Image Classification Using a Bigram Model

Rong Jin, Alexander G. Hauptmann. and Rong Yan

Feature representation and classification algorithm are two important aspects for the task of image classification. Previous studies focused on using color histogram and/or texture as the features for representing an image. Support vector machines (SVM) have been among the most successful classification algorithms for image classification. In this paper, we examine a new type of representational feature, namely the "bigram" feature, which computes a distribution for pixel pairs. Unlike color histogram based features, which treats each pixel as independent from others, the bigram feature scheme is able to take the correlations between pairs of pixels into account. In experiments over six different image categories, the bigram feature scheme appeared to be a better representation for image classification and achieved a better classification accuracy than either color histogram features, texture features or color correlograms.


This page is copyrighted by AAAI. All rights reserved. Your use of this site constitutes acceptance of all of AAAI's terms and conditions and privacy policy.