Text detection in images using sparse representation with discriminative dictionaries

a r t i c l e i n f o Text detection is important in the retrieval of texts from digital pictures, video databases and webpages. However, it can be very challenging since the text is often embedded in a complex background. In this paper, we propose a classification-based algorithm for text detection using a sparse representation with discriminative dictionaries. First, the edges are detected by the wavelet transform and scanned into patches by a sliding window. Then, candidate text areas are obtained by applying a simple classification procedure using two learned discriminative dictionaries. Finally, the adaptive run-length smoothing algorithm and projection profile analysis are used to further refine the candidate text areas. The proposed method is evaluated on the Microsoft common test set, the ICDAR 2003 text locating set, and an image set collected from the web. Extensive experiments show that the proposed method can effectively detect texts of various sizes, fonts and colors from images and videos. With the rapid development of digital devices, images and videos are now popular media in our daily lives. Texts, which are often embedded in images and videos, contain lots of semantic information useful for video comprehension. They can thus play an important role in content-based multimedia indexing and retrieval. In recent years, the automatic detection of texts from images and videos has gained increasing attention. However, the large variations in text fonts, colors, styles, and sizes, as well as the low contrast between the text and the often complicated background, often make text detection extremely challenging. A lot of efforts have been put on addressing these problems [1–11], and these can be roughly divided into four categories. The first category uses connected component analysis (CCA) [1], in which pixels with similar colors are grouped into connected components, and then into text regions. CCA is fast. However, it fails when the texts are not homogeneous and text parts are not dominant in the image. The second category is based on edges [2], which assume high-contrast differences between the text and background. It is fast, and can have a high recall. However, it often produces many false alarms since the background may also have strong edges similar to the text. The third category is based on textures [3], and assumes that texts have specific texture patterns. It is more time-consuming and can fail when the background is cluttered with text. The fourth …