Bird recognition from images is an important and challenging problem with many applications in the real world. The system will take as an input an image with a bird in the middle and it will return a label with the type of bird. The algorithm has to be fully automatic and it has to scale linearly with the number of bird classes.

The recognition process will have two stages: segmentation and classification. The segmentation process will accurately locate the bird in the image. I will explore the use of standard techniques such as region growing or split and merge. The main segmentation novelty will be the incorporation of priors into the segmentation process; these priors will be a model of shape and appearance variation that will be learned from training samples. The classification algorithm will be based in the recent approach bag of words and state-of-the-art Support Vector Machine classifiers. In addition to the bird recognition application, the techniques that we will develop will be useful for other image retrieval applications.