應用小波轉換作紋理影像之瑕疵檢測及合成; Applications of Wavelet Transforms on Textured Images:Defect Inspection and Synthesis

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/8722

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/8722

Title:	應用小波轉換作紋理影像之瑕疵檢測及合成;Applications of Wavelet Transforms on Textured Images:Defect Inspection and Synthesis
Authors:	郭正德;antony kuo
Contributors:	資訊工程研究所
Keywords:	瑕疵檢測;紋理;小波轉換;合成;Synthesis;Defect Inspection;texture;wavelet transform
Date:	2003-06-19
Issue Date:	2009-09-22 11:33:35 (UTC+8)
Publisher:	國立中央大學圖書館
Abstract:	隨著電子時代來臨，各種數位影像工具，如數位相機、數位攝影機、掃描器....等，功能及品質不斷提高，價格也日趨便宜，多媒體影像資料因此大量的增加，如何妥善儲存及管理這些影像資料，變得十分的重要，目前在影像資料特徵描述上，大都仍以顏色(colors)、形狀(shapes)、紋理(textures)來表示，因此有關紋理的分析研究便顯得非常重要。本論文主要目的在利用小波轉換(Wavelet transformation)的理論，研究紋理瑕疵檢測及紋理合成(synthesis)。在紋理瑕疵檢測方面，利用紋理正常部份與瑕疵在小波係數的分佈範圍不同，加以分離出來，在先前的紋理瑕疵檢測方法裏，大都必須訓練紋理正常部份，為了解決這個問題，本論文提出一個自動訓練的方法，藉由瑕疵與正常部份在影像上的一些特性，自動決定訓練區塊的方法，可以使影像在輸入的時候，重新取樣訓練，避免因為外在條件因素的變化，而使檢測發生錯誤。在紋理合成方面，利用小波轉換在影像的特性，在不同頻率區塊具有影像不同的資訊，抽取出紋理邊緣特徵，結合水平及垂直邊緣資訊，找出紋理的基本區塊，利用找出的基本區塊合成出原始影像，並利用本論文提出的瑕疵檢測方法驗證合成的效果。 Due to the emerging of computer technologies, the functions and quality of imaging devices, such as digital camera, digital camcorder, and scanner, have been continuously improved. Moreover, the cost of these devices is also rapidly reduced. The content conveyed by multimedia is thus more splendid and richer. The proper management of the image data is thereby more and more important. The features that describe image data are mainly represented by using color, shape, and texture. In this thesis, we will elaborate on the analysis of texture and its application in image analysis. The main purpose of this dissertation is to adopt the concept of wavelet transform and apply it to defect detection and texture synthesis in texture images. In texture defect detection, the defects can be discriminated according to the distribution ranges of wavelet coefficients between the normal and defective parts of texture images. In traditional texture defect detection methods, the normal parts of texture images have to be trained in advance. In this thesis, we propose a novel method to automatically determine the training regions based on the characteristics exhibited by normal and defective texture images. In this way, the detection error can be reduced because of the avoiding of environmental changes. In texture synthesis, texture edge features can be extracted according to the characteristics of wavelet transformation, that is different frequency bands will exhibit different information. By combining horizontal and vertical edge information, the basic blocks of textures can be built. Original images can be synthesized by the extracted basic blocks. Moreover, we utilize the proposed texture defect detection method to verify the synthesis results.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Size	Format

社群 sharing

Loading...