Geo AI Technical capacities

Bubble detection

  • Main AI capacities
  • Bubble with stroke.
  • Coloured and transparent bubble.
  • Elliptical, spiky and implied bubble.
  • Pixel-level contour detection.
  • Decomposition of multi-bubble and joint bubble
  • Combined with panel detection
  • Ordering of bubbles in panels.
  • Combined with character recognition
  • Recognition of the bubble tail position and direction.
  • Association of the bubble tail with the comic character.

Panel detection

  • Main AI capacities
  • Panel with a wide range of stroke and resilient to small openings.
  • Elliptical or implied panel.
  • Isolated drawing with space encompassing.  
  • Panel order computation.
  • Algorithm robustness
  • Overflowing panel from the page.
  • Incrusted and unclosed panel.

Text extraction

  • Main AI capacities
  • Text in speech and context bubbles.
  • Superimposed  text on  drawings (e.g. onomatopoeia).
  • Text in illustrations (e.g. store names, vehicle number).
  • Text bloc identification in multibubble.
  • Supported langages
  • Support for all Latin and Cyrillic languages.
  • Support for the main Asian vertical and horizontal texts (JP, KO…).
  • Detection and removal of furiganas in Japanese
  • Other languages can be implemented on demand.
  • R&D in progress
  • Bold and italic word tagging.
  • Onomatopoeia tagging and text recognition.
  • Refined Kanji recognition in Japanese manga.

Handwritten & cursive text recognition

  • Main AI capacities
  • Latin and Asian handwritten and cursive texts.
  • Capacity to switch to "training mode".
    (→ for an non recognised handwriting)
  • Inside or outside speech bubbles.
  • Outline curve text superimposed on drawing.
  • Linguistics post treatments
  • Post-OCR correction.
  • Spell checking with dictionaries.
    (→ to correct imprecisely written words)
  • R&D in progress
  • Curved text.
  • Japanese text within illustrations.

Long page support (→ webtoon)

  • Preprocessing implemented
  • Mono or sliced page image.
  • Recomposition of cut elements in a single "long" vertical page.

Double page detection

  • AI capacities not yet implemented in Geo Comix webapp
  • Graphic analysis of facing pages to detect double pages.
  • Ordering of panels and bubbles considering the spreed.

Character recognition

  • Main AI capacities
  • Graphical pattern of character.
  • Face localisation in the identified character pattern.
  • Propagation of anonymous identification in visual narrative type of documents such as comics, manga or webtoon.
  • R&D in progress
  • Character identification with real names.
  • Propagation of anonymous identification to non recurrent character and across albums.

Onomatopea & titration recognition

  • Main AI capacities
  • 'Very' graphical text position and most of the titration.
  • 'Little' distorted onomatopoeia text recognition.
  • Onomatopoeia with graphical consistency between characters.
  • Japanese katakana first level of support.
  • R&D in progress
  • Graphically 'complex' onomatopoeia.
  • Very distorted onomatopoeia.
  • Curved or completely round titration.

Emotion  recognition

  • R&D in progress
  • Face expression analysis.
  • Bubble type recognition.
  • Graphical pattern conveying emotion.
  • Character gender recognition.

Main concepts about Geo AI

Powerful technology produces simple, robust and effective results
while concealing high level complexity.

Combination of multiple methods for a super reliable result

Several methods are parallelized for each element (bubbles, panels...). Performing self-evaluation for each detection, Geo AI provides the best combination as output.

Connexion between AI components to enhance overall analysis

Components can exchange information to refine their detections and avoid missing relevant elements.

Comics/Webtoon file preprocessing

Our preprocessing chain is compatible with the maximum diversity of formats to best integrate with your workflow.

CBML, Geo AI output format

The Comic Book Markup Language (CBML), is the main output format of Geo AI and is a TEI-based XML vocabulary for encoding BD, comics, manga, webtoon,graphic novels, and related documents.

Beyond translation: all what Geo AI can do

L3i Laboratory - La Rochelle University

Accessible comics

Thanks to Geo's automatic detection of texts and characters, visually impaired people will finally be able to access comic books easily.

National Research Agency - LabCom

Better indexation

Exploit all the data extracted by Geo AI to significantly enrich the referencing of your comic books, as well as your own on-site search engine results.

European Union

Web/Print conversion

Our technology can be used to develop a semi-automated webtoon ↔︎ print conversion tool. Accelerate your designer adaptation work up to 50%!

European Union

Rich digital comics

Take advantage of Geo AI to enrich the ePUB production of your comic books: make them multilingual, ready for a guided view navigation, and much more.