Image Retrieval Application In The Network Of Sensitive Information In Real-time Early Warning System

With the continuous development of Internet technology, there comes more and more modes which are used for information swapping and spreading. For many businesses, they need to keep themselves informed about the sensitive information from network. "Network sensitive information real-time warning system" is a review of network information system, which developed for the business needs, it can extract and analysis text and image information on specify websites. This study belongs to the image search module in the system, it can discriminate official-document images from image folder, and recognize title characters with Chinese character recognition technology, so that the system could identify the title text to match the corresponding sensitive information and do a warning.This paper's description to the official-document image title recognize could be divided into 3 parts:image filtering, title characters extracting, and character recognition. Image filtering is a filter for the wide variety images, in order to filter out some images that do not have official-document image characteristics. The official-document images'characteristics include color distribution, red transverse line, and the image size. They are the judge standard to a official-document. Title characters extracting will extract separated title characters from image, it is the pre-conditions of feature extraction and recognition, this part can be divided into:layout analysis, character segmentation, standardization and refinement several steps. Character recognition is a process of feature extraction and recognition to each title characters, this part includes feature extraction and recognizer design, recognizer also includes coarse classification and word recognition two steps, we use the minimum distance classification for multi-level classification and identification.Appling the image retrieval and recognition methods, which described in this paper, to the system, allowing the system to retrieve text messages, at the same time, also has a function of official-document images retrieval, the applied situation shows that, our methods could retrieval official-document images and recognize the title information.