Super resolution for Camera Module

最近數位相機的使用是越來越普遍了， 而且也有越來越多人擁有具有相機功能的可攜式裝置， 如照相手機、個人數位助理、等等諸如此類設備。每個人都希望可以隨時為回憶留下照片記念。但是並不是每個人都可以擁有很好的相機功能， 所以拍出來的影像品質都不太好。因此我們需要一個強健性的超解析度演算法去重建回一張較高解析度的影像， 所以我們將介紹我們的方法去處理不同影像運動情況。但是在真實的情況可能會遭受不同的狀況， 例如相機本身快門速度的限制， 像機拍攝時的穩定度， 或是過快的移動物體…等等都會造成影像品質的下降。所以在我們的方法中將會專注在超解析度演算法中最重要的運動估計部份。 方法中將採用影像相減的作法去進行運動的估計。因此我們能找到一個精確的運動範圍去產生比較好的結果。Recently, digital still cameras are very popular, and many people have portable devices with camera module such as camera phone, PDA, and other devices. Each of them usually wants to take a picture for remembrance at any time. But some of their images have poor quality. Therefore we need a robust algorithm to reconstruct a high-resolution image, so we will introduce our method to conquer different image motion situations. But in real scene, super-resolution algorithm may suffer many difficulties such as camera shutter speed limitation, image jitter, fast motion objects, and so on. Because above situations will result in reduced image quality, we will focus on motion estimation which is the most important part in super-resolution algorithm. We propose an alternate approach using image subtraction method to estimate motion objects. Therefore we can find accurate object motion range to get a better result.目 錄 口試委員會審定書…………………………………………………………………..i 誌謝…………………………………………………………………………………..ii 摘要………………………………………………………………………………….iii Abstract...…………………………………………………………………………….iv 目錄…………………………………………………………………………………..v 圖錄………………………………………………………………………………….vi 表錄……………………………………………………………………………….....vi Section 1 Introduction………………………………………………….……………1 Section 2 Prior arts………………………………………………………….……....3 2.1 Irani and Peleg’s method…………………………………………………….3 2.2 Chen and Kuo’s method…………………………………………………….4 Section 3 Our method………………………………………………………………14 3.1 Modified Image subtraction method……………………………………….14 3.2 Finding real movement range and modified matching technique…………17 Section 4 Experiment Result………………………………………………………22 Section 5 Conclusion…………………………………………………….…………27 Section 6 Future Work…………………………………………………….……….28 Section 7 Reference ………………………………………………….……………29 圖 錄 Figure 1 Simulation results of interpolation using merely one image………………..1 Figure 2 Performance with 1st to 5th order of interpolation applied for initial guess…4 Figure 3 Local matching technique…………………………………………………...7 Figure 4 Global matching technique………………………………...………………..7 Figure 5 Gray scale results of Chen et al. [2] proposed method with a fixed scene and a simulated moving camera…………………………………………………………8 Figure 6 Color results of Chen et al. [2] proposed method with a fixed scene and a simulated moving camera……………………………………………………………9 Figure 7 Results of Chen et al. [2] proposed method with real scene captured by video camcorder Hitachi DZ-MV350A…………………………………………….10 Figure 8 Results of Chen et al. [2] proposed method with real scene captured by video camcorder Hitachi DZ-MV350A. ……………………………………………11 Figure 9 Flow-chart of Chen et al. [2] method..……………………………………12 Figure 10 Result of initial image subtraction in 2nd and 3rd frames with respect to first frame (reference frame) …………………………………………………………...15 Figure 11 Result of modified image subtraction in 2nd and 3rd frames with respect to first frame (reference frame)…………………………………………………………16 Figure 12 Flow chart of our proposed algorithm. …………………………………20 Figure 13 Experiment result with fixed camera location and stationary background with multiple moving objects.………………………………………………………22 Figure 14 Experiment result with fixed camera location and stationary background with moving object which is human been.………………………………………24 Figure 15 Experiment result with fixed camera location and stationary background with moving object which is human been.…………………………………………25 表 錄 Table 1The performance of our result compares with Cubic Interpolation and Chen et al.’s method………………………………………………………………………….2