CSE 252 C : Computer Vision

Nearest Neighbor (NN) search is a core operation that is performed in many approaches to object recognition. Once images in the database are represented as vectors or histograms (by whatever means), one way to label the unknown data is by comparing the novel images to the database of training examples; if there is a image that is “close,” then it is likely to have the sample label (i.e. be the same object). Although NN is a very simple classification scheme, when coupled with a proper dissimilarity measure, it often outperforms many sophisticated approaches to classification (for instance, SVM or boosting). The main drawback of NN, however, is its complexity, in both time and space. Brute force NN search of a database requires a comparison of the query to all the training