Synthetic Brazilian Documents Database (SBR-Doc Database)

Research Tasks

Signature Segmentation

2021-08-31 (v. 1)

Contact author

Celso A M Lopes Junior

Universidade de Pernambuco

+55 81 992469364

This work is licensed under a Creative Commons Attribution-NonCommercial-ShareAlike 3.0 Unported License.


Given an image of some document (input data), the model or technique applied should return such as a result an image containing only the pixels of the handwritten signatures, that is, an image with the same size of the input image with the handwritten signatures as foreground (white pixels), and the background (black pixels).

Baseline papers:

Speeding-up the Handwritten Signature Segmentation Process through an Optimized Fully Convolutional Neural Network;

FCN+RL: A Fully Convolutional Network followed by Refinement Layers to Offline Handwritten Signature Segmentation

Task 3



Database division:
Statistics from the dataset and experimental partition used in the Train and Test, where C1, C2, and C3 correspond to the 1st, 2nd, and 3rd Tasks, respectively.

To evaluate the methods, the following similarity metrics will be considered:

Dice Similarity Coefficient (DSC);
Scale Invariant Feature Transform (SIFT);
More information in the paper - Access to paper:
Competition on Components Segmentation Task of Document Photos




No comments on this dataset yet.

Add your comment

In order to comment on a dataset you need to be logged on
Register Now!


In order to rate this dataset you need to be logged on
Register Now!