A Novel Feature Extraction Method for Scene Recognition Based on Centered Convolutional Restricted Boltzmann Machines

摘要

Scene recognition is an important research topic in computer vision, whilefeature extraction is a key step of object recognition. Although classicalRestricted Boltzmann machines (RBM) can efficiently represent complicated data,it is hard to handle large images due to its complexity in computation. In thispaper, a novel feature extraction method, named Centered ConvolutionalRestricted Boltzmann Machines (CCRBM), is proposed for scene recognition. Theproposed model is an improved Convolutional Restricted Boltzmann Machines(CRBM) by introducing centered factors in its learning strategy to reduce thesource of instabilities. First, the visible units of the network are redefinedusing centered factors. Then, the hidden units are learned with a modifiedenergy function by utilizing a distribution function, and the visible units arereconstructed using the learned hidden units. In order to achieve bettergenerative ability, the Centered Convolutional Deep Belief Networks (CCDBN) istrained in a greedy layer-wise way. Finally, a softmax regression isincorporated for scene recognition. Extensive experimental evaluations usingnatural scenes, MIT-indoor scenes, and Caltech 101 datasets show that theproposed approach performs better than other counterparts in terms ofstability, generalization, and discrimination. The CCDBN model is more suitablefor natural scene image recognition by virtue of convolutional property.