Abstract
This paper presents a clustering-based color segmentation method where the desired object is focused on. As classical methods suffer from a lack of robustness, salient colors appearing in the object are used to intuitively tune the algorithm. These salient colors are extracted according to a psychovisual scheme and a peak-finding step. Results on various test sequences, covering a representative set of outdoor real videos, show the improvement when compared to a simple implementation of the same K-means oriented segmentation algorithm with ad hoc parameter setting strategy and with the well-known mean-shift algorithm.
Highlights
Digital videos are nowadays widespread on the World Wide Web or mobile phones but, whereas text documents are selfdescribing, their utility suffers as they do not give any explicit description of their content
The use of MPEG-7 dominant color descriptor (DCD) will be implemented, and to avoid an overfitting behavior, we introduce a new approach based on a perceptive saliency model [9]
Regarding the previous conclusion using dominant colors, lets compare the results obtained with this saliencybased approach
Summary
This paper presents a clustering-based color segmentation method where the desired object is focused on. As classical methods suffer from a lack of robustness, salient colors appearing in the object are used to intuitively tune the algorithm. These salient colors are extracted according to a psychovisual scheme and a peak-finding step. Results on various test sequences, covering a representative set of outdoor real videos, show the improvement when compared to a simple implementation of the same K-means oriented segmentation algorithm with ad hoc parameter setting strategy and with the well-known mean-shift algorithm
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have