數位相機之快速動態估測演算法設計

Abstract

本研究針對數位相機中所預設的動態估測演算法(三步搜尋演算法)進行改良。三步搜尋演算法雖然在實作上相當容易，但是在計算方面仍然佔據過多的計算量。為了減低計算量，鑽石形搜尋演算法被提出。雖然鑽石形搜尋演算法在PSNR與計算量上比三步搜尋演算法還好，但是其不規律的搜尋樣式卻是一個缺點。近年來，六角形搜尋演算法被提出來。原因為其搜尋樣式較趨向圓形，形成一個較規律的搜尋樣式，並且在計算量與畫面品質方面有更好的效果。在另ㄧ方面，此數位相機中含有一個硬體加速器，稱之為IMX。IMX在使用上有一個嚴格的限制，就是搜尋的樣式必須是規律的。因此利用六角形的特點並配合相機硬體的限制而提出一個適用於此相機平台的快速搜尋演算法，稱為預測式線性類六角形搜尋演算法。由實驗結果顯示，在計算量方面比原本的三步搜尋演算法平均減少42.07%的計算量，而在PSNR方面卻僅僅損失8.05%，於人類視覺上是可以被接收的。

This research aims to improve the three-step search algorithm (TSS) used in a digital still camera (DSC). Although the TSS is very convenient for programming, its computation is still too much. In order to reduce computation, diamond search (DS) is proposed. Even if the DS has a fine results in PSNR and the number of search points over the TSS, its search pattern is more complicated than the TSS. In recent years, hexagonal-based search algorithm (HEXBS) is proposed. Because its geometry shape tends to a circle and its search pattern is more regular than the DS, it performs better than the other algorithms in term of the PSNR value and the number of search points. On the other hand, the DSC includes a hardwareaccelerator called as IMX. The IMX has a strictly restriction in which the search pattern must be regular. Based on this restriction of DSC, we design a new algorithm called as predictive linear hexagon-like search algorithm (PLHLS). By experimental results, the PLHLS can efficiently reduce the amount of computation about 42.07% and only loses 8.05% in PSNR value compared with the TSS. In the view points of human vision, this result is acceptable.