A new approach to extracting sport highlight

摘要

Quantity and quality of sport highlights attract viewers to sport programs and portal sites. Making a good quality highlight requires an experience production team. It is a job that is very hard to automate because of the quality factor. The machines of the current generation are not intelligent enough to make a sound judgment on what establishes a highlight yet. This work focuses on this difficult problem. It introduces a new technique to automate the making of sport highlight that is of acceptable quality. Previous approaches to extracting the highlights employ image and/or audio processing technology to capture relevant contexts related to highlight moments, e.g., goal frame to hint at the scoring moment. However, the accuracy of such approaches has not been satisfactory. They generate many false positives that may render the highlights useless. To remedy this, we propose a new approach that combines on-line information retrieval with text extraction using optical character recognition (OCR) - a much more tractable problem than general image processing. The new approach makes better highlights because it is free of false positives by design.