纹理图像中重复纹理元素提取方法研究

发布时间：2018-08-23 09:22

【摘要】：纹理图像中重复纹理元素的提取是指将组成纹理图像的具有相同或相似特性的重复单元(即纹理元素)的数据结构提取出来。重复纹理元素提取的目的是识别纹理图像中具有相似视觉特征的区域,将图像的表现形式加以简化或者改变,使复杂的纹理图像归宿为简单独立的单个纹理元素的重复组合,以便图像更容易被计算机理解和分析,同时为主观重组设计纹理图像提供必要的前提条件。其中图像的分割通常参照如下规则:部分区域表示目标纹理元素的基本结构;其他区域则表示与目标存在一定差异的同质区域,即纹理背景区域。一方面,纹理元素的提取能够将纹理图像分解成独立的单元模块,而独立的纹理元素代表了纹理图像的基本构成,因此有效利用已抽取的纹理元素来分析原纹理图像的拓扑结构,是对纹理图像组成结构的一种有效研究方法;另一方面,提取出的纹理元素还可以用于纹理的合成,亦可以由此产生新的纹理图像,为纹理图像的转移、纹理图像的组合、纹理图像的设计打好基础,奠定有力的前提条件。为了实现纹理图像中重复纹理元素的提取,本文通过对现有重复元素提取方法的分析总结与研究,提出了一种交互式纹理图像中重复纹理元素提取算法,该算法能够在用户提供少量交互的情况下,较好地实现对纹理图像中具有相关性的颜色或纹理特征的重复纹理元素的同时提取。我们通过大量实验,充分验证了本文算法的有效性与实用性。本文算法的组织结构主要分为以下几点:1.算法采用颜色聚类方法将源纹理图像分割成独立且不连通的图像子块区域,并以此构建图像子块区域之间的连通关系;2.算法通过将颜色特征度量方式与纹理特征度量相结合的方式,定义一个更加鲁棒的相似性度量公式,实现对纹理特征与颜色特征共同作用的纹理图像中纹理元素之间的高质量相似性判断;3.算法通过结合颜色特征与纹理特征的相似性度量公式,进一步改进优化的图割模型,从而最终实现准确地捕获具有外观相似特征的纹理元素。本文从外观特征综合性着手,建立一种更有效更全面的度量机制,避免一种分割算法仅对具有一种特定外观特征的纹理图像有效的弊端,综合相应特征提取纹理图像中的重复纹理元素。通过综合比对大量实验数据,证明了本文算法针对前/背景颜色相近的纹理图像中的纹理元素的提取有较大改善,并且大大提高了现有图像分割算法的时间效率。
[Abstract]:The extraction of repeated texture elements in texture image is to extract the data structure which has the same or similar characteristics (i.e. texture element) of the texture image. The purpose of repeated texture element extraction is to recognize the region with similar visual features in texture image, simplify or change the representation of the image, and make the complex texture image a simple and independent combination of single texture elements. In order to make the image easier to be understood and analyzed by computer, it also provides the necessary prerequisite for subjective reconstruction and design of texture image. The image segmentation usually refers to the following rules: some regions represent the basic structure of the target texture elements, and others represent the homogeneous region which is different from the target, that is, the texture background region. On the one hand, the extraction of texture elements can decompose the texture image into an independent unit module, and the independent texture element represents the basic composition of the texture image. Therefore, using the extracted texture elements effectively to analyze the topological structure of the original texture image is an effective research method for the composition structure of the texture image; on the other hand, the extracted texture elements can also be used for texture synthesis. It can also produce new texture images, which can lay a good foundation for the transfer of texture images, the combination of texture images and the design of texture images. In order to extract repetitive texture elements from texture images, this paper presents an algorithm for extracting repetitive texture elements from interactive texture images by analyzing and summarizing the existing methods. The algorithm can extract the repeated texture elements of the relevant color or texture feature in the texture image at the same time with a small amount of interaction provided by the user. Through a large number of experiments, we fully verify the effectiveness and practicability of the proposed algorithm. The organizational structure of this algorithm is divided into the following points: 1. The algorithm uses color clustering method to divide the source texture image into independent and disconnected sub-regions of the image, and then constructs the connectivity relationship between the sub-regions of the image blocks. The algorithm defines a more robust similarity measurement formula by combining color feature metrics with texture feature metrics. The high quality similarity judgment between texture elements in texture image is realized. By combining the similarity measurement formula between color features and texture features, the algorithm further improves the optimized graph cutting model, so that the texture elements with similar appearance features can be captured accurately. In this paper, a more effective and comprehensive measurement mechanism is established to avoid the disadvantage that a segmentation algorithm is only effective for texture images with a specific appearance feature. The repeated texture elements are extracted from the texture image by combining the corresponding features. Through comprehensive comparison of a large number of experimental data, it is proved that the proposed algorithm can improve the extraction of texture elements in texture images with similar front / background colors, and greatly improve the time efficiency of existing image segmentation algorithms.
【学位授予单位】：长沙理工大学
【学位级别】：硕士
【学位授予年份】：2016
【分类号】：TP391.41

【参考文献】