請用此 Handle URI 來引用此文件:
http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/65964
標題: | 應用於可攜式裝置之即時影片切割技術與架構設計 Algorithm and Architecture Design of Real-time Video Segmentation in Mobile Devices |
作者: | Chieh-Chi Kao 高介其 |
指導教授: | 簡韶逸(Shao-Yi Chien) |
關鍵字: | 視訊切割,引導濾波器,硬體架構, video segmentation,guided filter,hardware architecture, |
出版年 : | 2012 |
學位: | 碩士 |
摘要: | 隨著通訊技術的進步與多媒體技術的發展,使用者可以在任何時間、地點透過可攜式裝置來連結到網路。各式各樣的應用相繼出現,如上傳照片和影片並和朋友分享、利用圖片來做搜尋、撥打網路視訊電話等等。物體切割技術在前述應用的前處理中,扮演了一個相當重要的角色。然而,受限於行動裝置的運算能力,並非每一個先前所提出的方法都能夠直接套用到可攜式裝置上。在本篇論文中,我們提出了應用於可攜式裝置之即時影片切割技術與架構設計。
在影像和視訊處理的領域中,物體切割是一個已經發展已久的研究題目,許多先前所提出的演算法已經可以達到很好的效果。然而,大部分的方法都需要使用者提供輸入有關於切割目標的資訊來輔助運算。在本篇論文中,我們提出了一個應用於可攜式裝置之即時影片切割技術,其包含了一個非監督式顯著物體偵測及切割的技術,可以在不需要使用者提供資訊的前提下切割出目標物體。實驗結果指出,由顯著特徵圖求得的顯著顏色模型能夠取代使用者輸入資訊來自動達成物體切割。本系統同時也提供了改善機制讓使用者能夠修正切割不夠正確的部分,藉由少許的使用者輔助將可大大提升切割的品質。我們所提出的顯著顏色模型不僅能夠應用到Min-cut演算法,同時也可以延伸到其他切割演算法,如matting或非參數模型。 因為受限於有限的運算能力,要直接在可攜式裝置上做高解析度的視訊切割是相當困難的。我們利用在較低的解析度上來做即時視訊切割,再將切割出來的物體罩放大至較高的解析度上。而我們選擇了引導濾波器(guided filter)來作為放大物體罩的方法,但是要增加更多的工作量在可攜式裝置上的中央處理器是不可行的。因此,我們提出了可以嵌入可攜式裝置的引導濾波器之硬體架構,可以協助達成高解析度即時視訊物體切割。就我們所知,本論文也是第一個對於引導濾波器的特殊應用積體電路設計實作。我們應用台灣積體電路公司的90奈米製程,本設計可運行在100MHz,能夠處理每秒30幀的FULL-HD(1920x1080)視訊,總邏輯閘數為92,895,總記憶體數量為3,206B。此外,和過往濾波器的特殊應用積體電路設計實作來比較,我們所提出的架構硬體效率也是最佳的。 With the development of communication technology and multimedia applications, users can use their mobile devices to access the Internet at anytime and anywhere. Various kinds of applications emerge like uploading photos and videos and sharing them with friends, searching things by photos, video calling through the Internet, etc. For applications mentioned above, object segmentation plays an important role as a preprocessing stage. However, due to the limited computation capability in mobile devices, not every segmentation approach proposed in the previous works can be adopted on mobile devices directly. In this thesis, the algorithms and architecture designs of real-time video segmentation for mobile devices are presented. Image and video segmentation is a well-developed topic in the image/video processing, and a number of previous works have been proposed with high performance. However, most previous works need user-assistance to provide the prior information of the target object in the segmentation. For the algorithm of real-time video segmentation for mobile devices, an unsupervised scheme combining the salient object detection and segmentation method is proposed in this thesis, which can segment the target object without any prior information from users. The experimental results show that the proposed salient color model derived with salient features can provide prior information with high confidence to generate precise segmentation automatically. The system also provides an effortless way to let user refine the segmentation result, which can greatly improve the performance of segmentation. The proposed color model of salient objects can not only be applied with Min-Cut algorithm, but also extended to more segmentation algorithms, like matting or non-parametric models. The limited computation capability in mobile devices makes it difficult to achieve real-time segmentation for high-resolution videos. We overcome this limitation by segmenting the object at QVGA and then up-scaling the segmentation result to higher resolution. The guided image filter is adopted as the up-scaling approach; however, it is infeasible to further allocate the workload to the CPU of the mobile platform. Therefore, the hardware architecture of guided image filter is proposed and can be embedded in mobile devices to achieve real-time HD video segmentation. To the best of our knowledge, this work is also the first ASIC design for guided image filter. With TSMC 90nm cell library, the design can operate at 100MHz and support for FULL HD (1920x1080) 30 fps with 92.9K gate counts and 3.2 KB on-chip memory. Moreover, for the hardware efficiency, our architecture is also the best comparing to other previous works with bilateral filter. |
URI: | http://tdr.lib.ntu.edu.tw/jspui/handle/123456789/65964 |
全文授權: | 有償授權 |
顯示於系所單位: | 電子工程學研究所 |
文件中的檔案:
檔案 | 大小 | 格式 | |
---|---|---|---|
ntu-101-1.pdf 目前未授權公開取用 | 8.52 MB | Adobe PDF |
系統中的文件,除了特別指名其著作權條款之外,均受到著作權保護,並且保留所有的權利。