臺灣博碩士論文加值系統

在影像曝光融合方法中，若來源影像發生位移或拍攝角度不同而有視差效應時將導致來源影像產生影像不一致的差距，其融合結果一定會出現問題。因此我們模擬當來源影像有著上下視差，其上下視差差距範圍在20~30 pixel之間，我們將這不一致的來源影像去做融合並且排除在融合結果時會發生的問題。在一般的影像曝光融合方法中其輸出結果品質是很依靠輸入影像資訊，因此當輸入影像發生位移或拍攝時手抖動現象，這都會導致合成結果會出現偽影或鬼影現象。而傳統影像融合方法會發生偽影或鬼影現象是因為直接將所有影像像素直接做融合。因此針對此問題我們提出了只針對感興趣區域（ROI）才做融合，並且我們也提出基於一個小規模的線性匹配以使來源影像具有對齊效果。最後我們還使用離散小波變換，以提高融合的品質。最後經過實驗測試，我們的融合結果與傳統的融合結果相比，在會發生偽影或鬼影效應的地方在我們的結果中是可以排除，而其它部分則保留著與傳統結果一樣的視覺品質，因此我們提出的方法是可以針對來源影像具有小差距時去做影像融合。

In the image exposure fusion process, if the source image displacement or shoot different angles have parallax effect will lead to the source image to generate an image inconsistent disparity, fusion results will have problems. Therefore, we simulated the source image has parallax, the disparity between the upper and lower have 20 ~ 30 pixels of range and we refer to these source images do fusion and solve problems in fusion results occur. The conventional exposure fusion suffers the artifact or ghost effects when there are source images misalignment or hand trembling, because the output image quality is only focus on input image information. In the conventional fusion method is let all image pixels directly to do fusion will produce artifact or ghost effect. Therefore, for this question we propose to only for the region of interest (ROI) to do fusion, and our also propose to do a small-scale linear matching in order to align the source images. Final we also to use the discrete wavelet transform to enhance the quality of fusion. Finally, after experimental tests, our results compared with the traditional fusion result, in occur artifacts or ghost effect places our results can be excluded, while other parts retain the same visual quality of with the traditional results, therefore, we proposed method can be used in the image fusion with small image disparity.

ACKNOWLEDGMENT I
中文摘要 II
ABSTRACT III
TABLE OF CONTENTS IV
I. INTRODUCTION 1
II. RELATED WORK 4
2.1 DISCRETE WAVELET TRANSFORM 4
2.2 EXPOSURE FUSION 6
2.2.1 PIXEL WEIGHT CALCULATION 6
2.2.2 MULTI-RESOLUTION FUSION 7
III. PROPOSED ALGORITHM 9
3.1 ROI DETECTION AND SELECTION 10
3.2 BUILD THE WEIGHT OF CE 12
3.3 UPDATE MOTION VECTOR DWT AND IDWT 12
3.4 CALCULATION METHOD OF FUSION 15
IV. EXPERIMENTAL TEST RESULT 20
4.1 OUTDOOR NIGHT SCENE 21
4.2 OUTDOOR DAYTIME SCENES 24
4.3 INDOOR WINDOW NATURAL LIGHT SCENE 27
4.4 INDOOR UNNATURAL LIGHT SCENE 30
V. CONCLUSION 33
REFERENCES 34
VITA 36

[1] T. Mertens, J. Kautz, F. Van Reeth, “Exposure Fusion”, Proc. of the15th Pacific Conference on Computer Graphics and Applications, Maui, Hawaii, pp. 382-390, Oct/Nov 2007.
[2] Tico, M., Gelfand, N.,Pulli, K., “Motion blur free exposure fusion”, In: IEEE International Conference on Image Processing(ICIP),pp.26-29(September 2010).
[3] J. Malik and P. Perona. Preattentive texture discrimination with early vision mechanism. Journal of the Optical Society of America, 7(5):923–932, May 1990.
[4] Yun-Nan Chang and Yan-Sheng Li, “Design of highly efficient VLSI architectures for 2-D DWT and 2-D IDWT”, IEEE Workshop on Signal Processing Systems,pp 133-140, 26-28 September 2001.
[5] P. E. Debevec and J. Malik, “Recovering High Dynamic Range Radiance Maps from Photographs,” in Proc. SIGGRAPH, 1997, pp. 369-378.
[6] Mallat, S., “A Theory for Multiresolution Signal: The Wavelet Representation”, IEEE Trans. Pattern Anal. Machine Intel., vol. 11, July 1989, pp.674-693.
[7] An J, Lee SH, Kuk JG, Cho NI: “A multi-exposure image fusion algorithm without ghost effect”. In IEEE International Conference on Acoustics, Speech, and Signal Processing. Prague; May 2011:1565–1568.
[8] T. Kartalov et al., “A real time global motion compensation for multi-exposure imaging algorithms”, International Conference on Computer as a Tool (EUROCON) pp. 1-4, 2011.
[9] T.-Y. Lin, “Bit-width tradeoffs in hardware discrete wavelet transform architecture”. Department of Electrical Engineering, National Chung Cheng University, Master Thesis, June 2006.
[10] F. Pece and J. Kautz “Bitmap movement detection: HDR fordynamic scenes”, Proc. CVMP, pp.1 -8 2010