Ronen Basri
My field of research is computer vision,
and I am focusing particularly on the problem of object recognition. Object
recognition is difficult because the appearance of objects may vary considerably
in different images due to changes in viewing position, illumination, non-rigid
deformation of the objects, and partial occlusion. A particular difficulty arises
when we attempt to recognize a novel object, e.g., a new type of a car, by comparing
it to other instances of its perceptual class, e.g., other kinds of cars. In
my research I am seeking to develop representations of objects and comparison
algorithms that enable a robust recovery of the identity of objects in spite
of these variations. In particular, I have developed methods for recognition
and shape reconstruction under complex illumination conditions. In addition,
I have developed a recognition method that identifies objects by matching their
volumetric parts to regions extracted from the image. Finally, object recognition
processes require a preprocessing stage of segmentation in which objects are
separated from their background. In recent years I have been involved in developing
a novel, multiscale segmentation algorithm. I am currently investigating methods
for using the hierarchical image descriptions obtained using this segmentation
algorithm to produce improved recognition schemes.
Recent Publications
- [with D. W. Jacobs] Projective alignment with regions. IEEE Trans. on Pattern Analysis and Machine Intelligence 23(5) (2001) 519-527.
- [with D. W. Jacobs] Lambertian reflectance and linear subspaces. IEEE Trans. on Pattern Analysis and Machine Intelligence 25(2) (2003) 218-233.
- [with M. Galun, E. Sharon, and A. Brandt] Texture segmentation by multiscale aggregation of filter responses and shape elements. 9th IEEE Int. Conf. Computer Vision (ICCV-03), Nice (2003) 716-723.