運用手機加速規之影像去模糊技術

Abstract

數位相機與智慧型手機越來越普遍，由於相機技術的進步，它們也相當容易操作，然而模糊的相片依然是一個尚待解決的問題。模糊相片通常發生於使用者拍照時晃動到相機，在長曝光時間或低環境光源的情況下特別容易發生。影像去模糊是一種將模糊影像還原成清晰且視覺上可接受的相片的技術。影像去模糊技術分為相當多種，近年來的技術著重在兩主要步驟：估計模糊函數和影像重建。之前的研究證實在影像邊緣的區域，估計模糊函數的效果會比在影像平滑的區域更佳。 在本篇論文中，我們利用手機加速規(accelerometer)的資訊估計模糊函數，透過手機加速規的資訊輔助，估計模糊函數的精準度可以顯著的增加。這類利用特殊硬體設備在相機上輔助影像去模糊的研究以前也曾出現過，我們和之前的方法最大的不同在於：我們結合手機加速規的資訊和盲反卷積(blind deconvolution)的演算法。我們所提出的新方法介於盲反卷積和非盲反卷積(non-blind deconvolution)方法之間，因此我們稱它為半盲反卷積(semi-blind deconvolution)。我們提出方法的去模糊效果勝過其他當今最先進的盲卷積影像方法或只用加速規資訊的影像去模糊。

Digital cameras and smart phones have become more common than before. They are simpler to operate due to the advance of the camera technology. However, blurring remains an unsolved problem. This often occurs when users sway the camera while taking photos, especially with long exposure time or in a low-light environment. Image deblurring is the method of reconstructing a sharp and visually plausible image from a blurred image. Image deblurring methods have different types, but recent deblurring approaches focus on two main stages: blur kernel estimation and image restoration. Previous studies have proved that blur kernel estimation algorithms perform better on the edge part of an image than on smooth part. In this thesis, we make use of the accelerometer in the cellphone to estimate the blur kernel. The accuracy of blur kernel estimation is significantly improved with the additional information of the accelerometer. Deblurring with the special hardware on the camera has also shown before. The main difference between our method and previous methods is that we blend the information of the accelerometer into blind deconvolution algorithms. This new method is an intermediate of blind deconvolution and non-blind deconvolution, so we call it semi-blind deconvolution. The deblurring results by our method surpass other state-of-the-art blind deconvolution methods or deblurring by accelerometers in real data.