For the interpretation of visual information the recognition of objects is crucial. Object recognition is complicated by a variety of photometric variations, including changes of shadows, shading, specularities and illuminant colour. For object recognition to be successful in real-world applications it is essential that robustness with respect to these photometric variations is obtained. This will prove important in many application fields such as surveillance, robotics, defence, manufacturing industry, and image and video search engines.
Although the greater part of image data is in colour format nowadays, most object recognition systems are still based on luminance alone. Robustness with respect to undesired photometric variations can be greatly improved by extending these algorithms to colour. In the field of colour vision invariants are derived from physical models of the reflection of light on surfaces. These invariants have however three drawbacks which have to be taken into account. Firstly, the invariant features are derived from physical models, which only hold in controlled environments. Secondly, the invariants become unstable and unusable in the absence of colour. Thirdly, in the absence of colours photometric invariance is unattainable.
In this proposal we aim to incorporate the photometric invariance theory into the computation of features for object recognition. Instead of full invariance, which results in unusable features when deviations to the physical model occur or when colour is absent, we aim at photometric robustness. We intend to obtain the robustness in two steps: 1. Enrich the existing features used in object recognition with photometric invariant features accompanied by confidence measures. 2. Apply machine -learning techniques to learn the most discriminative features. The machine learning techniques enable the object recognition system to choose for each object between luminance based features and photometric invariant colour features.
Fields of science
- humanitiesartsmodern and contemporary artfilm
- engineering and technologyelectrical engineering, electronic engineering, information engineeringelectronic engineeringrobotics
- natural sciencescomputer and information sciencesartificial intelligencemachine learning
- natural sciencescomputer and information sciencesartificial intelligencecomputer vision
- engineering and technologymaterials engineeringcolors
- natural sciencesphysical sciencesastronomyspace exploration
Call for proposal
See other projects for this call