Instructor : Bin Wang
This project aims to get a fast and better video segmentation based on machine learning.
Demo
My Contribution
-
Achieved a stable segmentation result on boundary consistency, based on the SVM training and the segmentation result of the first video frame by Mean Shift.
-
Implemented low-level feature extraction of images, such as variance and color, which enhanced the segmentation accuracy and robustness.
-
Improved the recognition of pattern repeated objects, such as zebra, by using the method of point-wise mutual information which was often used in the area of word association.