EFFICIENT UNCONSTRAINED STROKE DETECTOR

Result code in IS VaVaI

<a href="https://www.isvavai.cz/riv?ss=detail&h=RIV%2F68407700%3A21230%2F23%3A00366695" target="_blank" >RIV/68407700:21230/23:00366695 - isvavai.cz</a>

Result on the web

<a href="https://worldwide.espacenet.com/patent/search/family/054979874/publication/EP3380990B1?q=EP3380990B1" target="_blank" >https://worldwide.espacenet.com/patent/search/family/054979874/publication/EP3380990B1?q=EP3380990B1</a>

DOI - Digital Object Identifier

—

Result language

angličtina

Original language name

EFFICIENT UNCONSTRAINED STROKE DETECTOR

Original language description

Scene text localization and recognition, a.k.a. the text-in-the-wild problem, is a key component of many applications such as automated translation, image/video database indexing, assistance to the visually impaired, etc. So far, unlike printed document OCR, no method has reached sufficient accuracy and speed for a practical exploitation. It is, therefore, a need in the art for a fast and accurate text localization and recognition. Stroke detection has applications in extraction of structure information from images, such as bar codes. We propose a novel easy-to-implement stroke detector which is significantly faster and produces significantly less false detections than the detectors commonly used by scene text localization methods. Following the observation that text in virtually any script is formed of strokes, stroke keypoints are efficiently detected, see Figure 1, and then exploited to obtain stroke segmentations. We also propose an efficient classification step to eliminate segmentations which do not correspond to text fragments. Text fragment can be a single character, a group of characters, a whole word or a part of a character. The classifier exploits features already calculated in the detection phase and an effectively approximated "strokeness" feature, which plays an important role in the discrimination between text fragments and a background clutter. Last but not least, an efficient text clustering algorithm based on text direction voting is proposed, in order to aggregate detected segmentations into text line structures and to allow processing by subsequent stages, e.g. an OCR module. When the proposed detector is plugged into a scene text localization and recognition pipeline, the state-of-theart text localization results are maintained whilst the processing time is significantly reduced. The proposed detector has high application potential e.g. in real-time scene text detection in a video stream on mobile phones and embedded systems since a straightforward, single-thread non-optimized version of the algorithm runs in near real-time. Since all stages are scale and rotation invariant and they are not script-specific, a wide variety of fonts and scripts such as Latin, Hebrew, and Chinese can be detected. The detector can be used to produce repeatable local features for e.g. image matching and retrieval.

Czech name

—

Czech description

—

Type

P - Patent

CEP classification

—

OECD FORD branch

10201 - Computer sciences, information science, bioinformathics (hardware development to be 2.2, social aspect to be 5.8)

Project

Result was created during the realization of more than one project. More information in the Projects tab.

Continuities

P - Projekt vyzkumu a vyvoje financovany z verejnych zdroju (s odkazem do CEP)

Publication year

2023

Confidentiality

S - Úplné a pravdivé údaje o projektu nepodléhají ochraně podle zvláštních právních předpisů

Patent/design ID

EP3380990

Publisher

EPO_1 -

Publisher name

European Patent Office

Place of publication

Munich, The Hague, Berlin, Vienna, Brussels

Publication country

—

Date of acceptance

Jun 7, 2023

Owner name

České vysoké učení technické v Praze

Method of use

A - Výsledek využívá pouze poskytovatel

Usage type

A - K využití výsledku jiným subjektem je vždy nutné nabytí licence