Training and evaluation datasets

The original signature images are scanned at 600dpi and have 24bit RGB-color. We will also provide 600dpi grayscale (8bit) and binary formats, as well as 300dpi color, grayscale and binary formats.

The offline training dataset will be constituted of 1920 images, green ink on white paper (true color 24bit data, grayscale and binary data all in 600dpi and 300dpi, PNG format). The online training dataset will consist of 1905 ASCII-files with the format: pen tip positions X and Y, pressure Z, pen tilt (0-90 degrees) and azimuth (0-360 degrees). Sampling rate was 200 Hz. Filenames will reveal the true score.

The offline evaluation dataset will consist of 1953 images, black ink on white paper. The online evaluation dataset will consist of 1953 ASCII-files in the same format as described for the training data with similar sampling rate. A detailed description of the datasets will be provided in README files to the competitors.

Download training datasets

A preview of the NISDCC-signature data is available as well as a detailed description of the NISDCC data set. In order to download the data please register for the competition first and receive the login information (username and password.)

• NISDCC Signature Preview and NISDCC Readme.

• NISDCC Online Signatures
  • all-in-one file (download)

• NISDCC Offline Signatures
  • unprocessed images: 24bit RGB-color: 600dpi
    (download: part1, part2, part3, part4, part5, part6, part7)

• NISDCC Preprocessed Offline Signatures
  • binarized: 600dpi / 300dpi
  • 8bit grayscale: 600dpi / 300dpi
  • 24bit RGB-color: 600dpi / 300dpi
    
    

Download evaluation readme

The readme of the evaluation set is available now, so participants are able to see what kind of data their systems will be evaluated with: Evaluationset Readme.