Text Extraction Algorithm For Images Based On Wavelet Transforms

With the growing number of digital multimedia libraries, the need to efficiently index multimedia information is increased, and content-based multimedia indexing became the hot topic in recent years. If the text information embeded in digital images can be extracted and recognized exactly, it would play an important part in analyzing the information of images and images indexing based on content. On the other hand, the existing techonology also provided ripe techonology foundation for text extraction of images. Thus, how to extract text information of images fast and exactly to provide import for character recognition systems became the focal point of research.There exist texts with different languages, fonts and size in images, and the background of images is very complex, it brought much difficulty to extract text information of images. The goal of this paper is to lacate the text area in images fast and exactly, then transform the text area in binary images. So the text information can be provided to character recognition systems as single word for processing and recogniaztion in the later stages.Owing to the good properties in time-frequency and multiresolution analysis, text extraction of images based on wavelet transforms have cuurently become a well- discussed and important research aspect.The paper studies text extraction models for images based on wavelet transforms, a text extraction algorithm for images in wavelet domain is proposed. For the algorithm, two-dimensional wavelet transform is used to the image firstly and slipping windows are set to scan high-frequency subband, through computing the wavelet texture feature of the image in slipping windows, k -means cluster algorithm is used to extract text area. Then morphology operations and edge detection are put on the text area and original image respectively. According to the character of text area, they are located exactly. At last, text and background are segmented. The experimental result shows that the algorithm can extract texts information with different languages, fonts, size from the background image exactly.