Abstract:

Given that the bag of words model is quite sensitive to background noise and that visual words in the background are not relevant to objects, we propose an object recognition method which combines saliency detection with the bag of words model. Firstly, the region of interest from the original image is adaptively gained by using the graph-based visual saliency (GBVS) algorithm and the AC algorithm. The combination of the two detection methods can avoid incomplete region of interest. Secondly, we extract local features from the region of interest by using the scale invariant feature transform (SIFT) descriptor. Then, we use the peak density clustering algorithm to classify the features and generate a visual dictionary histogram by clustering local features. Finally, we employ the support vector machine (SVM) classifier to classify and recognize objects. Experiments on PASCAL 2007 and MSRC-21 databases verify the effectiveness of this method. Experimental results show that the proposed method can effectively improve the performance of object recognition.

Key words: bag of words model, saliency detection, density peak clustering, object recognition