Vocabulary Tree algorithm builds a tree structure by performing the off-line learning method using the large number of training image dataset. After constructing, we can retrieve a query image class very quickly by searching the tree structure. Becaus...
Vocabulary Tree algorithm builds a tree structure by performing the off-line learning method using the large number of training image dataset. After constructing, we can retrieve a query image class very quickly by searching the tree structure. Because of the great improvement for computation time reduction, this algorithm is widely applied for other object classification researches.
In this paper, we suggest a method which improves the retrieval performance via searching our tree with multiple times per one test data. The information which given by pre-matching array determines how the tree is visitied. Taking our new algorithm, we can reduce miss-classification rate considerably. On the other hand, disadvantages from computation time and memory allocation are not critical with respect to that of training time.