Dimension reduction into superpixels for image analysis and processing
With Rémi Giraud
Dimension reduction into superpixels for image analysis and processing
Superpixels locally group pixels into regions and have become very popular in image processing and computer vision applications. The aim is to exploit the local redundancy of information to lower the computational burden and to potentially improve the performances by reducing the noise of a processing at the pixel level. Superpixels can also be considered as a multi-resolution app roach that preserves image contours, contrary to standard regular down-sampling methods.
The presentation will first focus on the superpixel decomposition process itself. We answer to the limitations of state-of-the-art methods that may still highly fail to produce accurate pixel clustering in the presence of object contours, noise, or textures.
Then, we propose a new framework adapting standard regular patch-based approaches, e.g., non-local means, to these irregular regions. Results on several applications such as segmentation, classification and color transfer will be presented.
- Speaker: Rémi Giraud
- Monday 11 March 2019, 11:00–12:00
- Venue: MR13 Centre for Mathematical Sciences.
- Series: Cambridge Image Analysis Seminars; organiser: AI Aviles-Rivero.