陳銘憲Chen, Ming-Syan臺灣大學:電機工程學研究所帥宏翰Shuai, Hong-HanHong-HanShuai2010-07-012018-07-062010-07-012018-07-062009U0001-0508200915502500http://ntur.lib.ntu.edu.tw//handle/246246/188131當網路的傳輸頻寬不足，可能造成使用者接收到的影片解析度十分不佳，雖然行動裝置螢幕解析度越來越高，看到的串流影片品質卻無法跟著提升，也大大地降低了使用者經由手機觀看影片的意願。然而，在影片傳輸時，不管是傳送端還是接收端卻有閒置的CPU資源，因此我們希望將行動裝置計算量與影片品質做有效的交換，而能夠解決影片解析度不足的問題。　影片傳輸之主要瓶頸在於傳送頻寬不足，且我們必須考量行動裝置的電池耗能問題，以及能否及時計算出更高畫質的畫面。因此我們提出了一個新的編碼架構，我們限制傳送的影像解析度，在傳送端預先計算需要提高畫質所需資訊，再將得到的資訊跟著影片一起傳送到接收端。接收端在解碼完影片之後，只需將畫面與發送端的資訊結合在一起做一些簡單的計算，即可得到更好畫質的畫面。雖然傳統的內插演算法運算速度非常快，但它們通常伴隨著嚴重的缺點，包括格狀效應與影像的模糊等等。我們在這篇論文所提出的方法不但維持了低運算量, 並顯著改善了傳統演算法所存在的缺點。 研究結果顯示我們所提出的方法在邊緣清晰度與影像相似度上都優於傳統演算法，而在計算環境困難的嵌入式系統中也有不錯的表現。At the present situation, since the bandwidth of wireless network is limited, the video quality on mobile device is low. Although the display resolution of mobile device is higher and higher these days, the quality of video stream is not improved. The low video quality greatly reduces the willingness of people to use video services. In view of the fact that the computational resources are idle at the server end and the client end during the transmitting time, in this thesis, we make a trade-off between computational power of mobile devices and video quality to resolve the problem of video resolution insufficiency. The main bottleneck of video quality is the low bandwidth of transmission. Moreover, the battery power of mobile devices and the capacity of real-time upsampling are two chief considerations in the mobile environment. To enhance the video quality, we proposed a novel coding architecture. Initially, we determine the parameters that are needed for upsampling. Then, the parameters are sent along with video stream. When users receive transcoded video, they extract the frames with parameters and combine video frames with additional data together to display higher resolution. It is noticed that although traditional methods can provide high-speed upsampling, they process a lot of defects such as blocking effects and blur. The experimental results show that our approach can achieve higher visual quality than traditional methods. In addition, the proposed video upsampling algorithm is efficient and can be applied on mobile devices.口試委員會審定書 #文摘要 iiBSTRACT iiiONTENTS ivIST OF FIGURES viIST OF TABLES viiihapter 1 Inroductions 1hapter 2 Preliminaries 4.1 An Introduction of Image Upsampling 4.2 Non-Adaptive Image Upsampling 6.2.1 Nearest Neighbor Upsampling 6.2.2 Bilinear Upsampling 7.2.3 Bicubic Upsampling 8.3 Adaptive Image Upsampling 10.3.1 Image Interpolation by Pixel Level Data-Dependent Triangulation 11.3.2 New Edge-Directed Interpolation 13.3.3 Upsampling via Imposed Edges Statistics 15.3.4 Image Super-Resolution using Gradient Profile Prior 17.4 An Introduction of Edge Processing 18.4.1 First-order Approaches to Edge Detection 19.4.2 Second-order Approaches to Edge Detection 20.4.3 Canny Edge Detection 22hapter 3 System Architecture 25.1 Video Upsampling Server 25.1.1 Overviews 25.1.2 Modified Canny Edge Detection 27.1.3 Using Constrains of Temporal Continuity 30.1.4 Line Model Construction 34.1.5 Continuity Detection by Eight-neighbor Grouping 36.1.6 Look-Up Table Construction 37.2 Mobile Client 38.3 Image Reconstruction at Client End 39.4 Other Methods under Our System Architecture 41hapter 4 Experimental Evaluation 43.1 Image Restoration Analysis 43.2 Additional Data to Original Data Ratio Analysis 49.3 Efficiency Analysis of Video Upsampling 50hapter 5 Conclusions 52EFERENCE 531783068 bytesapplication/pdfen-US影像放大邊緣偵測嵌入式系統視頻轉換資料壓縮edge detectionvideo upsamplingmobile devicesvideo transcodingdata compressionthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/188131/1/ntu-98-R96921037-1.pdf