Matching corresponding points from unannotated images with Bayesian methods

In this work we present an incremental Bayesian model to learn the corresponding points from natural unannotated images. The training set is recursively expanded and the model parameters updated after maching each image. The semirandom set of nodes in the first image is matched in the second image, by sampling with particle filters the unnormalized posterior distribution, being the product of the Gabor filter based likelihood and a Gaussian prior for the shape of the nodes. For each matched node the model assigns an association probability for it to be associated with the object, and having matched few images, the nodes with low association probability are replaced with new ones to increase the number of object nodes. The results show that the model is able to accurately match the corresponding points in natural images.