Using Datagram in the K-views Model

The performance of the K-views template (K-views-T) algorithm is related to the size of a view template and the number of characteristic views in the set of characteristic views. If the size of a view template and the number of characteristic views are increased, the classification accuracy will be improved at the expense of the time complexity. To reduce the time complexity of the K-views-T algorithm and maintain the high classification accuracy, the algorithm can utilize the datagram in which the frequencies of characteristic views are cumulated and distributed in a histogram. Due to the use of frequency, a smaller size of the view can be used for maintaining similar classification accuracy. In a sense, this is very similar to the approach used in the LBP and Textural Unit in which a histogram depicting the distribution of all the frequency (i.e., number) for a texture patch is used for the classification. In the basic K-views-T algorithm, the decision is made by a single characteristic view whose center is located at the current pixel being classified. By using this new datagram in the K-views model, the decision is made by the distribution of all the views contained in a large patch (i.e., block) in which the current pixel is the center of the block. Hence, a new K-views datagram algorithm is developed based on the datagram concept. Due to the spatial template used for the view, a boundary-refined method is described to improve the boundary pixel classification.