基於邏輯估測之反光消除

Abstract

在本篇論文中，我們提出一個基於邏輯估測之反光消除方法用來處理兩張情景相同但反光區域不同的反光影像。假設兩張輸入的影像包含一些互補的資訊，這兩張影像即可以相互參考去修護反光的區域，即使兩張圖的反光區域有重疊的部分，我們提出的方法仍然可以將反光去除以及修補影像的結構。 首先，我們使用尺度不變特徵轉換(Scale-Invariant Feature Transform)去描述兩張輸入影像的特徵，然後利用這些資訊將其中一張輸入影像型變成另一張輸入影像的樣子，根據特徵點的對應關係，我們分別產生出一張反光重疊的影像(Specular-overlap Image)以及一張反光聯集的影像(Specular-union Image)。接著，我們是用這兩張產生出來的影像套入邏輯估測的模型，找出反光重疊影像以及反光聯集影像當中的反光區域。最後，我們利用顏色轉換及顏色混合的方法將反光重疊影像當中的反光區域修護完成。實驗結果顯示我們的方法可以自動地將反光去除，以及修護好反光區域當中的結構。

In this thesis, we proposed a specular reflection removal method based on logical estimation to deal with two specular images with the same scene but with different specular regions. Assume the two input images contain some complementary information. They can reference each other to recover the specular region. Even if there is a specular region appearing in both images, our method still can remove specularity and recover the structure in the image. At first, we use Scale-Invariant Feature Transform to describe the features of two input images and employ information to warp one of two input images to another one. According to the matching relationship of the feature points, we generate a specular-overlap image and a specular-union image. Then, we use these two images to fit a logical estimation model and find the specular region in both specular-overlap image and specular-union image. Finally, we recover the specular region of specular-overlap image by applying color transformation and color blending. Our results show that the proposed method has removed the specular reflection and recover the structure of contents in the specular region automatically.