低功率影像處理電路於腸胃道內視鏡之研究

Abstract

對於無線腸胃道內視鏡系統，我們已經成功發展出兩種應用於膠囊內視鏡或是吞嚥式影像膠囊的極低功率影像壓縮處理器。在無線膠囊內視鏡系統應用中，平衡壓縮端電池壽命/效能取捨是極為重要的。取代目前最先進的影像壓縮技術，我們首先提出一套以紅綠藍三原色為基礎的影像壓縮演算法，簡稱為GICam-I 且此演算法首先藉由移除傳統影像壓縮演算法內的解碼賽克(demosaicking)技術與色彩空間轉換(color-space transform)技術來簡化傳統影像壓縮演算法的計算複雜度。另外，為了更進一步延長無線腸胃道內視鏡系統壓縮端的電 池使用壽命，我們接著發展出一套改良式、極低功率、以次取樣 (subsample)技術為基礎的影像壓縮演算法，簡稱為GICam-II。藉由次取樣技術來改良GICam-I 演算法的計算負載，根據色彩敏感度分析的結果，我們成功利用次取樣技術去降低綠色訊號與藍色訊號的記憶體需求。 除了使用極低功率壓縮技術來節省在高解析度無線腸胃道內視鏡系統的射頻傳輸功率損耗。對於無線/有線腸胃道內視鏡系統，如何有效消除惱人的脈衝雜訊與強化腸胃道影像的銳利度是必然。為了克服這些問題，低-高-中(lower-upper-middle, LUM)濾波器是最適當的候選因為它本身同時具有平滑化與銳利化的能力。在LUM 濾波器的運算中，主要運算核心為排序濾波器(rank-order filtering, ROF)計算且LUM 濾波器需要不同的順序(rank)值來完成平滑化或是銳利化的任 務。因此需要一個有彈性的ROF 硬體來供任意選擇所需要的順序值 已完成LUM 濾波器的運算程序且我們已經提出一個以可遮罩式記憶 體為基礎的排序濾波器架構。可遮罩式記憶體結構又稱為雙細胞隨機 存取記憶體(dual-cell random-access memory, DCRAM)是一個伴隨著可遮罩式暫存器與雙細胞結構的靜態隨機存取記憶體(static random-access memory, SRAM)的延伸變化。本論文是第一個使用可遮罩式記憶體來實現排序濾波器，藉由一般排序濾波器演算法驅動，以此記憶體為基礎的硬體架構具有高彈性與高規則性且同時擁有低成 本與高效能之特色。這個架構能夠應用於任意順序的尋找以及包含遞 歸(recursive)或是非遞歸(non-recursive)的排序濾波器之變形。除了針對腸胃道影像可以消除惱人的脈衝雜訊及增加其影像銳利度外，本論文所提出的排序濾波器之運算速度也可以應付即時影像處理應用。

For wireless gastrointestinal endoscope systems, we have been successfully developed two kinds of ultra-low-power image compression processors applied for capsule endoscope or swallowable imaging capsules. In applications of wireless endoscope systems, it is imperative to balance battery life/performance trade-offs. Instead of applying state-of-the-art image compression techniques, we firrst proposed an RGB-based compression algorithm, called GICam-I and it ‾rstly simpli‾ed traditional image compression algorithms by removing the demosaicking technique and the color-space transformation. In addition, to further extend the battery life in wireless gastrointestinal endoscope systems, another improved ultra-lowpower subsample-based GICam image compression processor, called GICam-II, is proposed. By using the subsample technique to improve computational loads in the GICam-I, we successfully make use of the subsample technique to reduce the memory requirements of G1, G2 and B components according to the analysis results of color sensitivity.

Except using novel ultra-low-power compression techniques to save the power dissipation of RF transmitter in high-resolution wireless gastrointestinal endoscope systems. How to efficiently eliminate annoying impulsive noise caused by a fault sensor and enhance the sharpness is necessary for gastrointestinal (GI) images in wired/wireless gastrointestinal endoscope systems. To overcome these problems, the LUM filter is the most suitable candidate because it simultaneously has the characteristics of smoothing and sharpening. In the operational procedure of LUM filter, the mainly operational core is the rank-order filtering (ROF) and the LUM filter itself needs to use different kind of rank values to accomplish the task of smoothing or sharpening. Therefore, we need a flexible ROF hardware to arbitrarily select wanted rank values into the operation procedure of LUM filter and we have proposed an architecture based on a maskable memory for rank-order filtering. The maskable memory structure, called dual-cell random-access memory (DCRAM), is an extended SRAM structure with maskable registers and dual cells. This dissertation is the first literature using maskable memory to realize ROF. Driving by the generic rank-order filtering algorithm, the memory-based architecture features high degree of flexibility and regularity while the cost is low and the performance is high. This architecture can be applied for arbitrary ranks and a variety of ROF applications, including recursive and nonrecursive algorithms. Except efficiently eliminating annoying impulsive noises and enhance sharpness for GI images, the processing speed of ROF can also meet the real-time image applications.