Snap2Read/Comp2Watch: 增進手持裝置平台上的多媒體瀏覽體驗

Abstract

手持裝置服務的興起（例如：電子書、影片串流）以及手持裝置的盛行率不但顯露出在手機上閱讀�觀看的需求，也顯示了在手機服務開發上被看好的商業機會。然而，在不同的服務上都有其所面臨的挑戰。以在手機上閱讀電子書來說，不相容的閱讀器、不一致的電子書格式，甚至是有限的螢幕大小，都會導致在手持裝置上閱讀文件的不便。同時，要在手機上閱讀那些沒有數位副本的紙本雜誌是很困難的。因此，我們提出一個系統「Snap2Read」可以自動地切割手機上拍下的文件圖片（從紙本雜誌）並將它們轉成可閱讀的片段（patches）（像是文字、標題、圖片等等）然後將他們縮放、裁切成適合的大小以便讓使用者可以透過手機上的點擊就能夠很簡單地瀏覽數位化的雜誌頁面。另一個在手機上熱門的活動則是觀看影片，但是極小的手機螢幕尺寸、有限的頻寬，以及零碎的使用時間仍然阻礙了使用者的體驗：它們要不就是中斷了使用者的觀看過程，抑或是讓使用者無法一次瀏覽多樣的內容。而傳統的影片摘要技術並不能應用在有限的螢幕上，因此我們提出了「Comp2Watch」這個系統，發音近似於「come to watch」。這個名字也有著「將影片畫面組合成美術拼貼」以及「壓縮觀看時間」的意義。它考慮了感興趣的區域（ROI）因素讓使用者能夠快速地瞥過影片，並且我們也修改了價值函數（cost function）用來整合不同長寬比的樣板，我們也處理了因為有限空間而導致的單調排版（monotone layout）問題。實驗結果顯示使用者可以在沒有遺失太多週邊資訊的情況下獲得更清楚的畫面主體。

The rise of mobile services (e.g., electronic book, video streaming) and the prevalence of mobile devices reveal the needs for mobile reading/watching and the booming business opportunities in mobile service developments. However, there are different challenges among those services. For reading books on mobile devices, incompatible e-book readers, non-uniformed e-book formats, or even limited screen size causes the inconvenience of reading documents on handheld devices. Meanwhile, it is difficult to read physical magazines that do not have the corresponding digital copies. Therefore, we propose a system, Snap2Read, that can automatically segment the captured document images (i.e., from the physical magazines) in mobile phones into readable patches (e.g. text, title, image), and then scale them into suitable size so that users can easily browse the digitalized magazine pages via the mobile phone with simple clicks. Another popular activity on mobile is watching videos, but the small mobile screen size, low bandwidth, and fragmented watching time also hinder the user experiences: they either interrupt the watching process or limit users to browse many contents at the same time. Traditional video summarization techniques are suffering the small screen issue. Therefore, we propose a system, Comp2Watch, which is pronounced like “come to watch”. It implies the meaning of “composing the frames into a collage” and “compressing the watching time”. It puts ROI factors into consideration in order to help users take a quick glance at videos. Also, we modify the cost function to incorporate the templates with variable aspect ratios. We also address the monotone layout problem caused by the limited space. The experimental results show that users can obtain clearer subject without losing many contexts.