李明穗臺灣大學:資訊工程學研究所鞠肇宗Chu, Chao-TsungChao-TsungChu2010-05-172018-07-052010-05-172018-07-052009U0001-3007200916552400http://ntur.lib.ntu.edu.tw//handle/246246/183376在這本篇論文中，我們提出一個基於適性內容實現單張影像去霧化的方法。由於影像被霧化的程度與距離息息相關，而對於一張影像，每個特定區域(例如：樹、建築物或其他物體)裡面的像素，到相機的距離大致上都會一樣，所以我們假設每個區域被霧化的程度也會相同。在這樣的假設下，我們發展出一個對影像去霧化的方法。先，我們用暗原色先驗統計(Dark Channel Prior)的方法來預估霧光 (airlight) 的值。霧光(airlight)是在一張霧化的照片中用以代表霧的元素。基於我們觀察到在一張影像中，一個特定區域的像素裡面的像素，到相機的距離大致上都會一樣，所以每個區域的霧化轉換 (transmission) 值會是一樣的。所以我們用基於平均值移動分割(Mean Shift segmentation) 的方法分割我們的輸入影像。接著，我們使用一個函數來預估每一個區域的霧化轉換。預估完霧化轉換之後，我們用 Soft Matting 的方法把我們原先預估的霧化轉換對應圖變的更好。然後把霧化的影像給回復回來。實驗結果顯示出我們提出的方法，可以有效地對影像實現去霧，並且提供了一個準確預估的霧化轉換之值可以作為更深入的應用。In this thesis, we present a content adaptive method for single image dehazing. Since the degradation level affected by haze is related to the depth of the scene and pixels in each specific part of the image (such as trees, buildings or other object) tend to have same depth to the camera, we assume that the degradation level affected by haze of each region is the same. Based on this assumption, we develop a single image dehazing method. irst, we use Dark Channel Prior method to estimate the airlight vector which represents the haze component in hazy images. By the observation that pixels in each specific part of the image (such as trees, buildings or other object) tend to have same depth to the camera, the transmission in this region is also the same. So, we use Mean Shift segmentation to segment our input image into different regions. Then, we recover the scene radiance by using a cost function estimating the transmission in each region. After the transmission map was estimated, we use Soft Matting to refine this transmission map. Results demonstrate the proposed method’s power to remove the haze layer as well as provide a reliable transmission map which can be exploited for further usage.口試委員會審定書 #謝 i文摘要 iiBSTRACT iiiONTENTS ivIST OF FIGURES vihapter 1 Introduction 1.1 Introduction of Dehazing 1.2 Thesis Organization 3hapter 2 Related Work 5.1 Haze Degradation Model Overview 5.2 Related Work of Dehazing 6.2.1 Dark Channel Prior 7.2.2 Correction of Contrast Loss 8.2.3 Single Image Dehazing Methods 9.2.4 Multiple Image Dehazing Methods 11.2.5 Dehazing Methods with User Input 12.3 Mean Shift Segmentation 13.4 Soft Matting 14hapter 3 Single Image Dehazing using Mean Shift Segmentation 16.1 System Overview 16.2 Image Segmentation 17.3 Atmospheric light Estimation 19.4 Cost Function for Transmission Map Estimation 21.5 Refinement of Transmission Map using Soft Matting 25.6 Recovering the Scene Radiance 27hapter 4 Experimental Results 29.1 Resultant images 29.2 Comparison with other haze removal methods 41hapter 5 Conclusion and Future Work 45.1 Conclusions 45.2 Future work 46EFERENCE 50application/pdf2398870 bytesapplication/pdfen-US單一影像去霧化基於平均值移動分割(Mean Shift segmentation)暗原色先驗統計(Dark Channel Prior)影像回復Single Image DehazingMean Shift segmentationDark Channel PriorSoft MattingImage Restorationthesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/183376/1/ntu-98-R96922103-1.pdf