Design And Implementation Of Text Location Algorithm Based On Combination Of Several Algorithms

Text location in natural scene is a very difficult and challenging research topic. The pictures of natural scene contain a lot of useful information, such as store names, street names, traffic signs and so on. The text information accurate got from the pictures of nature scene is an important research content in digital image processing. To extract the text in the pictures of natural scene, first of all to find the text of the area, and then handed over to the text recognition module OCR can be identified. Although there are many researchers on the natural scene text recognition to do an in-depth study, but in view of the current state of development, natural scene text recognition accuracy far below the desired effect.In this paper, we conduct an exhaustive survey of the main text location methods and the current challenges facing the institute, categorize them, and analyze the text location methods advantages and disadvantages, and proposed the text location algorithm based on the integration of multi-methods. It considers not only the edge and shape of texts, but also the color information. The algorithm takes full advantage of the advantages of three methods: the method based on the edge, the method based on learning, the method based on region.The algorithm designed in this paper includes the pre-treatment, pyramid decomposition, edge detection, morphological operations, the restrictions of priori knowledge, Synthesis of the sub-images result, alternative text area extraction, neural network classification and connectivity analysis of nine steps in the region. First, the use of color edge detection methods on pyramid decomposition of sub-graph edge extraction, then the text Location with mathematical morphology, and thus get an alternative set of text area, and then set this option to mark the text area is divided into the text region and non-text region, as a neural network to learn the training set,In the edge extraction stage, in order to meet the specific requirements of the text region location, this paper proposed a CROstu Color edge detection method based on analysis of classical edge detection operator. The text areas are more clearly after this approach, and the characters have a good shape. In the neural network classification stage, we use the pixel RGB values of the "mi" shape in the square region as input features and a pixel point with using the BP network will be divided into a text pixel or non-text pixel. The approach can avoid the complex stage of constructing and the threshold selection in the classification procedures. The experimental results show that the proposed text location algorithm can not only more accurate locate the corresponding text region, but also has some theoretical significance and greater application value.