The Research Of Network Image Text Region Location Technology

The combination of Computer technology and Internet technology's development quickly form a unique network culture. A large number of digital information transmitted via internet brings a tremendous impact to people's daily life and work. On one side, such as book covers, envelope covers and other useful images requires computer automatically search for relevant text content, so as to release people's labor and achieve the demand of the digital office. Variant of indecent images transmitted via the internet needs to be inspected and filtered timely to eliminate the negative impact of internet communication. Text information is the most important in image. So, based on the image text's information retrieval comes into being naturally. Existing character recognition technology is well established, as long as accurately locate the text, the text can be separated and identified. So locating the text in the network image is the primary important thing.The main work involves how to detect, locate and determine the text region from the network image on all aspects and steps. Research work has focused on text's detection, location and discrimination from network image and the associated algorithm. There are no uniform norms and standards about network picture, therefore, the text also contains a variety of forms. The accuracy and applicability of the algorithm will be affected by text's different size, arrangement, language, color, brightness, background's complexity and the resolution of the network image itself. All these factors gave the location technology of text in network images more difficulties and challenges.On the aspect of text detection, this paper contrast the different algorithm based on connected component, edge detection, texture learning and machine learning. Finally, it takes the corner as the main feature of the text, and proposes a algorithm based on corner and Gaussian pyramid for automatic text detection and localization. This algorithm can automatically detect different sizes of text, which appears large, medium and small in one same network image. In order to achieve high recall rate, improve the integral projection method, proposed staging idea to locate different text region in the same picture. In order to improve search accuracy, using a mixture of characteristics of text-based heuristic rules. In the performance evaluation, mainly adopt recall rate, precision, and running time of evaluation criteria.The proposed algorithm is designed and completed through a series of literature-reading and the experimental contrast. Experimental results show that this algorithm can quickly and accurately locate the text regions in the network image, which includes covers of books and CD album transmitted via internet. The proposed algorithm with text segmentation, character recognition system can complete the text extraction and recognition of the network image, as to extract useful information while filtering bad information.