Abstract.
In recent years there has been growing interest in recogni- tion models using local image features for applications ranging from long range motion matching to ob ject class recognition systems. Currently, many state-of-the-art approaches have models involving very restrictive priors in terms of the number of local features and their spatial relations. The adoption of such priors in those models are necessary for simplifying both the learning and inference tasks. Also, most of the state-of-the-art learning approaches are semi-supervised batch processes, which consid- erably reduce their suitability in dynamic environments, where unanno- tated new images are continuously presented to the learning system. In this work we propose: 1) a new model representation that has a less re- strictive prior on the geometry and number of local features, where the geometry of each local feature is influenced by its k closest neighbors and models may contain hundreds of features; and 2) a novel unsuper- vised on-line learning algorithm that is capable of estimating the model parameters effciently and accurately. We implement a visual class recog- nition system using the new model and learning method proposed here, and demonstrate that our system produces competitive classification and localization results compared to state-of-the-art methods. Moreover, we show that the learning algorithm is able to model not only classes with consistent texture (e.g., faces), but also classes with shape only (e.g., leaves), classes with a common shape but with a great variability in terms of internal texture (e.g., cups), and classes of flexible ob jects (e.g., snake).1