Text Localization in the Image of Complex Background using Discrete Wavelet Transform

Objectives: To propose suitable method for localizing text in images. Methods/Statistical Analysis: Many methods of text localization use techniques based on region and textures using statistical approaches. In this paper, a domain transform based method has been proposed. The proposed approach based on wavelet transform utilizes the properties of wavelet in localizing the text. Particularly, it uses wavelet energy in differentiating text region and non text region in an image. Findings: The proposed approach was applied on many bench marked images to localize the text and found effective in achieving its purpose. Many existing methods of text localization based on region and texture, need special software or complex calculations and methods. Some of these methods require the heuristic functions like threshold and other inputs. Also, some methods are suitable for small text while some of them are suitable for large text. The proposed wavelet based method for text localization works well for small as well large text. It doesn’t need any special software. It doesn’t have any input constraints like threshold. Also it is quite simple, robust and more efficient in comparison with the existing approaches. Applications: Localizing text in images contributes much in the extraction as well as retrieval of information from an image in the web. Also, it is an important step in recognizing the text present in the image.