We propose a pairwise local observation-based Naive Bayes (NBPLO) classifier for
image classification. First, we find the salient regions (SRs) and the Keypoints (KPs) as
the local observations. Second, we describe the discriminative pairwise local observations
using Bag-of-features (BoF) histogram. Third, we train the object class models by using
random forest to develop the NBPLO classifier for image classification. The two major
contributions in this paper are multiple pairwise local observations and regression object
class model training for NBPLO classifier. In the experiments, we test our method using
Scene-15 and Caltech-101 database and compare the results with the other methods.
