Creating and sharing knowledge for telecommunications

An improved method to compute visual saliency in 3D video

Ferreira, L.F. ; Cruz, L. A. S. C. ; Assunção, P.A.

An improved method to compute visual saliency in 3D video, Proc Conf. on Telecommunications - ConfTele, Aveiro, Portugal, Vol. , pp. - , September, 2015.

Digital Object Identifier:


Recently, several methods to compute visual
saliency maps were proposed for 3D image, but for 3D video
no definite solution exists. This paper presents an improved
computational method for visual saliency in 3D video, which
includes human face detection and mapping to better estimate
where the viewers tend to hold their gaze for long periods. The
method is based on aggregation of four feature maps (spatial,
temporal, depth and human face) and center-bias weighting.
The paper examines the impact of the face map on the final
saliency map by comparing the performance of visual saliency
estimator with and without face map. For validation of the
proposed method, two publicly available databases of fixation
density maps were used as ground-truth for comparison. The
results show that the proposed method achieves better results
than other competing state-of-the-art methods.