The ‘Imagerie couleur’ team develops automatic image analysis methods that represent colour and/or multispectral information in natural scenes. Its researchers focus on:

• extracting texture descriptors from raw images acquired by colour or multispectral cameras equipped with a single sensor;
• the design of colour texture features and the semi-supervised selection of feature spaces for image segmentation;
• the recognition of objects observed under uncontrolled acquisition conditions using reflectance analysis estimated in the visible and near infrared range;
• dense estimation of the motion of vibrating objects on a sub-pixel scale.

Future work will focus on tensor modelling of multichannel images acquired by cameras with different characteristics.

The main fields of application that have been tackled recently include:

• precision agriculture, for example weed detection on agricultural plots;
• structural integrity monitoring, in particular using modal analysis.

The experimental developments were carried out using acquisition devices made available by the Equipex IrDIVE and the ‘Sciences et Cultures du Visuel’ Research Federation.