![Haralick: Algorithm
Gray-level co-occurrence matrix
- Element [i, j]: number of times pixel of
intensity i is adjacent to intensity j
- Divide matrix by number of comparisons
made (~ number of pixels in image)
http://murphylab.web.cmu.edu/publications/boland/boland_node26.html](https://image.slidesharecdn.com/9computervisionpatternrecognition-160105184048/85/Computer-Vision-Pattern-Recognition-51-320.jpg)
Computer Vision: Pattern Recognition
- 2. Credit: Saturday Morning Breakfast Cereal http://www.smbc-comics.com/?id=2088
- 3. Definitions Computer vision: the ability for a computer to understand the contents of images Pattern recognition: Classification of images Image processing: preparing images for human or computer consumption
- 4. Computer Vision: Applications Object recognition Facial recognition and expression recognition Surveillance Camera stabilization GIS Crowd / moving object tracking Robotics Computer graphics & Games
- 5. Tools Mahotas SciPy ndimage OpenCV Pillow scikit-image NumPy, Matplotlib, scikit-learn http://www.pyimagesearch.com/2014/01/12/my-top-9-favorite-python-libraries- for-building-image-search-engines/
- 6. Mathematical Concepts Convolution & Correlation Discrete Differentiation / Image Gradient Determinants & Trace Eigenvectors & Eigenvalues Statistics
- 7. Credit: Mubarak Shah, University of Central Florida
- 8. Credit: Mubarak Shah, University of Central Florida
- 9. Convolution vs Correlation Doesn’t matter which one you choose as long as your kernel is symmetric Convolution is associative & commutative
- 10. Credit: Mubarak Shah, University of Central Florida
- 11. Gradient Magnitude & Direction
- 14. Credit: Mubarak Shah, University of Central Florida
- 15. Image processing: Components Smoothing filters Edge detection Interest point detection Descriptors
- 16. Smooth Filtering Averaging Gaussian Filter Median Filter Bilateral Filter
- 17. Averaging Credit: Mubarak Shah, University of Central Florida
- 18. Credit: Mubarak Shah, University of Central Florida
- 19. Median Filter Non-linear filter Take all intensities within window Sort intensities Use intensity in the middle
- 20. Bilateral Filter Non-linear filter Uses Gaussian filtering, but adjusts the weights according to how closely a pixel within the window matches the intensity of the center pixel
- 21. Smooth Filtering: Summary Averaging: Blurs edges the most Gaussian Filter: Good for Gaussian noise, used often Median Filter: Good for salt & pepper noise Bilateral Filter: Preserves edges the most
- 24. Prewitt & Sobel Edge Detectors Prewitt Edge Detector (1970) - Gradient operator w/ averaging Paper: J.M.S. Prewitt. “Object Enhancement and Extraction”. 1970 Sobel Edge Detector (1968) - Gradient operator w/ weighted average Talk: I. Sobel. “Isotropic 3x3 Image Gradient Operator”. 1968
- 30. Marr-Hildreth Edge Detector (1980) Laplacian of Gaussian Inspired by neurophysiology: Neurons for the eyes perform Gaussian filtering Paper: D. Marr, E. Hildreth. “Theory of Edge Detection”. 1980
- 34. Canny Edge Detector (1986) Gradient of Gaussian Pure mathematical approach Paper: J. Canny. “A Computational Approach To Edge Detection”. 1986
- 37. Credit: Mubarak Shah, University of Central Florida
- 38. Credit: Mubarak Shah, University of Central Florida
- 44. Credit: Mubarak Shah, University of Central Florida
- 45. Credit: Mubarak Shah, University of Central Florida
- 46. Credit: Mubarak Shah, University of Central Florida
- 47. Credit: Mubarak Shah, University of Central Florida
- 48. Credit: Mubarak Shah, University of Central Florida
- 49. Credit: Mubarak Shah, University of Central Florida
- 50. Haralick Texture Descriptor (1973) Paper: R.M. Haralick. “Textural Features for Image Classification”. 1973
- 51. Haralick: Algorithm Gray-level co-occurrence matrix - Element [i, j]: number of times pixel of intensity i is adjacent to intensity j - Divide matrix by number of comparisons made (~ number of pixels in image) http://murphylab.web.cmu.edu/publications/boland/boland_node26.html
- 52. Haralick: Algorithm 4 co-occurrence matrices, one for each adjacency direction: horizontal, vertical, left and right diagonals
- 53. Haralick: Zernike moments (Features) ● Angular second moment ● Contrast ● Correlation ● Sum of differences: variance ● Inverse difference moment ● Sum average ● Sum variance ● Sum entropy ● Entropy ● Difference variance ● Difference entropy ● Info. measure of correlation 1 ● Info. measure of correlation 2 ● Max. correlation coeff.
- 54. Pyramids (1983) Gaussian Pyramid Application: Image compositing Laplacian Pyramid Application: Image compression Paper: E.H. Adelson, P.J. Burt. “The Laplacian Pyramid as a Compact Image Code”. 1983
- 55. Credit: Mubarak Shah, University of Central Florida
- 56. Credit: Mubarak Shah, University of Central Florida
- 57. Credit: Mubarak Shah, University of Central Florida
- 58. Credit: Mubarak Shah, University of Central Florida
- 59. Reduce & Expand operations performed using: Credit: Mubarak Shah, University of Central Florida
- 60. Credit: Mubarak Shah, University of Central Florida
- 61. Credit: Mubarak Shah, University of Central Florida
- 62. Credit: Mubarak Shah, University of Central Florida
- 63. Credit: Mubarak Shah, University of Central Florida
- 64. Credit: Mubarak Shah, University of Central Florida
- 65. SIFT & SURF SIFT: Scale-Invariant Feature Transform SURF: Speeded-Up Robust Features
- 66. SIFT (1999) Inspired by primate visual system (human, ape, monkey, etc.) Paper: D. Lowe. “Distinctive Image Features from Scale-Invariant Keypoints”. 2004
- 67. Credit: Mubarak Shah, University of Central Florida
- 68. Credit: Mubarak Shah, University of Central Florida
- 69. Credit: Mubarak Shah, University of Central Florida
- 70. Credit: Mubarak Shah, University of Central Florida
- 71. Credit: Mubarak Shah, University of Central Florida
- 72. Credit: Mubarak Shah, University of Central Florida
- 73. Credit: Mubarak Shah, University of Central Florida
- 74. Credit: Mubarak Shah, University of Central Florida
- 75. Credit: Mubarak Shah, University of Central Florida
- 76. Credit: Mubarak Shah, University of Central Florida
- 77. Credit: Mubarak Shah, University of Central Florida
- 78. SURF (2006) Paper: Herbert Ray, et al. “SURF: Speeded Up Robust Features”. 2006
- 79. SIFT vs SURF SURF has similar performance to SIFT, but is much faster When time isn’t critical, SIFT tends to outperform SURF
- 81. Acknowledgements: Videos M. Shah, University of Central Florida- Center for Research in Computer Vision YouTube Videos https://www.youtube.com/user/UCFCRCV A. Ng, Stanford University https://youtu.be/n1ViNeWhC24
- 82. Acknowledgements: Books J.E. Solem. Programming Computer Vision with Python. 2012 W. Richert, L.P. Coelho. Building Machine Learning Systems with Python. 2013 A. Rosebrock. Practical Python and OpenCV. 2014