利用多重解析邊緣梯度的資訊來增強靜態影像畫質

Abstract

本論文的目的在於結合tone mapping的技術和以梯度為基礎的陰影去除技術來增強靜態影像畫質，期望結果如同在均勻光源下所獲得的影像一致。因此可以讓辨認系統或是監控系統在不受光影的影響下而能獲得更佳的結果，或是將原本拍攝環境不佳的影像還原成在均勻光源下拍攝的影像。 本論文的方法延伸Fattal等人[8]的工作，Fattal等人的方法可以讓太亮或太暗區域的細節部分顯現出來。再搭配我們提出去除模糊陰影邊界的方法，也就是利用模糊邊界辨認器的演算法來偵測陰影的邊界，再調整該區域的梯度，就可以達成去除陰影的目的。在我們的實驗結果中發現這個方法是可行的。不過仍有少數的問題尚待解決，像在陰影處的顏色會有色偏的問題，在模糊邊緣的區域會有一點模糊以及光暈的現象，這些問題如果可以解決，相信辨認系統或是監控系統在未來會有更大突破。

The main purpose of this thesis is to combine the technique of tone mapping and gradient based shadow removal technique to enhance the image. We expect to gain the result which is similar to the same scene but with uniform lighting. Therefore, we could let the surveillance or the recognition system gain better result, or restore the image captured under poor condition to a better image as if it was captured under uniform lighting environment. The method of this thesis extends the work of Fattal et al. [8]. Their method could reveal the detail in the bright and dark areas. We simply employ the fuzzy edge classifier to detect the fuzzy edge area, and then adjust the gradient in the fuzzy edge area. In our experimental results, this method is proved feasible. But, there still are some problems remaining to be solved, such as “color shifting” in the shadow area, and blurring the fuzzy edge area and some remaining halo artifact. If these defects can be removed, the surveillance and recognition applications can become much practical.