Text Detection In Natural Scene Images

With the development of science and technology, human beings haveentered the information age, which represented by electronic computer science andtechnology is quietly changing people’s life. The integration of text information anddigital multimedia technology is no longer single presented, which enriched our lifeand gave us more choice space at the same time.In this dissertation, two questions we have mainly considered are the cornerpoint extraction for small text and the shortage of large text detection in textpositioning. This dissertation use the improved Harris corner detection method andthe stroke extraction method respectively to solve the problem that detect the smallcharacters in common corner detection. Meanwhile, the correct positioning of thecharacters is also achieved by the method of extracting the strokes.For small characters text, we use methods of the improved Harris cornerdetection method and morphological processing to detection text. So we get the dataof aspect ratio and area ratio, then we take them as the characteristics to establishbinary linear regression equation and get the solution of equations of the parameters.For the text area of insufficient diagonal point detection and too closed cornerdetection text area, this dissertation analyzes their numerical regularity of ratio withaspect ratio and area ratio, adding different constants in different cases,and as theresult we improve the detection equation.Using large character text stroke of consistent width and the characteristics of thevertical distance near the center point to the two edge, this paper complete the strokeextraction and subsequent stroke selection and region growing by using distancetransformation of distance and direction information. At the same time we removingpseudo strokes, I am more likely to finish filling inadequate stroke the edge to localthe great character text and more conducive recognize optical character.Finally, comparing with other literatures, this paper objectively analyzes theadvantages and disadvantages of each method,it summarizes the existing problemsand shortcomings which as a reference of research and experimentation in the feature.