-
Notifications
You must be signed in to change notification settings - Fork 16
Tools
In this section various tools that are available for HTR related processes are described.
Provided via docker: docker.laypa
Source: https://github.com/knaw-huc/laypa
[inferencing]
Requires: images of documentments
Provides: empty PageXML and 1 channel pixel maps of the baselines
[training]
Requires: PageXML with the images
Provides: a trained model
\
Laypa can be used of baseline detection, start / end baseline detection and region detection.
For more information see: https://github.com/knaw-huc/laypa
Provided via docker: docker.loghi-tooling
Source: https://github.com/knaw-huc/loghi-tooling
Language: Java
Runs on: CPU
Requires: 1 channel pixelmaps and empty PageXML (output of Laypa
Provides: PageXML containing baselines
Baseline-extraction from pixelmaps (uses pixelmaps to detect baselines,java/CPU)
Provided via docker: docker.loghi-tooling
Source: https://github.com/knaw-huc/loghi-tooling
Language: Java
Runs on: CPU
Requires: 3 channel pixelmaps
Provides: PageXML containing baselines
Baseline-extraction from pixelmaps when start/end is being used. It uses pixelmaps created by P2PaLA in start/end mode or pixelmaps from Loghi-linedetection to detect the actual baselines. It provides the second stage for the two-stage baseline detection. The three channels refer to baseline, start of baseline and end of baseline. The combination of these three elements makes it possible to disentangle baselines that are touching or very close together. Typical examples where this helps are: marginalia that are close to the main text and two columns that are close together. These easily can go wrong with a single channel approach, but are separated quite nicely with a three channel approach.
Provided via docker: docker.loghi-tooling
Source: https://github.com/knaw-huc/loghi-tooling
Language: Java
Runs on: CPU
Requires: PageXML containing baselines and original scans
Provides: images of segmented text lines
- Text line segmentation (requires baselines, java based/CPU)
Textline polygons are updated in the PageXML to reflect the boundaries better.
Provided via docker: docker.htr
Source: https://github.com/knaw-huc/loghi-htr
[inferencing]
Requires: images of segmented text lines and a trained model
Provides: txt files containing text line image filepath and transcription
[training]
Requires: images of segmented text lines and transcription
Provides: a trained model
Full help can be viewed by calling "python3.8 main.py -h"
This is the core of Loghi HTR and contains neural networks that be trained to read handwritten and printed characters from images of text lines.
Provided via docker: docker.loghi-tooling
Source: https://github.com/knaw-huc/loghi-tooling
Language: Java
Runs on: CPU
Requires: txt files containing text line image filepath and transcription
Provides: PageXML
Merges output from Loghi HTR with existing PageXML or creates new PageXML.
Provided via docker: docker.loghi-tooling
Source: https://github.com/knaw-huc/loghi-tooling
Language: Java
Runs on: CPU
Requires: PageXML
Provides: PageXML
- Region detection based on textline clustering(optional, java/CPU)
- Applying reading order (optional/java/CPU)
- Cleaning of common errors (optional/java/CPU)
Provided via docker: docker.loghi-tooling
Source: https://github.com/knaw-huc/loghi-tooling
Language: Java
Runs on: CPU
Requires: PageXML
Provides: PageXML
Detects language based on either preset trainset or custom trainset. The language is detected on TextLine-level, TextRegion-level and Page-level.
Provided via docker: docker.loghi-tooling
Source: https://github.com/knaw-huc/loghi-tooling
Language: Java
Runs on: CPU
Requires: PageXML
Provides: PageXML
This can segment Textlines from existing PageXML into Words. It uses interpolation Segmenting of Textlines into Words using interpolation.
This produces PageXML containing just the text-regions.
-
(requires segmented textlines, python/GPU and java/CPU)
-
Loghi-HTR (requires segmented textlines, python/GPU)
-
Loghi-HTR-mapper (requires segmented textlines, python/GPU and java/CPU)
-