臺灣博碩士論文加值系統

在此篇論文當中，我們將介紹一個階層式降雜訊的演算法，而此演算法能夠有效地消除雜訊並且同時保留影像的細節。
這個階層式的方法首先會將一張影像轉換為向下取樣的四張圖片，其向下取樣的程度皆不同，因此這四張不同大小的圖片粗略而言分別隱含有原始圖片中不同頻率的資訊。因此我們不必將原始影像轉換到頻率域即可對圖片的不同頻率做不同的處理。
接下來，我們會在這四張子圖中分別找出影像的邊，也就是一般所謂的影像細節，並且再對細節及非細節處分別做不同的平滑處理。
當所有操作在這四張子圖都完成時，我們會將它們重新合併而成為一張消除完雜訊的影像。

In this thesis, we introduce a hierarchical noise reduction algorithm which can reduce the noise and preserve the details of the images.
This hierarchical method converts an image into a set of four downscaled images which contain information of different frequency in original image. Hence we can process the image in different frequency layers without transferring to frequency domain.
Afterwards, we find edge pixels in each sub-image and apply different smoothing methods to edge pixels and non-edge pixels.
After handling all four layers of original image, four sub-images are combined to produce a de-noised image.

Contents
口試委員會審定書 i
誌 謝 ii
摘 要 iii
Abstract iv
Contents v
List of Figures vi
List of Tables viii
Chapter 1 Introduction 1
1.1 Motivation 1
1.2 Introduction to Noise Types 2
1.3 The Characteristics of Image and Noise 3
1.3.1 Digital Image Characteristics 3
1.3.2 Noise Characteristics 5
Chapter 2 Analysis of Noise 10
2.1 Noise Level Measurement 10
2.2 Noise Models [5] 11
2.3 Related Works [8] 13
2.3.1 Wavelet Analysis [16] 13
2.3.2 Bilateral Filtering [5] 13
2.3.3 Kernel-Regression Based [13] 14
Chapter 3 Original Hierarchical Method 15
3.1 Introduction to Hierarchical Method 15
3.1.1 Background of Hierarchical Method 15
3.1.2 Overview of the Hierarchical Method 15
3.1.3 Detailed Description of Hierarchical Method [12] 16
3.1.4 Flow Chart 27
3.2 Drawbacks of the Original Method 28
3.2.1 The Patent 28
3.2.2 The Application Program 28
Chapter 4 Our Proposed Method 30
4.1 Multi-Threshold 30
4.2 Different Mask Sizes 30
4.3 Pixel Distance 32
4.4 Noise Control 32
4.5 Merge Coefficient 33
4.6 Unsharp Mask 34
4.7 Flow Chart 34
Chapter 5 Experiments and Results 36
5.2 Images with Color-SNR 36
5.3 Images without C_SNR 44
Chapter 6 Conclusion and Future Work 48
6.1 1.1 Conclusion 48
6.2 1.2 Future Work 48
Reference 49

List of Figures
Figure 1.1 Noise types. (a) Fixed pattern noise. (b) Banding noise. (c) Random noise [9]. 3
Figure 1.2 RGB color space [17]. 4
Figure 1.3 YCbCr color space [18]. 4
Figure 1.4 Noise separated into luminance noise and chroma noise [10]. 6
Figure 1.5 Different frequencies of noise [10]. 6
Figure 1.6 Different magnitudes of noise [10]. 7
Figure 1.7 Histograms of different-magnitude noise [10]. 8
Figure 1.8 Noise appearances. (a) Noise appearances with different ISO speeds (taken with Epson PhotoPC 800). (b) Noise distributions in different channels. (c) Noise appearances on different luminance areas [10]. 9
Figure 2.1 Salt-and-pepper noise applied to original image. 11
Figure 2.2 Gaussian noise applied to original image. 12
Figure 2.3 An example of wavelet transform. 13
Figure 2.4 The bilateral filter [5]. 14
Figure 2.5 The concept of kernel-regression based filtering [13]. 14
Figure 3.1 An example of multi-scaled frequency image. 17
Figure 3.2 Mask for judging edge pixels. 18
Figure 3.3 Adjacent pixels of pixel of interest. 19
Figure 3.4 Mask for eliminating mislabeled edge pixel cluster. 20
Figure 3.5 An example of edge cushioning. 21
Figure 3.6 Luminance gradient masks. 22
Figure 3.7 Color gradient masks. 25
Figure 3.8 Flow chart of original hierarchical method [12]. 27
Figure 3.9 An example of color mistaken. 29
Figure 4.1 Multi-threshold 30
Figure 4.2 Different masks for judging edge pixel. 31
Figure 4.3 Results of different masks 31
Figure 4.4 Results of different parameter C 33
Figure 4.5 Results of different merge coefficients. 33
Figure 4.6 The effect of unsharp mask. 34
Figure 4.7 The flow chart of our proposed method. 35
Figure 5.1 Result 1. 37
Figure 5.2 Result 2. 38
Figure 5.3 Result 3 39
Figure 5.4 Result 4. 40
Figure 5.5 Result 5. 44

List of Tables
Table 5.1 Processing parameters of Figure 5.1 to Figure 5.4. 41
Table 5.2 C_SNR and voting results of twenty images. 43
Table 5.3 Processing parameters of Figure 5.5. 45
Table 5.4 Voting results of twenty images. 47

[1]M. Almond, “Noise Reduction Tool Comparison,” http://www.michaelalmond.com/Articles/noise.htm, 2005.
[2]A. Buades, B. Coll, J.M Morel, “ A Non-local Algorithm for Image Denoising, ” Proceedings of IEEE Int. Conf. on Computer Vision and Pattern Recognition, San Diego, CA, Vol. 2, pp: 60 – 65, 2005.
[3]R. C. Gonzalez and R. E. Woods, Digital Image Processing, 2nd Ed., Prentice-Hall, Upper Saddle River, NJ, 2002.
[4]R. M. Haralick and L. G. Shapiro, Computer and Robot Vision, Vol. I, Addison Wesley, Reading, MA, 1992.
[5]Y. L. Huang, “Noise Reduction Using Enhanced Bilateral Filter,”2006.
[6]KammaGamma, “RGB Noise Reduction,” http://kammagamma.com/articles/rgb-noise-reduction.php, 2007.
[7]KammaGamma, “Lab Noise,” http://kammagamma.com/articles/lab-noise.php, 2007.
[8]S. Lansel, “Recent Trends in Denoising Tutorial: Publications,” http://www.stanford.edu/~slansel/tutorial/publications.htm, 2007.
[9]S. McHugh, “Digital Camera Image Noise: Concept and Types,” http://www.cambridgeincolour.com/tutorials/noise.htm, 2008.
[10]S. McHugh, “Image Noise: Examples and Characteristics,” http://www.cambridgeincolour.com/tutorials/noise2.htm, 2008.
[11]Y. Nakami, “Image Noise Reduction,” US Patent#7324701 B2, 2008.
[12]A. Petrosyan and A. Ghazaryan, “Method and System for Digital Image Enhancement,” US Application#11/116,408, 2006.
[13]H. Takeda, S. Farsiu, and P. Milanfar, "Kernel Regression for Image Processing and Reconstruction," IEEE Transactions on Image Processing, vol. 16, no. 2, pp. 349-366, Feb. 2007.
[14]Wikipedia, “Noise Reduction,” http://en.wikipedia.org/wiki/Noise_reduction, 2008.
[15]Wikipedia, “Image Noise,” http://en.wikipedia.org/wiki/Image_noise, 2008.
[16]Wikipedia, “JPEG 2000,” http://en.wikipedia.org/wiki/JPEG_2000, 2008.
[17]Wikipedia, “RGB color space,” http://en.wikipedia.org/wiki/RGB, 2008.
[18]Wikipedia, “YCbCr,” http://en.wikipedia.org/wiki/Ycbcr, 2008.