Master Thesis

Key-point Matching for Non-Rigid Endoscopic Image Streams

Work default illustration


  • If you are interested in the proposal, please contact with the supervisors.


Deformable key-point-based image matching is the task of finding correspondences between key points detected in two or more images [1,2] that are related by an unknown non-rigid motion. This task is particularly difficult for very large elastic deformations, such as the images obtained from endoscopic videos [3]. In this project, we will study this type of images and evaluate the most relevant strategies (from classical descriptors to more novel CNN-based approaches [4]) to match non-rigid images, analyzing the failure cases and proposing a solution. This master thesis will be carried on in the Institut de Robòtica i Informàtica Industrial, at the Universitat Politècnica de Catalunya (under direction of Dr. Antonio Agudo and Dr. Francesc Moreno).


Candidates with a background in mathematics, medical image and good programming skills (Matlab/C++) are particularly encouraged to apply.

For additional information, please contact Dr. Antonio Agudo at

[1] D. Lowe. Distinctive image features from scale-invariant keypoints. In IJCV 60: 91-110, 2004.

[2] E. Rosten and T. Drummond. Machine learning for high-speed corner detection. In ECCV, 2006.

[3] A. Agudo, F. Moreno-Noguer B. Calvo and J.M.M. Montiel. Sequential non-rigid structure from motion using physical priors. In TPAMI, 38(5): 979-994, 2016.

[4] E. Simo-Serra, E. Trulls, L. Ferraz, I. Kokkinos, P. Fua and F. Moreno-Noguer. Discriminative learning of deep convolutional feature point descriptors. In ICCV, 2015.

The work is under the scope of the following projects:

  • RobInstruct: Instructing robots using natural communication skills (web)