lec7 prelim

lec7 prelim - Lecture 7: Sept 15, 10 HW2 posted? Last Class...

Info iconThis preview shows pages 1–5. Sign up to view the full content.

View Full Document Right Arrow Icon
USC CS574: Computer Vision, Fall 2010 Copyright 2010, by R. Nevatia 1 Lecture 7: Sept 15, 10 • HW2 posted? • Last Class – Mean Shift Segmentation – Normalized Cuts • Today’s objective – Tutorial on OpenCV – Energy Minimization Methods – Edge Detection
Background image of page 1

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
USC CS574: Computer Vision, Fall 2010 Copyright 2010, by R. Nevatia 2 Energy Minimization Methods • General Idea – Task is to label each pixel as being foreground or background – Label depends not only on the property of the pixel (e.g. intensity, color or some other features of the neighborhood) but also on the labels of the pixels in the neighborhood for smoothness • Commonly expressed as the task of minimizing an energy function • Data and loss functions need to be defined by the designer – For example, data energy can be the absolute difference between observed intensity and expected intensity – Smoothness energy can be defined as being sum of terms for each neighbor considered (common to use four neighbors). For each neighbor, for example, the energy term is zero if the two neighbors have the same label, and another value if they are different. This value may be different for different labels (background or foreground) and for different neighbors.
Background image of page 2
USC CS574: Computer Vision, Fall 2010 Copyright 2010, by R. Nevatia 3 Optimizing the Energy Function • In general, the energy function is not convex and may have multiple minima; thus finding a global minimum may be combinatorial. • For certain class of functions, global minimum can be found by computing Max_Flow/Min_Cut in a graph constructed from the energy function – Details in posted tutorial (borrowed from Niethammer) – Details of the Max Flow algorithm are for information only; they will not be tested on in exams or assignments.
Background image of page 3

Info iconThis preview has intentionally blurred sections. Sign up to view the full version.

View Full DocumentRight Arrow Icon
USC CS574: Computer Vision, Fall 2010 Copyright 2010, by R. Nevatia 4 Comments on Optimization Described method is of the class of “augmented path” methods. Empirical studies show that, for the types of energy functions commonly used in
Background image of page 4
Image of page 5
This is the end of the preview. Sign up to access the rest of the document.

Page1 / 16

lec7 prelim - Lecture 7: Sept 15, 10 HW2 posted? Last Class...

This preview shows document pages 1 - 5. Sign up to view the full document.

View Full Document Right Arrow Icon
Ask a homework question - tutors are online