1 / 38

Image Enhancement [DVT final project]

Image Enhancement [DVT final project]. Speaker: Yu-Hsiang Wang Advisor: Prof. Jian -Jung Ding Digital Image and Signal Processing Lab Graduate Institute of Communication Engineering National Taiwan University. Outline. Target Possible enhancement methods Interpolation

teal
Download Presentation

Image Enhancement [DVT final project]

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript

  1. Image Enhancement [DVT final project] Speaker: Yu-Hsiang Wang Advisor: Prof. Jian-Jung Ding Digital Image and Signal Processing Lab Graduate Institute of Communication Engineering National Taiwan University DISP Lab, Graduate Institute of Communication Engineering, NTU

  2. Outline • Target • Possible enhancement methods • Interpolation • Adaptive osculatory rational interpolation • Enhancement • Bilateral Enhancers • Non Local Means • Conclusion • Reference

  3. Target • Scale image from SD (Standard-definition) to HD (High-definition). • Given: 720x480, YCbCr, 4:2:2, 8 bits per channel. • Target: 1920x1080, YCbCr, 4:4:4, 10 bits per channel.

  4. Possible enhancement methods • Sharpness • Noise reduction • Edge smoothness • Skin-tone enhancement • Texture enhancement • Super resolution (SR) Multi-Image SR [1] Single-Image Multi-Patch SR [1]

  5. Interpolation:Bilinear • 1. Interpolate R1: • 2. Interpolate R2: • 3. Interpolate P: [2]

  6. Interpolation: Adaptive osculatory rational interpolation • The interpolation kernel function of adaptive osculatory rational interpolation (AORI) is more accurately approximate to the ideal interpolation not only in space domain but also in frequency domain. • Apply 4 points to interpolate 1 point in one direction.

  7. Interpolation: Adaptive osculatory rational interpolation • The interpolation function • where r(x) is the interpolated value, g(xk) are the sample values, RI(x) is the interpolation kernel.

  8. Interpolation: Adaptive osculatory rational interpolation • The interpolation kernel [M. Hu and J. Q. Tan, ”Adaptive osculatory rational interpolation for image processing,” Journal of Computational and Applied Mathematics, vol. 195, pp. 46-53, 2006.]

  9. Interpolation: Adaptive osculatory rational interpolation • The original 1920x1080 image:

  10. Interpolation: Adaptive osculatory rational interpolation • The interpolated image by AORI:

  11. Interpolation: Adaptive osculatory rational interpolation • The original 1920x1080 image:

  12. Interpolation: Adaptive osculatory rational interpolation • The interpolated image by AORI:

  13. Enhancement: Bilateral Enhancers • Extended from the bilateral filter (BF). • BF: • A nonlinear filter adopts a low-pass Gaussian filter for both the domain filter and the range filter. • Smooth the noise while preserving edges.

  14. Enhancement: Bilateral Enhancers • Bilateral filter (BF) • with the normalization factor • where r is the input image, h is the output image, Ω(x) is a subset of the input image r with x and ξ are the pixels coordinates. (row, column) [C. Tomasi and R. Manduchi, “Bilateral filtering for gray and color images,” in Proc. ICCV, pp. 839–846, 1998.]

  15. Enhancement: Bilateral Enhancers • Bilateral filter (BF) • with the normalization factor • The function c operates on the spatial domain designed as • The s operates on the range domain designed as

  16. Enhancement: Bilateral Enhancers • Problem of Bilateral filter: • Only edge-preserving and de-noising. • Do not enhance the sharpness of an image.

  17. Enhancement: Bilateral Enhancers • Bilateral enhancers (sharpness + smoothness) • The c function is set as the same as in BF. (Gaussian function on spatial domain) [C. Gatta and P. Radeva, “Bilateral enhancers,” IEEE ICIP, pp. 3161-3164, 2009.] vs

  18. Enhancement: Bilateral Enhancers • Bilateral enhancers (sharpness + smoothness) • The p composes of two parts (p = ps + pe): • the edge-preserving smoothing (ps) • the selective sharpening (pe) vs

  19. Enhancement: Bilateral Enhancers • The edge-preserving smoothing (ps) • ηs: adjusts the intensity of the blurring. • σs: controls how strong should be an edge to be preserved from the blurring. • If the intensity difference is small, Gaussian smoothing is performed.

  20. Enhancement: Bilateral Enhancers • The selective sharpening (pe) • ηe and σe have similar meaning as for the edge-preserving smoothing. • If the intensity difference is small, no enhancement is performed.

  21. Enhancement: Bilateral Enhancers • The interpolated image by AORI:

  22. Enhancement: Bilateral Enhancers • Enhanced the previous page’s image by BE.

  23. Enhancement: Bilateral Enhancers • The original 1920x1080 image:

  24. Enhancement: Bilateral Enhancers • The interpolated image by AORI:

  25. Enhancement: Bilateral Enhancers • Enhanced the previous page’s image by BE.

  26. Framework of System

  27. Enhancement: Non Local Means • Purpose: Noise reduction. • Given an interpolated image • The estimated value • W is a search window of fixed size (we choose 5x5 here) centered at pixel i. • The weights depend on the similarity between the pixel i and j. [A. Buades, B. Coll, and J.-M. Morel, “A non-local algorithm for image denoising,” IEEE CVPR, Jun. 2005.]

  28. Enhancement: Non Local Means • The weighted Euclidean distance • Nk denotes a square neighborhood of fixed size (we choose 3x3) and centered at a pixel k. • α is the standard deviation of the Gaussian kernel. • G(N) is the Gaussian kernel of the same size as Nk.

  29. Enhancement: Non Local Means • The weights are defined as • Z(i) is the normalizing constant • h denotes a degree of filtering. (It controls the decay of the weights function of the Euclidean distances.)(h = 2.5)

  30. Enhancement: Non Local Means • Scheme of NL-means strategy. [6]

  31. Enhancement: Non Local Means • Advantage: • NL-meanscompares the gray level of pixels. • Compare the geometrical configuration in a whole neighborhood. • Disadvantage: • Do not perform sharpness. • Blur some edges.

  32. Enhancement: Non Local Means • The interpolated image by AORI:

  33. Enhancement: Non Local Means • De-noise the previous image by NL-means.

  34. Enhancement: Non Local Means • The interpolated image by AORI:

  35. Enhancement: Non Local Means • De-noise the previous image by NL-means.

  36. Enhancement: Non Local Means • Enhanced the previous page’s image by BE.

  37. Conclusion • Adaptive osculatory rational interpolation is more accurately approximate to the ideal interpolation. • Bilateral enhancers performs well at sharpness and smoothness. • Non local means is mainly used for noise reduction.

  38. Reference • [1] D. Glasner, S. Bagon, and M. Irani, “Super-resolution from a single image,” ICCV, pp. 349-356, Sep. 2009. • [2]http://en.wikipedia.org/wiki/Bilinear_interpolation • [3] M. Hu and J. Q. Tan, ”Adaptive osculatory rational interpolation for image processing,” Journal of Computational and Applied Mathematics, vol. 195, pp. 46-53, 2006. • [4] C. Tomasi and R. Manduchi, “Bilateral filtering for gray and color images,” in Proc. ICCV, pp. 839–846, 1998. • [5] C. Gatta and P. Radeva, “Bilateral enhancers,” IEEE ICIP, pp. 3161-3164, 2009. • [6] A. Buades, B. Coll, and J.-M. Morel, “A non-local algorithm for image denoising,” IEEE CVPR, Jun. 2005.

More Related