The final meeting of the MobilAI project has been held at the end of March. Let’s focus on this project and its benefits for ARIADNEXT.
Image recognition on cell phones: a challenge based on optimizing network use
Image analysis and recognition technologies, from photos and videos, are becoming more and more powerful. More and more services are using these techniques coupled with deep learning for various applications: identity verification, augmented reality, e-shopping, etc.
The cameras available on our smartphones allow us today to obtain images of sufficient quality to be exploited. The recovered images are then processed either via a cloud, with a processing speed impacted by the availability of the mobile network, or locally on the phone, most often with basic methods that are far from the state-of-the-art, due to the rapid progress made in the field in recent years.
Advanced processing (recognition, deep network analysis, …) on cell phones is still limited. It is necessary to address this technological challenge to allow fast image processing, to improve the user experience, to democratize the use of smartphones for this type of service, and to address the problems of mobile networks by performing more and more processing locally on the phone.
MobilAI: a French consortium to optimize image recognition on mobile phones
In 2018, four French organizations, including ARIADNEXT, decided to collaborate to address this issue. To improve the efficiency of image processing on the network, we have chosen to use recent and high-performance approaches based on deep learning artificial intelligence algorithms. The main challenge of this project was the industrialization of these methods, previously used mainly in academic research, but little for commercially available applications.
The companies Lamark (now Imatag), Quai des Apps, and ARIADNEXT, all three using mobile image processing in their products, have joined forces with Inria in Rennes for the MobilAI project. This two-year project, funded by the Brittany region and the Pays de la Loire region to reply to the call of the “Images et Réseaux” cluster, began in September 2018 and ended in March 2021.
The benefits of MobilAI for image recognition
MobilAI has allowed the advancement of the-state-of-the-art in the field of deep learning applied to image recognition. In particular, new architectures of deep neural networks have been proposed to meet the constraints of the mobile environment, in terms of processing time and storage space. These new convolutional networks constitute the new state-of-the-art for the specific case of identity document recognition.
The benefits of MobilAI for ARIADNEXT
MobilAI has enriched the mobile version of the identity control system proposed by ARIADNEXT, IDCheck.io. The main product of the company, developed within the framework of the collaborative project IDFraud, confirms the authenticity of an identity document from a simple photo of it.
The acquisition of cutting-edge expertise in deep learning for image recognition applications will also be an important knowledge base for future developments of the company.
Finally, this project has led to several publications and scientific communications, including a publication by ARIADNEXT researchers.
Within the framework of the project, three applications were studied:
– Classification of identity documents’ images (ARIADNEXT)
– Reading of text contained in identity documents, from an image (ARIADNEXT)
– An augmented reality application to better accompany an art exhibition (Quai des Apps)
The two applications developed for ARIADNEXT have been introduced in its product IDCheck.io.
Beyond the framework of the project, contributions in the field of health (for example for remote visual diagnosis) or education (interactive learning support) are envisaged.
Collaborate with ARIADNEXT
Innovation and research have been at the heart of ARIADNEXT’s activities since the company was founded. Our team of experts is constantly seeking to push the current limits in the areas of identity verification and fraud detection, in order to offer ever more effective solutions. Image analysis, character recognition, biometrics, deep learning, artificial intelligence… we are convinced that together we can go further!
We are open to any type of collaborative research. Do not hesitate to contact us!
This project is supported by Brittany region.