Scene search based on the adapted triangular regions and soft clustering to improve the effectiveness of the visual-bag-of-words model.