System that allows you to quickly check a logo for plagiarism

Search for similar images

ИНТЭЛС logo
Отрисовка логотипа для проверки
Поиск похожих изображений

About client

01/
INTELSONLINE is a large patent office that since 1988 has dealt with the preservation of intellectual property

INTELSONLINE has a online service @poisknakov.ru

Who is it for?
experts in the field of intellectual property
ИНТЭЛС logo
The system checks your trademark against other possible duplicates online
creative offices
companies registering a trademark or trademark

Tasks the client came with:

02/
INTELSONLINE came with a ready request for a neural network:
We have finalized it and expanded the capabilities of the service
The client had a non-working, "raw" algorithm
improvement
support
01
02
The project came from our EORA MAGE which is a image search across several knoweldge domains

Interesting fact

Development process

03/
Patent experts:
The process took
selected 100 trademarks and selected 50 similar ones for each.
Our development team:
We trained the nerual network using this dataset. Initially the neural network made decisions based on examples rather than a similiraity criteria
4
months
Result:
a neural network trained on 5 thousand characters connects to the database of existing trademarks - there are 1.5 million of them. The system checks the user's logo for similarity to them.

System operation diagram

04/
User uploads a logo image:
The neural network compares it with trademarks registered in the CIS countries, the Baltic States, as well as the Database of the World Intellectual Property Organization
Our system produces a list of trademarks with a visually similar logo - potential plagiarism:
Вимм-Билль-Данн logo
The list is ranked - the most similar trademarks are shown first. This saves the experts' time.
Cписок товарных знаков с визуально сходным логотипом

Project team

05/
Computer Vision Engineer
MLOps engineer
Project manager
Data scientist
BackEnd developer
ИНТЭЛС logo

Difficulties of the project

06/
In data science, there is always a risk that nothing will work out, simply because the right tools have not yet appeared. It usually takes a long time to find them.
But in this case everything went according to plan. We corrected developer errors and used a succesful algorithim that gave us 80% accuracy.
Were poorley timed

What did we accomplish

07/
01
achieve 80 percent accuracy
show the most relevant images in the first hundred of results
save the customers time by removing the need for patent attorney work
02
03

Mass media wrote about us

08/
Article on the company's website
«How to find all similar images in 5 seconds»
первыйБит logo
"... As a result, we stopped at the development from the Russian company EORA, which creates dialogue systems (chat bots) with high recognition accuracy, as well as solutions with OCR - what we need. Their product for image recognition and comparison is now called EORA MAGE. "

Our plans

09/
speed up the service
submit new pictures to the database so that the neural network can train on them
Contact us