法进行颜色渲染。 最后对原始图像的纹理部分进行处理，将其与矢量化后的平滑部分进行融合，得到最终的矢量化图像。数值试验结果表明了方法的合理性和有效性。31398 毕业论文关键词 矢量化 纹理提取 K-Means 聚类 扩散曲线 B 样条拟合 颜色渲染 Title Image vector method based on texture extraction and optimization Abstract Vector graph is the image that uses the geometry feature to draw the image, with points, lines, or some key points color storing image information. Compared to on a series of color dot matrix composition pattern bitmap, vector graph with small files, infinite zoom without mosaic or distortion, can be easily programmed and geometry editing, and is conducive to the 3D reconstruction or animation editing based on vector graphics. In the application of CAD/CAM software, 3D printing technology, medical, animation and aerospace and other fields has clear practical significance, and in every application, it is usually possible to have only bitmap data. In view of this, this paper is to keep the bitmap features and the advantages of the premise, based on bitmap image of the vector method of discussion. However, the texture of the slightly more complex bitmap image, our numerical experiments show that the traditional image vector method is directly used to combine color rendering system and the results are not satisfactory. Therefore, this article begins from the bitmap image features. First of all, through the analysis and comparison of various texture extraction algorithm, selection method of texture image pre processing is completed, the original bitmap processing part of structure and texture, and then to the two parts in order processing. Secondly, the vector and color rendering of the main structure is processed by using the method of diffusion curve. After the pre segmentation image processing, its conversion to grayscale, detected by Canny edge extraction of image edge and detail information of the, K-means clustering and B spline fitting these curves information, extracts pixel values of the key points. Based on the vector information, color rendering is performed by solving the diffusion curve of Poisson equation. Finally, the texture of the original image is processed, and then fused with the smooth part of the vector, and the final vector image is obtained. The results of numerical experiments show the rationality and validity of the method. (责任编辑：qin) |