A Computational visual saliency Model for Perceptual video coding

Abstract

The detection of visually salient information in videos is advantageous in a lot of application like video quality evaluation and video retrieval. In this research work, we present an efficient method for calculating visual saliency 9f videos for PVC. The proposed algorithm is based on block level search method. Both spatial and frequency domain saliency masks are constructed and integrated to produce the final masks. In spatial domain the mask is obtained using a variant of gbvs method whereas in the frequency domain four different characteristics are extracted from dct coefficients. Both spatial and frequency domain masks are integrated and threshold ING of mask is performed. The method is evaluated on 24 video sequences. The results shows that the proposed method is efficient in comparison to other state of art techniques.