The Research Of Text-Region Extraction Based On Color Image

With the rise of digital libraries and the development of multimedia information retrieve and Internet technology, extracting the text from color image has been attached importance to more and more, and applied to the information management, such as, Web search, automatic processing of complex documents and various image database retrieve. The text embedded in color image has significant meaning, which usually forms brief description or explanation to the relative image. Recognizing and analyzing the text regions which are automatically located and exacted from image are very helpful for understanding , indexing and retrieving image.To the text in color image, the arrangement of the characters is sequential and the color of the font is basically same. The most importance is there exists distinct edge contour between the characters and the background. So, when exacting the text from color images, it is very important to use the above-mentioned specialty of the text.In this thesis , CEDO(Color Edge Detection Operator)operator is given on the basis of modifying the color edge detection operator,which is provided in literature [11]. CEDO is a color edge detection operator which operates on the overall three channels of RGB color space. Firstly it extents the edge detection operator of gray image-Roberts edge detection operator to edge detection of color image, considering the three channels of every pixel synthetically byusing Euclida distance, and obtains a gray-level edge image I1. Then, uses Laplace edge detection operator to detect I1, and obtains a gray-level edge image I2. On the base of I2, the suitable threshold is found, and thegray-level edge image I1 is converted to binary edge image with thisthreshold. Consequently, the edge extraction of color image is completed.Using the feature: there exists distinct edge contour between the characters and the background in color image, we give a color image text extraction algorithm-TEA(Text Extraction Algorithm), which makes the use of CEDO toprocess edge detection of color image, then, adopts morphologic dilation technology to form region image, and combines the area-growth algorithm to mark the candidate text blocks, finally, determinates the correct text block by analyzing the density of text block and the space frequency feature. So the text exaction of color image is preferably accomplishedIn this thesis, after intensive description of text extraction from color image, the system based on a simplification model is investigated. It includes the detail processing: image preprocess, color edge detection, threshold calculation, binary processing, candidate text region formation and extraction. Experiments verify the validation of TEA and CEDO.