指導教授：傅楸善臺灣大學：生醫電子與資訊學研究所林婷涵Lin, Ting-HanTing-HanLin2014-11-262018-07-052014-11-262018-07-052014http://ntur.lib.ntu.edu.tw//handle/246246/261853本論文提出一個基於雙離散小波轉換即時重建動態模糊影像的方法。將模糊圖像透過兩次小波轉換後，能夠清楚的觀測出清晰圖像與模糊尺度並藉以修正。在本論文中，我們利用Zhang所提出的雙小波轉換 (Double Discrete Wavelet Transform) 出發，利用自相關 (Auto-correlation) 的與登山式搜尋法 (Hill Climbing Algorithm) 方式估測模糊尺度。並利用估測出的尺度資訊推測真實圖像高頻訊號。實驗結果顯示我們提出的方法與其餘方法相較下有較快的執行速度並能即時執行。 處理速度可達30張/秒，可用於即時系統中。Motion blur is a common phenomenon in photography. Blur may render the image useless and lead to inaccurate result. In many inspection industries, blur will cause large error and cannot be ignored. In this thesis, we propose a real-time method to improve motion blur for the medical capsule defect inspection system. We simplify the problem of dealing with motion blur with constant velocity k and moving horizontally. Based on double Haar wavelet transform, we estimate the length of blur kernel with auto-correlation and hill climbing searching algorithm. With the information of blur kernel, we are able to correct the proper coefficients in wavelet domain iteratively. We compare the deblur results and execution time of previous work and our method and achieve satisfactory results.誌謝 i 口試委員審定書 ii 中文摘要 iii ABSTRACT iv Chapter 1 Introduction 1 1.1 Discrete Discrete Wavelet Transformation 3 1.2 Autocorrelation 3 1.3 Hill Climbing Algorithm 4 1.4 Deducing Wavelet Coefficients of Latent Image 4 1.5 Thesis Organization 5 Chapter 2 Related Works 6 2.1 Blur Overview 6 2.2 Deblur Method 7 2.2.1 Hardware Approach 7 2.2.2 Frequency Domain 8 2.2.3 Inverse Deconvolution 9 2.2.4 Wiener Deconvolution 10 2.2.5 Richardson Lucy Deconvolution 11 2.3 Blur Kernel Estimation 12 2.3.1 Defocus Blur Kernel 13 2.3.2 Motion Blur Kernel 14 2.3.3 Camera Shake Blur Kernel 15 2.3 Wavelet Domain Method 16 Chapter 3 Background 18 3.1 Discrete Wavelet Transformation 18 3.2 Double Discrete Wavelet Transform 20 3.3 Haar Wavelet Transform 20 Chapter 4 Methodology 22 4.1 Overview 22 4.2 DDWT Analysis on Blur Model 23 4.3 Estimate Kernel Length: Autocorrelation 27 4.4 Hill Climbing Algorithm 28 4.5 Correct the uj from vij 31 4.6 Equipment 32 Chapter 5 Experiment 33 5.1 Deblurring on a simple signal 33 5.2 Deblurring on Test Data 36 5.3 Deblurring on Real Data 37 Chapter 6 Conclusion and Future Work 43 Reference 451426630 bytesapplication/pdf論文公開時間：2019/07/16論文使用權限：同意有償授權(權利金給回饋學校)去模糊小波轉換自相關最佳化登山法thesishttp://ntur.lib.ntu.edu.tw/bitstream/246246/261853/1/ntu-103-R01945021-1.pdf