Database of images for low complexity sign detection and text localization method for mobile applications

Listed in Datasets

By Katherine L. Bouman1, Golnaz Abdollahian2, Mireille Boutin2, Edward J. Delp2

1. University of Michigan, 2. Purdue University

This database contains images that were used for training and testing of our text detection and localization algorithm.

Download Bundle

Version 1.0 - published on 01 Feb 2018 doi:10.4231/R7ZP44BW - cite this Content may change until committed to the archive on 01 Mar 2018

Licensed under Attribution 3.0 Unported

image_and_groundtruth.gif isolating_regions.gif

Description

This database contains images that were used for training and testing of our text detection and localization algorithm presented in “A Low Complexity Sign Detection and Text Localization Method For Mobile Applications.” The contents of this database are as follows:

  • signs-N800-training: This directory contains 81 images of road signs, flyers, and posters. These images (size: 480 by 640) were all taken using a VGA camera on a Nokia N800. These images were used for training our algorithm’s thresholds.
  • signs-N800-training: This directory contains 81 images of road signs, flyers, and posters. These images (size: 480 by 640) were all taken using a VGA camera on a Nokia N800. These images were used for training our algorithm’s thresholds.
  • signs-N800-testing: This directory contains 160 images of road signs, flyers, and posters. These images (size: 480 by 640) were all taken using a VGA camera on a Nokia N800. These images were used for testing our algorithm’s performance.
  • signs-N800-GT1: This directory contains 241 images of ground truth files used to objectively measure the number of false positives and false negatives found in each output image. In this set of ground truth images, each character in the targeted sign region was manually segmented from the rest of the image. The most prominent sign region in the image was manually chosen as the targeted sign region. Each character is a single connected component region separated from any other character’s connected component region.
  • signs-N800-GT2: This directory contains 241 images of ground truth files used to objectively measure the number of false positives and false negatives found in each output image. In this set of ground truth images, each sign region was manually segmented from the rest of the image. All sign regions have been identified, not just the targeted sign region.

The paper goes into more detail about how we use these files to train and test our algorithm’s performance. K. L. Bouman, G. Abdollahian, M. Boutin and E. J. Delp, "A Low Complexity Sign Detection and Text Localization Method for Mobile Applications," in IEEE Transactions on Multimedia, vol. 13, no. 5, pp. 922-934, Oct. 2011. doi: 10.1109/TMM.2011.2154317

Cite this work

Researchers should cite this work as follows:

Tags

Notes

The images (size: 480 by 640) were taken using a VGA camera on a Nokia N800. They are in TIFF format.

The Purdue University Research Repository (PURR) is a university core research facility provided by the Purdue University Libraries, the Office of the Executive Vice President for Research and Partnerships, and Information Technology at Purdue (ITaP).