Extraction And Recognition Based On K-means And Neural Network Algorithm For Image Text

In recent years, with the rapid development of digital technology, information technology and multimedia technology, there are a large number of digital images, remote sensing images, ad images, signpost signs and microblogging (web) filled with our lives. The images contain a wealth of semantic information, intelligent control systems and information detection and retrieval system which are processing objects. Extraction and recognition of the contents of these pictures is a great help for people’s daily life and work, and also reduces the workload of manual labor. Although domestic image segmentation and recognition technology, in the field of academic and industrial applications has made some achievements, in practical applications, the identification and understanding of computer image text messages still can not meet the requirements of the people.By learning and analysing of previous work, text extraction and recognition include the detection of text area, the segmentation of the text area, text extraction and text recognition, Due to the lower resolution color image in some images itself, more complex background, brightness influence and position, shape and color with uncertainty.for color image text extraction and recognition of the problems, in this dissertotion, a method based on K-means clustering and neural networks is used. Firstly, analyzesthe characteristics of the image, and use the technologies of image analysis, image segmentation, image enhancement to detect the text area of images, and then segment the text area from the color image, magnificate the image of the text area with the wavelet and then use the k-means method to cluster color image to become the text image of single-color background, make the single-color image binarized and the text image slicing, and finally use the method of neural networks to identify the text. In the experimental verification, the method in a certain extent can effectively solve the problems of the character recognition of the image with complex background.