數位攝影機之數位式影像穩定系統及視訊編碼器之整合

Abstract

近幾年來，由於科技的發展，許多的數位商品紛紛問世，並已經融入了我們生活之中。由於電子式的影像擷取裝置被廣泛的應用在各個層面，因此數位影像的處理技術愈趨重要，譬如影像壓縮技術，影像穩定系統等等。其中影像穩定系統，能夠提供給使用者有更清晰、更高品質的畫面享受。 數位式影像穩定系統的架構大致上可以分為兩部分：總體移動估計以及總體移動補償。總體移動估計主要是當攝影機震動時，去量測影像框架之間的總體移動；而總體移動補償則是根據計算出的總體移動，在感光元件上取出穩定而且連續的影像。由於總體移動估計需要耗費大量的計算，然而，在影像編碼器或是解碼器裡面皆含有相關的資訊，例如以區塊作為基礎的移動量估計等，如果可以利用這些資訊跟影像穩定系統去做整合，便可以使得計算更有效率。 本論文研究的重點，將針對數位式影像穩定系統進行研究，希望以影像穩定系統的演算法作為基礎，與視訊編碼技術MPEG-4做一即時運算的結合。在此論文中，我們提出三種整合的架構，皆是利用MPEG-4編碼器或是解碼器所得到的運動資訊，去計算出影像系統中總體移動估計的部分，以避免重複去做移動估計的運算，並對整合所得的結果去做各個層面的分析與比較。

The movement of a video camera caused by hand jiggle introduces jerky image motion that is often annoying to human eyes. The problem can be solved by using an image stabilization system to compensate for the hand motion. As opposed to optical image stabilization which uses gyro-sensors to detect hand motion and shifts a corrective lens inside the lens system (or alternatively shifts the image sensor while keeping the lens fixed) , digital image stabilization detects the induced image motion based on the video data and shifts the image display window accordingly to compensate for the hand motion. For many digital video camera designs, digital image stabilization is a cost effective solution and is the subject of discussion in this thesis. A digital image stabilization system that can reduce inter-frame hand-shaking movement and smooth the intentional camera movement in a video sequence is proposed. The video sequence is processed to estimate the local motion vectors by block-based motion estimation and to obtain the global motion vector by clustering. The global motion vector between frames is real-time low-pass filtered to remove high frequency motion component caused by hand-shaking. A simplified stabilization algorithm facilitates the implementation on BF561, a DSP processor of Analog Device. For a digital video camera, digital image stabilization system and video coder are two important components, which both require motion information of the captured image sequence to perform their respective tasks. Since motion estimation is a computationally intensive operation, we propose three schemes for integrating the electronic image stabilization system with the video coder. The technical issues involved in the integration are discussed, and the simulation results are shown to illustrate the effectiveness of the integration schemes.