Research On Scene Image Text Detection Method Based On Sparse Representation And Morphological Component Analysis

With the rapid development of Internet and smart phones,image and video information has increased dramatically.Image retrieval and image classification have become a hot research point in the field of computer vision.As the images and video often contain a lot of text,these text usually carries abundant semantic information,which makes text detection and text recognition are of great significance in image retrieval and classification.For text images,text recognition is heavily dependent on the results of text area detection,so text detection techniques attract a large number of researchers.However,due to variations of text and complexity of background,detecting text in scene images becomes a critical and challenging task.In view of these problems,this paper study the problems and obtain the following research results:(1)The effect of complex background and noise interference is studied in this paper,and a method of text detection based on sparse representation and morphological analysis on the scene image is proposed.The method first uses morphological analysis and sparse representation to study the discriminant text and background dictionary,and then the text part of the image to be detected is reconstructed by the learned dictionary,and finally performs text detection on the reconstructed image.This method translates text detection problem into sparse and robust representation by the complete dictionary.(2)In this paper,a method of on-line discriminant dictionary learning is proposed for text detection on scene image for the shortcomings of low dictionary learning speed and lack of adaptability in the method of text detection based on sparse representation and morphological analysis on the scene image.In view of these problems,this paper studies the method of online discriminant dictionary learning for text detection on scene images,reduces the computational complexity and improves the adaptability and discriminability of the dictionary.(3)In this paper,the experiments on ICDAR2003/2011/2013 and MSRA-TD500 are carried out.It is found that the proposed method has better robustness in scene image text detection compared with the popular method.The two proposed methods are also compared in this paper.The method of on-line discriminant dictionary learning is superior to the first dictionasry method,whether in dictionary learning accuracy or detection effect,have achieved good results.