Coordinate coding on the Riemannian manifold of symmetric positive-definite matrices for image classification

Mehrtash Harandi, Mina Basirat, Brian C. Lovell

Research output: Chapter in Book/Report/Conference proceedingChapter (Book)Researchpeer-review

1 Citation (Scopus)

Abstract

Over the years, coding—in its broadest definition—has proven a crucial step in visual recognition systems [4, 7]. Many techniques have been investigated, such as bag of words [1, 9, 16, 18, 19, 31], sparse coding [21, 34], and locality-based coding[33, 35]. All these techniques follow a similar flow: Given a dictionary of code words, a query is associated to one or multiple dictionary elements with different weights (i.e. let@tokeneonedot, binary or real). These weights, or codes, act as the new representation for the query and serve as input to a classifier (i.e., support vector machine (SVM)) after an optional pooling step.

Original languageEnglish
Title of host publicationRiemannian Computing in Computer Vision
EditorsPavan K. Turaga, Anuj Srivastava
Place of PublicationCham Switzerland
PublisherSpringer
Chapter16
Pages345-361
Number of pages17
ISBN (Electronic)9783319229577
ISBN (Print)9783319229560
DOIs
Publication statusPublished - 2016
Externally publishedYes

Cite this