Gaussian process learning for image classification based on low-level features

Recently, Gaussian Process for Machine Learning (GPML) has received increasing attention in the machine learning community. In this paper, a method implementing GPML for image classification is proposed. This algorithm uses low-level image features that can be easily and quickly extracted. The proposed algorithm is tested on the well-known object-category data sets (Caltech 256) and is compared with Least Squares Support Vector Machines (LSSVM). The major contributions of this paper is that it proposes a feasible framework to implement GPML for image classification and introduces a novel color feature extraction procedure based on color coherence vector, which is suitable for supervised learning. Influence of different low-level features on GPML and LSSVM is also investigated in the experiments.