Doctorate student on AI and image/video analysis M/F
Created at the end of 2012, b<>com (private Institute of Research and Technology) supplies technology solutions and accelerates innovation for organizations wishing to enhance their competitiveness using digital technology.
b<>com unites talented individuals from multiple disciplines and cultures including; augmented reality and virtual reality, immersive video and audio, artificial intelligence, cybersecurity, 5G networks, the internet of things and cognitive sciences, e-Health, etc.
Combining the best of the industrial and university worlds, b<> com’s researchers and engineers enjoy a creative and innovative work environment at its Rennes campus and sites in Paris, Brest and Lannion. With its advanced engineering team and considerable scientific resources, b<>com offers its customers the perfect set of ingredients and tailor-made solutions that make all the difference in today’s turbulent socioeconomic context.
Thesis title : Automatic detection of logos and graphics in videos using AI-based methods
The thesis aims to design and develop solutions for the automatic detection of logos and graphics in video content in order to provide an accurate alpha mask separating the underlying image from the overlaying graphics.
In many cases, logos and graphic elements are embedded in video content, such as TV logos, text banners, etc. As these elements have different characteristics from those of the underlying images (texture, color, motion), they may require different processing when converting the videos, for example in terms of resolution or dynamic range conversions (SDR to HDR).
Although several works exist on logo detection, they focus on still images, rely on known logo and brand databases and aim to detect the location but not to precisely separate the graphical element from the image. As such, they are not suitable for application to video content.
The thesis focuses on this particular problem, addressing graphical elements at different levels of complexity: fixed logos, text banners or larger textual elements, and finally graphical elements potentially containing transparency. To detect and separate these elements, traditional image processing approaches will be considered, as well as deep learning solutions.
Graduated from a Master 2 or an engineering school, the candidate has the following skills :
- Knowledge of mage/video processing techniques
- Python programming
- Experience in AI/deep learning techniques
- Independent and organized
- Able to analyze and synthesize information
- Writen and spoken english
- Duration: 3 years PHD contract
- Start date: September / October 2022
- Location: Rennes