Character Recognition Of Complex Background Color Images

Made from one of the most appropriate concept to start now, OCR (Optical Character Recognition, Optical Character Recognition) technology has experienced nearly a century, and now, for the simple text background can be efficient and accurate computer can be read into understand the electronic text. With the development of increasingly sophisticated technology, application of more and more widely, increasing market demand, a variety of character recognition software and tools has been introduced. However, the classic character recognition technology is only obtained by scanning the background for a simple, high image resolution and contrast are good recognition rate. However, in real life there are many scenes with text, such as road signs, bus stop, name, Description and so on, want to get the text in a natural scene, apparently relying on the way scanning is not practical. At present, although there are many natural scenes for the recognition of the Chinese, but the results are not as expected in the ideal.Usually obtained by the image capture device into the picture with natural scenes and text images. As the complexity of natural scenes, resulting in one of the text background is quite complex, but also because the location, lighting, and other causes are the text font, size, brightness, contrast and so are not ideal, which makes growth difficult position the text area , directly affecting the text area extraction and character recognition results.This paper studies the hardware devices in a limited grasp of the various scenarios under the conditions of the text under the images. How to effectively capture devices from simple access to the complex background color image text extraction and recognition of great significance. This study not only enrich the theory of image processing, but also in life, such as images and video for the Internet environment, search, traffic management on the identification plate, the digital library literature, the blind and other travel information for all areas of the text , there is a great application and commercial value. So, this also makes it has become a hot research topic internationally.Complex color images in the context of the text information into the computer can be recognized and processed text messages. Including image preprocessing, the text area location and extraction, character recognition three parts. To color images under complex background to recognize the text, the text must first locate a region, then part of them to identify the characters. In this paper the main text by examining the existing location and identification, to analyze the advantages and disadvantages, Gabor filter proposed joint marginal density of the text region detection feature extraction, the use of stroke direction based on the statistical characteristics of the characters in feature extraction and recognition.Regional location in the text part, Gabor filtering method of the joint marginal density detection using Gabor filter or separate edge density test results alone are not ideal, because the individual character of each of them will still be confused with non-text area, but their characteristics After making these features combine to complement each other, can clearly distinguish the text area. Positioning the text area, the text of which do binarization, skew correction, text segmentation, normalized, after a single quantitative standard size character images, then use the following directions based on the statistical characteristics of stroke locate the characters in the text area to feature extraction and recognition, and achieved good results.Experimental results show that the proposed text location method can more accurately determine the complex color images in the context of the text area and locate the basic integrity of the text, locate the area of the text basically can be identified. Has some theoretical and practical value.