植基於模糊預測模型之自動對焦演算法

Abstract

數位相機，本身是一個複雜的系統，含有許多的模組，如自動對焦、自動曝光、資料傳輸、影像處理、等等，其中自動對焦演算法在使用上的先決條件為即時性，而一般的搜尋法，在目前的百萬像素級的搜尋時間過於冗長，針對此問題提出本研究結合預測的方法，來改善此一問題。 本研究針對自動對焦上的問題，提出一個可以改善自動對焦速度及提高自動對焦可靠度的演算法，而演算法利用離線差分方程預測模型之優越的預測特性，可預測轉折點之效果與以少數的取樣點即可獲得預測曲線的趨勢，再利用模糊推論的方法，將離散差分方程預測模型的預測結果，作為推論的依據，進而可以決定步距的大小，大幅降低對焦所需的時間及減少取樣的點數。 在本研究中，我們提出模糊離散差分方程預測模型進行數位相機的離線測試與機上測試，其結果均能精準的預測對焦曲線的趨勢與最佳對焦點之區間。Digital camera is a complex system. There are many modules included in a digital camera system, such as auto-focus, auto-exposure, data transmission, image process, and so on. Real time searching is the most important prerequisite of auto-focusing algorithm. To overcome this problem, we propose a novel method that combines with prediction. This paper proposed an approach of auto-focusing algorithm, which can improve the velocity of the auto-focus, and it can make the auto-focus more reliable. We chosed Discrete differential equation prediction model (DDEPM) as prediction method for the turning point. By the experimental results, it can be found that the model has a better prediction result in the turning point, and it can improve the auto-focusing velocity. Utilizing the result of Discrete differential equation prediction model (DDEPM) as Fuzzy inference input variables, and we can use the fuzzy output to detemine the sapn.The method can reduce the time which camera focuses and captures a clear image and it can decrease the sampling. We use this method to on-line and off-line test in camera. The results of testing show that the Fuzzy DDEPM can predict the Focus curve tendency and the appropriate focus region.