透過您的圖書館登入
IP:18.117.153.38
  • 學位論文

以離散餘弦轉換為基礎之迭代反投影影像超解析研究

DCT-based Iterative Back-projection Algorithm for Image Super-Resolution

指導教授 : 李錫捷

摘要


目前在數位影像超解析度的研究中主要可以分成兩類,分別是以單一影像為基礎的解析度提升,以及使用多張影像作為參考的解析度提升,本研究的方法是採取單一影像作為解析度提升的方式在進行的。在本論文進行解析度提升的研究過程中,我們分別導入了數位影像處理中應用於頻率域以及空間域的處理技術,像是在進行影像初步的解析度放大過程中,我們使用了離散餘弦轉換所特有的低頻區能量緊密特性作為影像解析度提升上的再取樣基礎。而在得到了解析度經過初步放大後的初始高解析度影像後,為了針對影像在經過頻率域處理後可能會產生的混疊效應以及使用低頻資訊作放大後所特有的高頻遺失效應,我們另外使用空間域上的技術來對影像作混疊修補及高頻補償的處理。如此一來,影像在經過頻率域的低頻放大以及空間域的高頻補償後,便能夠得到一幅高品質的高解析度影像作為輸出。本論文所提出的演算法不論是在肉眼的主觀認知上,或是鋒值訊噪比的評比數值上,均比常見的使用多項式插補作解析度放大的演算法的結果還要來得有更好的視覺效果以及顯著的評比數據的提升。

並列摘要


There are two main approach of the image super-resolution research in digital image processing. one is based on the single image, and the other is used by multiple images which inter-reference each other. In the two different approaches, this paper is belong to the first one. In our research approach, we use the frequency and spatial domain methodologies of digital image processing respectively. The first step to get the initial high resolution image is through the energy compaction characteristics of Discrete Cosine Transform. Though we may get some extra side effects such like aliasing and high-frequency losing in those images by using the frequency domain transform technic, hence we then use the spatial domain processing technics for dealing with those side effects and at the same time doing high-frequency feedbacks synchronically. After this ways of the second step processing that image passes through the low-frequency resampling and high-frequency feedbacks, we shall and will get a high quality and high resolution image as our approach output. The super- resolution algorithm we proposed in this paper can clearly find not only the subjective viewing quality, but also the objective value of PSNR results a satisfied improvement. Finally, the approach of ours exactly have a good improvement from the existing interpolation algorithm like Bicubic.

參考文獻


[37] 張宸瑋, “高效率影像視訊超解析度方法”, 國立台灣大學, 資訊工程研究所碩士論文,
[23] 蕭傳楷, “藉由大量樣本強化影像解析度”, 國立台灣大學, 資訊工程研究所碩士論文,
[50] 張書樊,“運用高頻資訊補償之超解析度演算法研究”,國立中央大學,
[56] 解昇維,“VC-1解碼器於PAC DSP3.0之實作研究”, 國立成功大學, 電腦與通信工程研究所
[51] 黃一民,“超解析技術應用於影像序列放大之研究”,國立成功大學

延伸閱讀