Skip to content. Skip to navigation
CIM Menus
 

Optical flow estimation by complex factorization over quadrature feature pairs


Yanyan Mu
Centre for Intelligent Machines McGill University

February 22, 2018 at  8:30 AM
McConnell Engineering Room 437

For any visual perception task, extract invariances from input is a inevitable. Investigation on how to achieve this has been studied by computer vision, machine learning and neuron scientists over the last decades. From those research, we know that the visual information is processed in a hierarchical structure. At the lower level of the hierarchy, input from optical nerves are represented as local features such as edge at different orientation and motion vector. Those information propagation layer by layer to the higher level, they are represented as more complex patterns such as object classes and motion trajectories. There are many exist hierarchy model to simulate this visual process. However, the exact computation model is not clear even for the lower level, it's still an mystery. One missing piece in those model is the decomposition of static patterns and their motion. This research is focus on the lower level representation for decouple static pattern and motion independently.