Detection of Inserted Text in Images

Part of the Mathematics for Industry book series (MFI, volume 4)


It is possible to embed text into images using an image-editing software, and this can be used to make misleading or unfounded claims in advertisements, which do not comply with advertising standards. To monitor the large volume of advertising that now exists on the Internet, it is desirable to automatically detect and read the text inserted into images. Here we describe a technique to determine regions of images corresponding to inserted text using the FAST algorithm, by finding corners in the image that lie along a straight line, which we term a supercorner. We then create a graph by connecting supercorners and apply cost functions describing the geometrical relations between the corners. Using a graph cut algorithm, we can separate the text from the background. Using this method in a sample set of 130 images with inserted text, we were able to detect 81 % of the inserted text with a false positive rate of only 4 %.


Supercorner Fast Graph cut Inserted text Energy function Edge cost 


  1. 1.
    Boykov Y, Veksler O, Zabih R (2001) Fast approximate energy minimization via graph cuts. In: IEEE Transactions on PAMI, pp 1222–1239Google Scholar
  2. 2.
    Gopalan C, Mamjula D (2008) Contourlet based approach for text identification and extraction from heterogeneous textual images. Int J Comput Sci Eng 2(4):202–211Google Scholar
  3. 3.
    Rosten E, Drummond T (2005) Fusing points and lines for high performance tracking. In: 10th IEEE international conference on computer vision, 2005. ICCV 2005, vol 2, pp 1508–1511Google Scholar
  4. 4.
    Gatos B, Pratikakis I, Kepene K, Perantonis SJ (2005) Text detection in indoor/outdoor scene images. IEEE Trans Pattern Anal Mach Intell 127–132Google Scholar
  5. 5.
    Kolmogorov V, Zabih R (2004) What energy functions can be minimized via graph cuts? In: 1st international workshop on camera-based document analysis and recognition, vol 26, no 2, pp 147–159Google Scholar
  6. 6.
    Rosten E, Drummond T (2006) Machine learning for high-speed corner detection. In: European conference on computer vision, pp 430–443Google Scholar
  7. 7.
    Wayne K (2004) Max flow, min cut. In: Algorithms and data structures, Princeton University, COS 226Google Scholar
  8. 8.
    Zhao X, Lin K-H, Liu Y, Huang TS (2011) Text from corners: a novel approach to detect text and caption in videos. IEEE Trans Image Process 20(3):790–799CrossRefMathSciNetGoogle Scholar

Copyright information

© Springer Japan 2014

Authors and Affiliations

  1. 1.The University of Electro-CommunicationsChofuJapan
  2. 2.The University of Electro-Communications/JST CRESTChofuJapan

Personalised recommendations