What does visual scene recognition mean

Jump labels

Partially supervised learning of models for visual scene recognition

Research project funded by the German Research Foundation (DFG)

Project duration: 16.03.2015 - 15.03.2018

Visual object and scene recognition is an important research area in the field of machine vision. The highly complex modeling of current processes is trained on extensive random samples with monitored learning processes. Since these have to be completely annotated for this purpose, considerable effort arises, which makes the method only applicable to a limited extent. Alternatives to supervised learning are partially or completely unsupervised methods, which, however, when viewed in isolation, each have significant fundamental disadvantages. Therefore, a meta-learning process for object and scene categorization is to be developed in this project. In this, different, partially monitored learning processes should interlock in such a way that a detailed analysis of natural scenes is possible with as little manual effort as possible. The essential aspects of the meta-learning process are the unsupervised determination of proto-categories for scenes, the web-based learning of relevant object models, the partially monitored learning of contextual relationships in partial scenes and the active learning of structured scene models. The generalization ability of the models, the robustness and self-assessment possibilities of the classifiers and the interaction between models of different granularity must be treated as additional challenges.