Abstract:

A novel approach based on the edge improved center symmetric local binary pattern (EICS LBP) and the statistical domain color pairs of edge as visual features combined with the extended probabilistic latent semantic analysis (PLSA) model for scene classification is presented. First, the features are extracted from edge dense sampling regions as visual words, and then these visual words are formed by clustering respectively.  After that, the bag of words model is used to represent the image. And then, the potential semantic is excavated by the extended PLSA model. Finally, the confusion matrix is obtained by K nearest neighbors (KNN) classifier. Experiment results show that this method achieves higher accuracies, especially performs well in the color images with much edge contours and also it does not require experts to annotate the scene content in advance.