Physics-Based Vision Algorithms

Physics-Based Vision Algorithms

Project type: PhD project

Period: -0001 - -0001

Focus Area: Development of vision algorithms based on the physical laws of image formation. In this way, the image formation process is taken into account and optical effects such as shading, shadows, specularities, inter-reflections, etc. can be discounted/tolerated by the algorithm implementing the vision task. Main results of this research work include an edge detection algorithm (C3E), two segmentation algorithms (IS2R, C3S), a set of methods for estimating the lighting parameters and an algorithm for CCD-based radiometric camera calibration and intensity uncertainty estimation (R2CIU).

IS2R segments an image by (partially) reconstructing the shape of the scene. As a result, at the end of the process the segmentation and a description of the scene shape are available. Unfortunately, the image formation model is limited by the shape reconstruction requirement of the algorithm, and only ambient illlumination and body reflection are allowed. C3 E and C3S are the result of a study about how colour channels are coupled in uniform reflectance image areas and achieve more ambituous segmentations. Both look for image pixels where the coupling is lost because of a reflectance transition arising there, leading to an edge map (C3E) from which a segmentation can be then obtained (C3S). All three algorithms make use of intensity uncertaintes estimated by R2CIU to find reflectance transitions in an adaptive way with respect to image noise.

Project Leader

Project Collaborators

Related Publications