Chinese environment-oriented complex scene text positioning method

Abstract

The invention discloses a Chinese environment-oriented complex scene text positioning method, which mainly solves the problem that scene text positioning in a complex background has a high false alarm rate in the prior art. In the method, the advantages of an edge-based text positioning method and a texture-based text positioning method are combined, and edge features and texture features are combined by stages. The method comprises the following steps of: detecting a text area by using the edge features, namely performing edge extraction and binarization processing on a down-sampled color image, connecting the edges of characters to form a block through morphological operation, extracting the features of each communicated domain, and removing a great number of non-character communicated domains by using a cascaded threshold classifier to acquire candidate character communicated domains; and extracting the texture features of the candidate character communicated domains, and further judging whether the candidate character communicated domains are character communicated domains or not by using a back propagation (BP) network classifier. The method has the advantages of high text positioning accuracy and high speed under the conditions of complex illumination and backgrounds, and can be used for text automatic extraction and identification in a complex scene.

Description

technical field [0001] The invention belongs to the technical field of image processing, relates to an image text positioning method, and can be used for automatic text extraction and recognition. Background technique [0002] Natural scenes contain rich text information, such as shop signs, road signs and various warning or prompt signs, etc. If people can extract these information and do further processing, it will have a broad application prospect. For example, the text can be recognized by the optical character recognition OCR module, which can realize the retrieval of images or videos; or the text-to-speech TTS module can be used to recover the text in the form of sound, which will bring great benefits to international tourists and the blind. Great convenience. Under various demands, extracting text information from images has become a very practical topic. The text positioning is the premise of correct text information extraction, and it is the key part of the whole ...

AI Technical Summary

Problems solved by technology

The positioning method based on the connected domain is more accurate, especially for larger characters, and the speed is faster, but when the text background is complex, some text-like objects are difficult to be classified; at the same time, how to select an appropriate threshold for character connectivity Domain confirmation is also a difficult point

The positioning method based on edge detection has good positioning for images with clear text background or dense text, and the calculation speed is relatively fast; but when there are many strong edge objects in the image, it often produces a high false alarm rate , with low accuracy

[0005] The texture-based positioning method considers text as a special texture. This method can extract texts of different sizes, different languages ​​and different fonts in different resolution images. It has certain versatility, but it has a large amount of calculation and positioning accuracy. not high disadvantage

[0006] From the above analysis, it can be seen that any one of the above methods alone cannot achieve satisfactory results.

Aiming at the complexity of natural scenes, researchers from various countries have made various attempts and researches. Although they have achieved certain results, most of the methods are aimed at scene images with low complexity of video text or background, and cannot adapt to illumination. Inhomogeneity, low contrast and complex background, etc., the localization method is less robust

Images