a
   
     
Bloom's taxonomy
Please Note > This page will be under construction throughout the semester.
 


Syllabus

Unless otherwise stated, page numbers in the readings column below refer to the required textbook, C. M. Bishop, Pattern Recognition and Machine Learning, Springer 2006. Additional material will be handed out in class or posted on the table below as appropriate.

Note: papers in this table are linked through their DOI, so the link will work for you only if you or your institution have proper access privileges. For Duke students, this typically means that the link will work from a Duke computer, but not from elsewhere.
Module
Description
Readings
Software and Data
Introduction Purpose, state of the art, issues, concepts Introductory slides of the CVPR 2007 course on recognition and learning  
Basic concepts of machine learning and pattern classification A conceptual framework and some toy examples pp. 1-66 Matlab code for univariate and naive DP classifier (updated on 9/8/08)
Elements of image formation and processing We dig into the SIFT operator, both to understand key image processing concepts and methods, and learn two useful algorithms. Lowe's paper on SIFT. Notes on image formation: pp. 1.5-1.7, 1.8(bottom)-1.9, 1.15-1.21. Notes on image processing (all). Andrea Vedaldi's (UCLA) implementation of the SIFT operator.
More invariance than SIFT Affine-invariant regions The Mikolajczyk and Schmid paper. The Wikipedia entry on the Harris detector.  
First example of object recognition Bags of features The Sivic and Zissermann and Nister and Stewenius papers  
Segmentation Felzenszwalb's segmentation algorithm The paper by Felzenszwalb and Huttenlocher on segmentation. Pedro Felzenszwalb's segmentation code. [Compilation tips]
Maximally stable regions A more data-driven way to extract regions of interest The paper by Matas et al Andrea Vedaldi's (UCLA) implementation of the maximally stable region detector
Factorization and clustering method for defining "visual words" PCA, EM, LSA, and non-negative matrix factorization to approximate data Background on EM and LSA (a.k.a. PCA): Ch. 9 and 12. A paper by Lee and Seung on part detection with NMF.  
Unsupervised learning with "bags of features" EM, Probabilistic Latent Semantic Analysis (pLSA) and Nonnegative Matrix Factorization (NMF) to approximate count matrices Hofmann's paper on pLSA for text, and a computer vision version for images.  
Weakly supervised learning with constellation models Generative models of shape and appearance A paper by Fergus et al.  
Supervised learning with templates Recognition without parts Two papers by Viola et al on detecting faces in images and pedestrians in video The Intel Open Source Computer Vision Library has code for boosting, Haar descriptors