摘要 |
A method is described that greatly reduces the computational cost of determining the globally optimal match between a template and one or more samples. For every sample in a search range, intermediate distance measures between the template and each sample are first computed in one designated dimension. The computed distance measures are then sorted according to their magnitude. The sample with the minimal distance measure is selected to accumulate a new distance measure in the next higher dimension. This new distance measure is recorded and the samples are sorted again according to the updated distance measures. The above process is repeated until a minimal distance measure has been computed in all dimensions. For motion estimate, this method can reduce the number of computational operations by about 90% to 99%. This method can also provide a list of globally best matches. By sacrificing the global optimality, this method can satisfy the time constraint required by some applications and provide a suboptimal match.
|