Generalized Super-Vector Coding for Image Classification

DOI

10.2991/aiie-15.2015.95How to use a DOI?

Keywords

image classification; SIFT; bag-of-features; super-vector coding; support vector machine

Abstract

Semantic understanding of images remains an important research challenge in machine intelligence and statistical learning. It mainly includes two steps: feature extraction and classification. This study mainly aims to explore a generalized feature extraction framework motivated by the popularly used Bag-of-feature (BOF) and super-vector coding using local descriptor, which is intuitively time –consuming for computation. In the other hand, the effortless on only exploring color and edge histogram with uniformly quantized space, which are conventional statistics, make less progress in image understanding field. Therefore, This study investigates a generalized framework based on the accessible color or edge information via adaptively modelling the explored space of a specific application, and then extracts the representation statistics (histogram of the data-driven model) and deviation statistics (the statistics of reconstruction error) for image representation. Compared to the uniformly quantized strategy such as the conventional histogram, the proposed framework can represent the image more faithful and compact, and then lead to more discriminant representation for images. With the extracted data-driven statistics, a simple linear support vector machine (SVM), which is especially efficient for large-scale database, can be effectively utilized for achieving acceptable recognition performances. Experiments on two databases: SIMPLICity and OMRON validate that our proposed strategy can achieve much better recognition performances than the conventional and the state-of-the art methods.

Copyright

© 2015, the Authors. Published by Atlantis Press.

Open Access

This is an open access article distributed under the CC BY-NC license (http://creativecommons.org/licenses/by-nc/4.0/).

Download article (PDF)

TY  - CONF
AU  - M. Nakajima
AU  - Y.W. Chen
AU  - X.H. Han
PY  - 2015/07
DA  - 2015/07
TI  - Generalized Super-Vector Coding for Image Classification
BT  - Proceedings of the 2015 International Conference on Artificial Intelligence and Industrial Engineering
PB  - Atlantis Press
SP  - 341
EP  - 344
SN  - 1951-6851
UR  - https://doi.org/10.2991/aiie-15.2015.95
DO  - 10.2991/aiie-15.2015.95
ID  - Nakajima2015/07
ER  -