一個有效的閃光燈攝影增強法

Abstract

如果在偏暗的環境下要拍照的話，通常我們會使用閃光燈，藉以補光以得到更多的影像細節，但不幸的是這樣卻會破壞相片中的色彩。要是不使用閃光燈的話，雖然可以得到色彩真實的影像，但是其需要較長的曝光時間或須以高ISO值拍攝，這會導致影像的模糊或產生嚴重雜訊。所以學者提出藉由結合一張沒有使用閃光燈與另一張有使用閃光燈的兩張相片來計算產生一張同時具有真實色彩與細節豐富的影像。
由於來源影像為分別使用閃光燈的影像與以較長曝光時間拍攝但沒有使用閃光燈的影像，所以很明顯的除了有一般位移、旋轉、縮放等影像幾何失真外，還會有嚴重光照度(illumination)的明暗變化與陰影，更可能會有物體的存在，這使得傳統點對應(point-mapping)與Ward影像等人提出基於平均臨界值(mean threshold bitmap)的影像校正法並不能適用於這類的影像。本論文中，我們首先針對使用閃光燈的影像序列，提出一個強健的影像校正(image registration)法，其是基於快速傅立葉轉換(FFT)，所以能有效克服影像序列中光照差異大，甚至有移動物體的問題。
接著本論文提出一個基於YS(下標 b)S(下標 r)色彩空間的閃光燈攝影增強法來改善現有方法中色彩分離的部份，實驗結果顯示所提方法在不影響合成影像品質的條件下，能節省大約50%的記憶體使用量。由於現在的數位相機的解析度都大幅的提升，超過千萬像素的相機已是相當普遍，這使得所提方法能省下大量的記憶體空間，且有效加快影像合成時所需的計算時間。此外我們也發現所提YS(下標 b)S(下標 r)色彩分離方法在相同的記憶體使用量下，比基於YS(下標 b)S(下標 r)色彩空間的閃光燈攝影增強法能保留更多的細節資訊。Under dark illumination, a photographer is usually faced with a frustrating dilemma: to use the flash or not. A picture relying on the available light usually has a warm atmosphere, but suffers from noise and blur. On the other hand, flash photography cause unacceptable artifact such as flat and harsh lighting, although it provides more details. Some researchers proposed the detail transfer approach to enhance the flash photography by combining a picture taken with the available light and one taken with the flash. However, they do not focus on the problem of image registration.
Image registration is important for images taken with handheld cameras, for example, changing the camera settings often results in camera motions. Since the flash/no-flash images do have significant differences due to the change in illumination, in this paper, a robust image registration method based on FFT is proposed for registration of such image pairs. An image pair with moving objects is used in this paper to demonstrate the robustness of the proposed method.
We also figure out the problem of the conventional methods of flash photography enhancement, a large amount of memories are required during the computation. It can be a serious computational burden when the size of the input images is large. In this paper, we propose a new method based on YS(subscript b)S(subscript r) color space to alleviate this problem. Experimental results show that the advantages of the proposed method include: (1) It saves up to 50% of the memory space; and (2) It preserves more details and sharpness than the approach based on YC(subscript b)C(subscript r) color space when the same quantity of memory is used.