Multi-Model Estimation and Its Application to the Problem of Finding Repeated Texture in Images
In computer vision multi-model estimation is the task of estimating multiple geometric transformations from a set of features extracted from images. The task is difficult because it is unknown beforehand whether a model generated some subset of the data. Furthermore, the data can be corrupted with outliers, which are erroneous data that are mistakes of data acquisition. This talk will present two estimators that solve the multi-model estimation problem in the context of computer vision. In particular, estimators will be presented that segments scene planes and repeated coplanar texture. However, the problem formulation is general and can be applied to a wide variety of tasks that contain data with multiple relevant structures.