Jagtap, Mahesh (2014) Interactive image segmentation. MTech thesis.
Segmentation of objects from still images has many practical applications. In the past decade, combinatorial graph cut algorithms have been successfully applied to get fairly accurate object segmentation, along with considerable reduction in the amount of user interaction required. In particular, the Grabcut algorithm has been found to provide satisfactory results for a wide variety of images. This work is an extension to the Grabcut algorithm. The Grabcut algorithm uses Gaussian mixture models to fit the color data. The number of Gaussian components used in mixture model is however fixed. We apply an unsupervised algorithm for estimating the number of Gaussian components to be used for the models. The results obtained show that segmentation accuracy is increased by estimating the Gaussian components required, prior to applying the Grabcut algorithm.
|Item Type:||Thesis (MTech)|
|Uncontrolled Keywords:||Interactive image segmentation, Gaussian mixture models, Minimum description length, Expectation maximization, Mincut/maxflow algorithm|
|Subjects:||Engineering and Technology > Computer and Information Science > Image Processing|
|Divisions:||Engineering and Technology > Department of Computer Science|
|Deposited By:||Hemanta Biswal|
|Deposited On:||11 Sep 2014 11:03|
|Last Modified:||11 Sep 2014 11:03|
|Supervisor(s):||Sa, P K|
Repository Staff Only: item control page