The strength and the originality of our team comes from the wide spectrum of image based research topics covered by all of our members: from acquisition to quality assessment including analysis and processing. This global approach is unique compared to other French and European laboratories.
Our works essentially lean on representations of polynomial or wavelet families (X-let) and processings based on the partial derivative equations.
We develop a set of analysis tools from Clifford's algebras to manipulate geometrically the vector data, as well as from the concept of the monogenic signal allowing to give a frame « signal » for color image processings. We develop an original modeling allowing to fix a theoretical frame to redefine the classic operators of multiband image processing.
Another topic concerns the modeling of the geometry for multivalued data. We develop a strategy based on the graph structures to define a multiscale analysis for multispectral images with notions of distances. It is applied for example in applications of image restoration. In a complementary way, we formalized a digital method of pattern detection in multispectral images in the frame of compressive sensing.
Digital tools developed in ICONES team are generalized to include a temporal dimension in the analysis. The team is also interested in the estimation and analysis of the apparent motion in image sequences by variational approaches. In this context we have shown that models based on orthogonal polynomial bases allow to detect robustly singular points in 2D vector fields.
Our works are based on the physical modeling of the Human Visual System, the metrology and interactions between light and materials.
First, studies have been conducted on physical models of interactions between light and materials constituting an acquired scene by focusing on the study of surfaces in terms of topography and roughness, and colorimetric photometric properties in connection with IG and SIR teams of Asali axis.
This work is now continued in the frame of the new CPER by the establishment of the Center of Optical Metrology (CeMOP). Indeed, ICONES team is fully committed to the development of this platform in collaboration with the PPrime Institute in the two following themes:
The consideration of the metrological characterization of the acquisition chain led us to produce a complete set of non-linear processing tools for filtering and mathematical morphology based on distance functions for the quantitative analysis of spectral images. The writing is generic, vector and full-band. It has helped to produce perceptual attributes of texture of color spectral image. These works are internationally valued especially in Division 8 (Image Technology) of the CIE (International Commission on Illumination).
Physical modeling of the Human Visual System is operated for the assessment of the perceived quality of reproduction media or processings (compression, inpainting, color correction) and for the development of bio-inspired processings. Solutions have been proposed for the quality of video surveillance systems under the Ministry of Interior and studies have been conducted with the Home Office (United Kingdom) to ensure the quality of the identification of people.
The natural evolution of ICONES team is to study 3D contents. That is why we model the main physiological phenomena related to stereopsis for the development of a 3D saliency model and a prediction model of the binocular energy. Their effectiveness has been proven in compression and in the development of fully-3D quality metrics. The quality of 3D content is closely linked to the concept of visual comfort, a model was proposed for his prediction. The studies contributed to the development of recommendations on the effects of 3D have been recently published by ANSES (Health Safety Agency for Food, Environment and Labour). Moreover, to exploit these attributes, we have strengthened our activities on the research of multimedia information, including the categorization of images with high-level added value semantics. Furthermore, our work is also interested in the contribution of visual saliency and in bio-inspired content characterization models.