網路攝影機應用已經逐漸被使用於社交行為。順應趨勢，此應用更進一步廣泛的加入在桌上型電腦、筆記型電腦、平板電腦及智慧型網路電視等許多消費型電子產品中。在使用者操作中，經常會使用對焦的功能以使畫面中感興趣部分更清晰。在定焦式網路攝影機使用上，如果拍攝畫面過遠或太近，經常會發生畫面模糊的狀況。因此需要一個自動對焦功能以使畫面中某部分達到特定期望的清晰度需求。目前市面上的自動對焦網路攝影機，幾乎皆為系統單晶片架構，其組成元件為影像訊號處理器、影像感測器及搭配音圈馬達等元件所建構成的攝影機模組。基於成本考量下，已有很多不同自動對焦方法被提出來克服在消費型網路攝影機的資源限制(例如緩衝記憶容量及運算複雜度)。其中，爬山演算法因具有低複雜度及容易在硬體上實現特性，因此被廣泛的使用。然而，傳統爬山法經常會有對焦於影像場景中高頻影像區域的問題。這主要會造成當一個畫面中，感興趣區域的影像複雜度低於其它影像內容時，發生對焦錯誤的情況。本研究針對此問題，提出自適性權重估測機制來改善傳統爬山法，藉由影像差異法偵測出移動前景物體，接著給予該區域適當的權重，最後再針對前景物體重新進行對焦。經實驗結果顯示，此新的方法對於不同複雜度的影像類型及在低光源環境中，有顯著的效能提升，並且不需要大量且複雜的運算。另外，此方法非常適用於低成本的消費型網路攝影機產品中。Webcams are actively used for social interaction. Accordingly, this sort of application has been successfully visualized in many consumer electronics products, such as notebooks, tablets and smart TVs. In the context of user-driven processing operations, focusing is probably the most frequently used function developed to obtain a particular desired view of a scene. In the case of fixed-focusing webcams, the captured scene will exhibit blurring contents if the image scene is out of focus. It hence drives the demands for achieving a particular desired clarity in the scene with an auto-focusing function.Almost all of the auto-focusing webcams in the market are based on a system-on-chip(SOC) framework, where a integrated camera module includes lens, complementary metal-oxide-semiconductor(CMOS) sensor, image signal processor(ISP) and voice coil motor(VCM). Due to the consideration of cost, many different auto-focusing approaches have been developed to overcome the limitations pertaining to resources available to consumer webcams. Among these, hill-climbing algorithm is most widely adopted because of its simplicity and easy implementation in hardware for real-time applications. However, traditional hill-climbing solution is limited to focusing on high-frequency blocks of an image scene. This results in a miss-focusing when the complex of region-of-interest (ROI) is much simpler than other contents of an image scene.Motivated by this, we propose an improved hill-climbing algorithm using an adaptive weighting estimation mechanism in this thesis. Experimental results show that this new solution can perform outstandingly well for various scenes and different light conditions without requiring high computational cost. Moreover, the developed scheme is well-suited for implementation in low-cost consumer webcams.
