Research On News Video Caption Extraction Based On Corner Density Detection And Two Times Binary

In recent years, with the rapid development of Internet technology and video technology research, there are more and more video data on the network, so it is difficult for someone to find the video that who interested in vast amounts of video data accurately, so the video search technology came into being, the researchers propose a way for video retrieval:Content-based Video Retrieval (Content-Based Video Retrieval, CBVR). The basic idea of CBVR is extracted the characteristics of video data to make up the index which describe the content of video data, but some underlying information in video image such as image texture, color can not describe the content of video accurately, so the researchers tend to focus on high-level semantic of video step by step. Video caption includes most high-level semantic information in video and these information can describe the main content of video accurately, so if we can extract the video caption effectively, it will be of great help for video retrieval. However, there is low resolution in video generally, and video caption always embed in complex background image, so there is no way to use OCR to identify the video image caption directly, it is necessary to overcome this difficulty before extracting the video caption accurately.This paper reviews the development video caption extracting technology, and analysis the current method of video caption extracting, to resolve the difficulty of video caption extracting, this paper present some new method for video caption extracting:First is the way of video caption detection that based on the density of corner. This method uses the feature that there are more corner information in caption area in the video image, first detect the corner information in the image, and then select the intensive corner area to be the candidate caption regions, and finally accordingly to some heuristic rules filter out some disturbance area to get the final caption area. Second is the method of second binarization that based on their own charactetristics. The general method for image binary is global threshold and local threshold method, but whether it is the global threshold or the local threshold, the result of binary is relevant about the threshold, and the threshold will affect the binary results. This paper presents a method of second binary that based on their own strock and color charactetristics,first get the stroke diagram by using caption stroke characteristic, complete the first binary,then using the color information filters noise in the stroke diagram to get the final result of binary.