Please use this identifier to cite or link to this item: http://hdl.handle.net/11455/51554
標題: An Effective Image Inpainting Method and Its Application to Remove Dust Spots and Hot Pixels from Digital -SLR Images
一個有效的影像修補演算法及其在消除DSLR感光元件染塵與熱燥點去除之應用
作者: 吳俊霖
關鍵字: 資訊工程--硬體工程
Image inpainting
應用研究
Edge structure reconstruction
Sensor dust
Hot pixels
Vector filter
影像修補
邊緣結構重建
感光元件染塵
熱燥點
向量濾波器
摘要: Image inpainting is an iterative method for repairing damaged pictures or removing unnecessaryelements from pictures. A new inpainting algorithm is proposed for removing large objects fromdigital images in this study. In the first step, edges are detected around the missing area, based on thecontours of the segments that result from the JSEG segmentation. In the second step, the edges ismodeled as locally circular shapes, and the image structure is then reconstructed by coupling edgesthat are strongly related to each other, thereby reconnecting the pairs of edges that were part of thesame object contour. Finally, a fast exemplar-based inpainting approach is used to fill the missingarea, the image skeleton is then preserved.The digital cameras are becoming more and more popular nowadays. The digital SLR not onlyallows photographers to use inter-changeable lenses to get different ranges of zoom and depth of field,but also gives the users more controls, it helps to take a better picture. Recently the number of digitalSLR users is rising steadily as the equipments drop in price. However, unlike film cameras, currentdigital SLR suffers from a frustrating weakness: sensor dust and hot pixels. In this study, we developan automatic spots removal algorithm based on the inpainting technique. Several examples on realimages are given to demonstrate the effectiveness of the proposed method.The images with long exposure time usually are affected by serious noise. This kind of noise iscalled hot pixels, which comes from the individual sensors on the CCD/CMOS with higher thannormal rates of charge leakage. A two-stage noise reduction algorithm for hot pixels is proposed inthis study. In the first stage, a hybrid vector filter for color image is developed to remove the impulsenoise and Gaussian noise. The goal of the second stage is to remove the large hot pixels in the R andB channels. We first propose a noise detection algorithm to identify the hot pixels, it uses the Sobelfilter and the morphology operations. The proposed inpainting algorithm is then used to fill holes ofhot pixels in images, it achieves accurate propagation of linear structures.
影像修補是專門用來修補被破壞影像的一項技術,其應用的範圍包括舊照片上裂痕的修補、移除不想要的物件以及附加在圖片上的文字與戳印等。在本研究中,我們將要發展一個有效的影像修補演算法,其能修補較大面積的影像破壞,且能正確保留影像中的線性結構。首先我們先以影像分割法將影像中的連續邊緣資訊找出,接著以圓弧來估計讀進來的邊界,並重建出破損部分的影像骨幹結構。最後利用基於範例影像修補演算法來完成影像修補,其藉由給予落在線性結構的patch 較大的優先權,使其優先填補,而能有效確保線性結構的完整。隨著單眼相機在消費市場上越來越普遍,使用人數逐年增加,這使得數位單眼相機容易出現感光元件染塵,與長曝光攝影所產生的熱燥點的問題變得不容忽視。為了解決CCD 染塵問題,最近有一些相機廠商發明出「感光元件自我清理架構」,利用壓電元件產生超音波震動,讓粉塵脫落,不過對於黏度較高的粉塵,則效果不彰。本研究針對這個問題提出一個能夠自動偵測粉塵的演算法,同時利用所發展的影像修補的技術,將影像上令人困擾的粉塵去除,並且維持影像中的材質與線性結構的完整。在另一方面,由於數位相機在長時間曝光攝影時,會使得CCD 和相機電子元件產生過熱現象,而會以隨機斑點的形式出現電子干擾或雜訊,此種雜訊我們通常稱為熱燥點。我們發現通常此種雜訊依其特性可以分為兩種,一種為近似脈衝雜訊的雜點,通常為1 pixel 大小。另一種雜訊通常散佈在紅藍頻道,大小較大,可以達數個pixels。為了有效去除第一種雜訊,本研究提出一混合式向量濾波器,其結合了以脈衝雜訊偵測為基礎的向量中值濾波器與適應權重向量濾波器,此濾波器能有效復原受雜訊干擾的數位彩色影像,而達成雜訊的移除和邊緣保存。第二種熱燥點比一般雜訊要大得多,而我們也將發展一個能夠自動偵測熱燥點的演算法,同時利用所提影像修補的技術,將影像上令人困擾的熱燥點去除,並且維持影像中的材質與線性結構的完整。
URI: http://hdl.handle.net/11455/51554
其他識別: NSC97-2221-E005-081
文章連結: http://grbsearch.stpi.narl.org.tw/GRB/result.jsp?id=1686123&plan_no=NSC97-2221-E005-081&plan_year=97&projkey=PB9709-2486&target=plan&highStr=*&check=0&pnchDesc=%E4%B8%80%E5%80%8B%E6%9C%89%E6%95%88%E7%9A%84%E5%BD%B1%E5%83%8F%E4%BF%AE%E8%A3%9C%E6%BC%94%E7%AE%97%E6%B3%95%E5%8F%8A%E5%85%B6%E5%9C%A8%E6%B6%88%E9%99%A4DSLR%E6%84%9F%E5%85%89%E5%85%83%E4%BB%B6%E6%9F%93%E5%A1%B5%E8%88%87%E7%86%B1%E7%87%A5%E9%BB%9E%E5%8E%BB%E9%99%A4%E4%B9%8B%E6%87%89%E7%94%A8
Appears in Collections:資訊科學與工程學系所

文件中的檔案:

取得全文請前往華藝線上圖書館



Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.