Research And Application Of Background Filtering Technology Based On Text Detection

This paper studies the text-based text image background filtering technology related issues: corner detection algorithm for text image, text positioning method, image reconstruction, based on summarize and analyze the existing theories, propose this technology solution and achieve the application of the program.Firstly, this paper introduces the text positioning algorithm recent research situation, technology application areas, summarizes the commonly used text positioning algorithm. Analysis and comparison of image edge extraction Sobel, Roberts, Prewitt operator features and applications. Research Harris algorithm, SUSAN algorithm theory, by analysis get the issues of Harris corner detection algorithm needs to be improved.Based on above analysis, put forward a Harris corner detection algorithm optimization solution, image corner detection using Sobel operator instead of the difference operator for image filtering processing, dynamic calculation of the amount of image corner threshold instead of the hand-setgiven threshold to select the corner, by the above two steps to optimize the Harris algorithm. The test results show that the Sobel algorithm ensures a sufficient number of corners, the calculation of the dynamic threshold to improve the scope of application of the Harris algorithm.Secondly, through the analysis of the text corner with the text pixel distribution features, text pixels distribution has regional connectivity, arranged in regular features, the experiments show that text corner also has the above characteristics, according to these characteristics to the text corner denoising, the positioning of the text area, can get more satisfactory results.In the text area position using projection method, through statistical and analysis the number of images each row or column corner, get the text area of image, also use this method to split large text area, this method can be proved by experiments to text location about filter corner image.This article also proposes a reserved original image text color of the image reconstruction method, the method of positioning the text specific coordinate using image filling reconstructed pixel values of the original image, so as to ensure a reconstructed image in the filter background retain the original color of the text.Finally, achieve the technical solutions proposed in this paper, and gave details of the prototype system of the program, after the test results were analyzed to prove that the solution has a good practical value.