Training-and Segmentation-Free Intuitive Writer Identification with Task-Adapted Interest Points

Abstract : Identifying the writer of a document establishes its authenticity or authorship and has several applications, notably in forensic and historical document analysis. Previous research has shown the potential of Interest Points (IP) for writer identification, but existing methods require segmentation or training. This paper evaluates the performance of intuitive features computed directly from IP properties rather than extracting descriptors at their locations; allowing for a training-free approach. Secondly, we show that adapting detectors to the specific task of writer identification is not only vital for performance but also allows for segmentation-free approaches. Experiments on widely-used datasets show the potential of the method applied self-contained and when combined with existing methods. Limitations of our method relate to the amount of data needed in order to obtain reliable models.
Document type :
Conference papers
Complete list of metadatas

Cited literature [11 references]  Display  Hide  Download

https://hal.univ-antilles.fr/hal-01165881
Contributor : Philippe Simon <>
Submitted on : Saturday, June 20, 2015 - 4:07:27 PM
Last modification on : Monday, June 22, 2015 - 2:02:42 PM
Long-term archiving on: Tuesday, April 25, 2017 - 6:20:12 PM

File

IGS_2015_submission_15.pdf
Files produced by the author(s)

Identifiers

  • HAL Id : hal-01165881, version 1

Collections

Citation

Angelika Garz, Marcel Würsch, Rolf Ingold. Training-and Segmentation-Free Intuitive Writer Identification with Task-Adapted Interest Points. 17th Biennial Conference of the International Graphonomics Society, International Graphonomics Society (IGS); Université des Antilles (UA), Jun 2015, Pointe-à-Pitre, Guadeloupe. ⟨hal-01165881⟩

Share

Metrics

Record views

257

Files downloads

182