MRI slices at different depths 3d brain scan Ω Z 3 C 3 dimensions for space 3d

Mri slices at different depths 3d brain scan ω z 3 c

This preview shows page 111 - 123 out of 124 pages.

MRI slices at different depths 3d brain scan: Ω Z 3 C 3 dimensions for space 3d pixels are called voxels (“volume elements”) 88
Image of page 111

Subscribe to view the full document.

Image representation – Semantic gap Semantic gap in CV tasks Gap between tensor representation and its semantic content. 89
Image of page 112
Image representation – Feature extraction Old school computer vision Semantic gap: initial representation of the data is too low-level, Curse of dimensionality: reducing dimension is necessary for limited datasets, Instead of considering images as a collection of pixel values (tensor), we may consider other features/descriptors: Designed from prior knowledge Image edges, Color histogram, Local frequencies, High-level descriptor (SIFT). Or learned by unsupervised learning Dimensionality reduction (PCA), Parameters of density distributions, Clustering of image regions, Membership to classes (GMM-EM). Goal: Extract informative features, remove redundancy, reduce dimensionality, facilitating the subsequent learning task. 90
Image of page 113

Subscribe to view the full document.

Image representation – Feature extraction Example of a classical CV pipeline 1 Identify “interesting” key points, 2 Extract “descriptors” from the interesting points, 3 Collect the descriptors to “describe” an image. 91
Image of page 114
Image representation – Feature extraction – Key point detector Key point detector Goal: to detect interesting points (without describing them). Method: to measure intensity changes in local sliding windows. Constraint: to be invariant to illumination, rotation, scale, viewpoint. Famous ones: Harris, Canny, DoG, LoG, DoH, . . . 92
Image of page 115

Subscribe to view the full document.

Image representation – Feature extraction – Descriptors Scale-invariant feature transform (SIFT) (Lowe, 1999) (Source: Ravimal Bandara) Goal: to provide a quantitative description at a given image location. Based on multi-scale analysis and histograms of local gradients. Robust to changes of scales, rotations, viewpoints, illuminations. Fast, efficient, very popular in the 2000s. Other famous descriptors: HoG, SURF, LBP, ORB, BRIEF, . . . 93
Image of page 116
Image representation – Feature extraction – Descriptors SIFT – Example: Object matching 94
Image of page 117

Subscribe to view the full document.

Image representation – Feature extraction – Bag of words Bags of words (Source: Rob Fergus & Svetlana Lazebnik) Bag of words : vector of occurence count of visual descriptors (often obtained after vector quantization). Before deep learning: most computer vision tasks were relying on feature engineering and bags of words. 95
Image of page 118
Image representation – Deep learning Modern computer vision – Deep learning Deep learning is about learning the feature extraction , instead of designing it yourself. Deep learning requires a lot of data and hacks to fight the curse of dimensionality ( i.e. , reduce complexity and overfitting). 96
Image of page 119

Subscribe to view the full document.

Quick overview of ML algorithms
Image of page 120
Quick overview of ML algorithms What about algorithms? (Source: Michael Walker) 97
Image of page 121

Subscribe to view the full document.

Machine learning – Quick overview of ML algorithms Quick overview of ML algorithms In fact, most of statistical tools are machine learning algorithms.
Image of page 122
Image of page 123

What students are saying

  • Left Quote Icon

    As a current student on this bumpy collegiate pathway, I stumbled upon Course Hero, where I can find study resources for nearly all my courses, get online help from tutors 24/7, and even share my old projects, papers, and lecture notes with other students.

    Student Picture

    Kiran Temple University Fox School of Business ‘17, Course Hero Intern

  • Left Quote Icon

    I cannot even describe how much Course Hero helped me this summer. It’s truly become something I can always rely on and help me. In the end, I was not only able to survive summer classes, but I was able to thrive thanks to Course Hero.

    Student Picture

    Dana University of Pennsylvania ‘17, Course Hero Intern

  • Left Quote Icon

    The ability to access any university’s resources through Course Hero proved invaluable in my case. I was behind on Tulane coursework and actually used UCLA’s materials to help me move forward and get everything together on time.

    Student Picture

    Jill Tulane University ‘16, Course Hero Intern

Ask Expert Tutors You can ask 0 bonus questions You can ask 0 questions (0 expire soon) You can ask 0 questions (will expire )
Answers in as fast as 15 minutes