Publication: Sparse Radial Sampling LBP for Writer Identification
Authors: Anguelos Nicolaou, Andrew D. Bagdanov, Marcus Liwicki, and Dimosthenis Karatzas
Published in: ICDAR 2015 (oral presentation)
In this paper we present the use of Sparse Radial Sampling Local Binary Patterns, a variant of Local Binary Patterns (LBP) for text-as-texture classification. By adapting and extending the standard LBP operator to the particularities of text we get a generic text-as-texture classification scheme and apply it to writer identification. In experiments on CVL and ICDAR 2013 datasets, the proposed feature-set and a simple end-to-end pipeline demonstrate State-Of-the-Art (SOA) performance. Among the SOA, the proposed method is the only one that is based on dense extraction of a single local feature descriptor. This makes it fast and applicable at the earliest stages in a DIA pipeline without the need for segmentation, binarization, or extraction of multiple features.
Experimental data and reproducabillity:
In order to allow for the reproducabillity of most experiments described in the article, SRS-LBP histograms on the most recent datasets are provided in csv format.
- Precomputed SRS-LBP features on several datasets:
- ICDAR 2013 Writer Identification Dataset, precomputed SRS-LBP histograms
- ICFHR 2012 Writer Identification Dataset, precomputed SRS-LBP histograms
- CVL 2013 Writer Identification Dataset, precomputed SRS-LBP histograms
- KHATT Database train-set 1000 writers contributing 4 samples each in arabic, precomputed SRS-LBP histograms
- IAM Database, 301 writers contributing 2 samples each, precomputed SRS-LBP histograms
End-to-end pipeline srs_wi_pipeline.py This script was the used to perform most measurements reported in the article. It requires python 2.7 and assumes a linux system. Execute it with no arguments for help.
- The feature extractor used to produce the above csv files is also available as a command line binary.
bibtex entry :
Any work in this page other than source code or program binaries is licensed under a Creative Commons Attribution 4.0 International License. When applicable atribution should be in the form of a citation.