As a foundation of computer vision applications, image segmentation is not only crucial for the vision tasks such as object recognition and tracking, but also important for the higher-level image analysis, such as semantic analysis. In recent years, researchers have proposed a few algorithms which are able to over-segment an image into a number of small pixel groups, called superpixels. The superpixel provides another way to represent an image which, in some cases, is more effective than using the pixels. In this talk, I will present our recent experiments of employing different metrics on the manifolds of superpixel-based covariance matrix and some graph cut algorithms, which leads to competitive segmentation performance on a benchmark dataset.
Last modified: Tuesday, 03-May-2016 09:16:23 NZST
This page is maintained by the seminar list administrator.