不規則圖像轉換及基於分散誤差之影片縫切割技術;Irregular Image Transformation and Balanced Error Distribution for Video Seam Carving

NCUIR > College of Electrical Engineering & Computer Science > Graduate Institute of Computer Science and Information Engineering > Electronic Thesis & Dissertation > Item 987654321/74702

Please use this identifier to cite or link to this item: http://ir.lib.ncu.edu.tw/handle/987654321/74702

Title:	不規則圖像轉換及基於分散誤差之影片縫切割技術;Irregular Image Transformation and Balanced Error Distribution for Video Seam Carving
Authors:	張閎翰;Chang, Hon-Hang
Contributors:	資訊工程學系
Keywords:	重新定位技術;縫切割;能量圖;顯著圖;Retargeting;Seam Carving;Energy Map;Saliency Map
Date:	2017-07-31
Issue Date:	2017-10-27 14:36:44 (UTC+8)
Publisher:	國立中央大學
Abstract:	近年來，越來越多的圖像和視頻重新定位技術被提出來，特別是基於使用 seam carving、warping 或結合兩者之技術，成功地在多媒體領域上豐富我們的日常生活。雖然這些技術之已發展多年，但這些影像重新定位技術仍僅適用在於將不同尺寸之矩形與矩形間作圖像轉換，其不能輕易地重新將矩形圖片轉換成另一個圓形、多邊形及其他形狀。因此，本論文著提出一適用於不規則影像至轉換之圖像轉換系統命名為 CMAIR，此變形轉換系統能夠將矩形圖像重新定位為不同形狀的圖像，並能突顯主要興趣的顯著物件。在我們的圖像轉換系統中，我們另外提出了一種獨特的不規則插值方法來產生四個可能的目標圖像，並可經由我們們所提出之影像評估機制，再考慮圖像顯著性下來確定最佳候選圖像作為最終輸出。經由實驗結果發現，我們的實驗結果不僅可以將源圖像放置到不同的目標形狀的蒙版中，而且還可以盡可能保留和突出顯著的物件，幫助我們能夠瀏覽更清晰之影像。本文所提出的算法其重點在於，能夠在視圖片內容中考量不規則形狀之影像內容物之結構，並將影像其內容物合理的放置於所給定之框架中。在影片的重新定位部分，本研究另外提出一個影片雕刻技術，目的在於加強相連影像於時間域中保持其影像內容物之連續性，並避免影片內容物之顯著主體被破壞，加強視覺無違和之視頻產出。同時本研究特別針對其複雜影像內容物的視頻 (包含存在透視效果之消失縣、存在大量線性結構之影像內容）來證明此演算法的實用性。由本論文所提供之視訊結果中可以明顯的發現，該圖片或影像在影像縮小之過程中能有效地保持及維護其圖像或影格內容之線性結構，並在影片中的各影格維持其視覺之連續性；由圖像之結果可以證明本研究所提出的不規則縮放技術，讓矩形圖片之影像內容能夠有效的向內填入於不規則之外框，以不規則之影像外型來呈現。為了驗證本研究方法之可行性，http://video.minelab.tw/ DETS/EDSeamCarving/ 及 http://video.minelab.tw/DETS/VRDSF/ 介紹本論文視訊與圖像在尺寸調整後其影像內容物之比較來驗證本論文方法的貢獻。;In recent years, more and more image retargeting techniques have been proposed to facilitate our daily life, in particular those based on the use of seam carving, warping or the combination of them. Although these techniques are rather sophisticated, they only work for a specific pattern during retargeting. For example, these techniques can only retarget the source picture into the same shape of a square, but cannot be easily reshaped into a circular, a polygon or other shapes. This paper focuses on creating a graphics editing system, named CMAIR (Content and Mask-Aware Image Retargeting), for image retargeting, which can retarget the source images into diff erent shapes of image to highlight the salient objects of primary interest. CMAIR eff ectively supports removal of unimportant pixels, and frames as many surrounding objects inside the provided mask as possible. In this paper, we propose a unique irregular interpolation method to produce four possible target images, and an evaluation mechanism to decide the best candidate image as the final output with the consideration of image saliency. The results show that not only the source image can be placed into different targeted shapes of mask, but also the salient objects are retained and highlighted as much as possible. As a result, the salient objects become more clear in our eyes. Besides, a video retargeting method named BED is proposed in this study, and the proposed algorithm in this thesis focuses on maintaining the structure of straight lines and irregular shape of objects without deforming complex image contents, which may be altered in traditional seam carving of complex image or video. In addition, the proposed mechanism maintains visual continuity such that the resulting videowill not look shaky due to sudden changes in the background. Practical applicability of the proposed method was tested using both regular videos and special videos which contain vanishing lines (i.e., perspective eff ects). Experimental results demonstrate that the proposed CMAIR can convert the rectangular image to another irregular shape of image, and all the salient objects of the targeted images are preserved. In our video results using BED, the BED approach not only resizes the video with the retention of important information, but also maintains the structural properties of objects in varios kinds of videos. The demonstration website at http://video.minelab.tw/DETS/ EDSeamCarving/ and http://video.minelab.tw/DETS/VRDSF/ provides comparison results that illustrate the contribution of our method.
Appears in Collections:	[Graduate Institute of Computer Science and Information Engineering] Electronic Thesis & Dissertation

Files in This Item:

File	Description	Size	Format
index.html		0Kb	HTML	164	View/Open

社群 sharing

Loading...