Purpose of this study is to automatically segment goods catalog videos into individual goods section and to make it feasible to retrieve them when inquired. In this paper, we propose utilization of video captions superimposed on the video image to identify the goods names and perform the video segmentation. The video OCR software we propose in this paper is composed is video caption frame extraction, character region extraction, character binarization, character recognition and goods name identification. The individual goods section is segmented when a new name of commercial goods is recognized. Our software focuses on the binarization method that employs a floating adaptive three level threshold method.
展开▼
