【摘要】：紋理圖像中重復紋理元素的提取是指將組成紋理圖像的具有相同或相似特性的重復單元(即紋理元素)的數(shù)據(jù)結(jié)構(gòu)提取出來。重復紋理元素提取的目的是識別紋理圖像中具有相似視覺特征的區(qū)域,將圖像的表現(xiàn)形式加以簡化或者改變,使復雜的紋理圖像歸宿為簡單獨立的單個紋理元素的重復組合,以便圖像更容易被計算機理解和分析,同時為主觀重組設計紋理圖像提供必要的前提條件。其中圖像的分割通常參照如下規(guī)則:部分區(qū)域表示目標紋理元素的基本結(jié)構(gòu);其他區(qū)域則表示與目標存在一定差異的同質(zhì)區(qū)域,即紋理背景區(qū)域。一方面,紋理元素的提取能夠?qū)⒓y理圖像分解成獨立的單元模塊,而獨立的紋理元素代表了紋理圖像的基本構(gòu)成,因此有效利用已抽取的紋理元素來分析原紋理圖像的拓撲結(jié)構(gòu),是對紋理圖像組成結(jié)構(gòu)的一種有效研究方法;另一方面,提取出的紋理元素還可以用于紋理的合成,亦可以由此產(chǎn)生新的紋理圖像,為紋理圖像的轉(zhuǎn)移、紋理圖像的組合、紋理圖像的設計打好基礎,奠定有力的前提條件。為了實現(xiàn)紋理圖像中重復紋理元素的提取,本文通過對現(xiàn)有重復元素提取方法的分析總結(jié)與研究,提出了一種交互式紋理圖像中重復紋理元素提取算法,該算法能夠在用戶提供少量交互的情況下,較好地實現(xiàn)對紋理圖像中具有相關性的顏色或紋理特征的重復紋理元素的同時提取。我們通過大量實驗,充分驗證了本文算法的有效性與實用性。本文算法的組織結(jié)構(gòu)主要分為以下幾點:1.算法采用顏色聚類方法將源紋理圖像分割成獨立且不連通的圖像子塊區(qū)域,并以此構(gòu)建圖像子塊區(qū)域之間的連通關系;2.算法通過將顏色特征度量方式與紋理特征度量相結(jié)合的方式,定義一個更加魯棒的相似性度量公式,實現(xiàn)對紋理特征與顏色特征共同作用的紋理圖像中紋理元素之間的高質(zhì)量相似性判斷;3.算法通過結(jié)合顏色特征與紋理特征的相似性度量公式,進一步改進優(yōu)化的圖割模型,從而最終實現(xiàn)準確地捕獲具有外觀相似特征的紋理元素。本文從外觀特征綜合性著手,建立一種更有效更全面的度量機制,避免一種分割算法僅對具有一種特定外觀特征的紋理圖像有效的弊端,綜合相應特征提取紋理圖像中的重復紋理元素。通過綜合比對大量實驗數(shù)據(jù),證明了本文算法針對前/背景顏色相近的紋理圖像中的紋理元素的提取有較大改善,并且大大提高了現(xiàn)有圖像分割算法的時間效率。
[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.
【學位授予單位】：長沙理工大學
【學位級別】：碩士
【學位授予年份】：2016
【分類號】：TP391.41

【參考文獻】

相關期刊論文 前10條

1 伊聰聰;吳斌;張紅英;;一種改進的Grabcut圖像分割方法[J];小型微型計算機系統(tǒng);2014年05期

2 張會;劉士榮;張波濤;;一種基于顏色和紋理的顯著區(qū)域提取算法[J];華中科技大學學報(自然科學版);2013年S1期

3 戴海濤;唐作其;張正平;;基于Mean Shift聚類的圖像分割研究[J];通信技術;2011年12期

4 張立和;朱莉莉;米曉莉;;結(jié)合Gabor紋理特征的局域化多通道水平集分割方法[J];電子學報;2011年07期

5 蔣建國;郭艷蓉;郝世杰;詹曙;李鴻;Ian Ross;;貝葉斯框架下的非參數(shù)估計Graph Cuts分割算法[J];中國圖象圖形學報;2011年06期

6 趙在新;成禮智;;結(jié)合半局部信息與結(jié)構(gòu)張量的無監(jiān)督紋理圖像分割[J];中國圖象圖形學報;2011年04期

7 周愛武;于亞飛;;K-Means聚類算法的研究[J];計算機技術與發(fā)展;2011年02期

8 肖春霞;初雨;張青;;高斯混合函數(shù)區(qū)域匹配引導的Level Set紋理圖像分割[J];計算機學報;2010年07期

9 廖建勇;郭斯羽;黃梓效;;基于Mean Shift聚類的最大熵圖像分割方法[J];計算機仿真;2009年09期

10 韓曉紅;胡_g;;K-means聚類算法的研究[J];太原理工大學學報;2009年03期

相關博士學位論文 前2條

1 桂彥;可視媒體編輯與重用關鍵技術研究[D];上海交通大學;2012年

2 楊小兵;聚類分析中若干關鍵技術的研究[D];浙江大學;2005年

，

本文編號：2198607