影像縮放內插方法及硬體化可行性之研究

Abstract

本論文研討之影像縮放內插演算法及其可行之硬體化技術乃近年來影像科技發展之趨勢所需。不論是在日常家庭生活早已佔有不可或缺地位的電視機或是辦公室常見的電腦顯示器，已在沉默中由數位薄型螢幕取代了傳統的類比陰極射線管(CRT)顯示器；而人手一部的行動電話，亦拜平板顯示技術進步所賜，其功能早已進展到可播放數位影像。然而如何在這些大小尺寸不ㄧ的顯示媒體上顯示同一解析度的影像或是在一固定解析度顯視器上顯示不同解析度圖像訊號便成為了一個重要的研究議題。 理論上，在有限頻寬信號分析的研究中，應用Sinc函數作內插是最理想的內插選擇，運用到影像縮放處理時，則是將影像看成二維的信號，分別從水平方向及垂直方向進行內插。然而因其函數特性過於複雜而使其無法在硬體化實現上得到經濟的實作方式。ㄧ般來說，對於影像以內插方式增加或降低其解析度基本常用的內插核心不外乎有最鄰近像素內插(Nearest Neighbor)、雙線性(Bi-Linear)內插及雙立方(Bi-Cubic)內插。其中又以雙立方內插核心函數之特性與Sinc最為接近。 我們選擇雙立方內插核心函數為我們硬體實作之研究，理由除了其函數特性外，我們也以大量的圖像做為分析演算法之優劣的基礎，我們所得到的數據亦支持了學理上的說法。在影像縮放內插演算法之硬體化可行性之研究上，我們以硬體實作首要考量之經濟及處理速度為出發點，設計了查表式的影像縮放硬體架構，並針對所需解晰度轉換所需之查表容量做最佳化研究，以達經濟及處理速度之目的，其中對於內插函數權值的調整是目前已公開之文獻中未曾探討的領域。

Effective procedures for image scaling which may be suitable for hardware implementation are investigated in this thesis. Since the display technology is changing rapidly from analog to digital and CRT to flat panel display, there are often various resolution, in display specifications in our daily life, such as these for huge size TV or hand-held devices, such as cell phone. How to display an image with satisfactory in different display devices with different resolutions become big issue. In this thesis, we study the image scaling technique, starting from on the interpolation kernel "Sinc" that signal processing theory regarded as "ideal," then other common methods like "Nearest Neighbor," "Bi-Linear" and "Bi-Cubic." Subsequently, the "Bi-Cubic" interpolation kernel for possible hardware implementation, not only because it is closer to "ideal" mathematically, but also supported by experimental data. We implement the hardware architecture by table-look-up approach and also discuss different ways of optimizing the table size.