For the robotics scene recognition problem, the relationship between continuous scenes can be modeled by a
hidden Markov model based context model. Unlike traditionally used generative method to learn this model, a sparse
Bayesian learning machine is adopted to model the posterior probabilities of image features. Then by combining the sparse
model with the hidden Markov model using Bayes theory, a discriminative learning method of the context scene recognition
model is proposed. The experiments on a real scene database show that the obtained scene recognition system possesses
good recognition performance and generalization ability.