Improving Spatial Saliency Using Affinity Model and Temporal Motion

摘要

Saliency map has been applied in diverse fields such as image segmentation, object detection, image scaling and so forth. Over the past decades, a variety of spatial saliency generation methods for still images have been introduced. Recently, motion saliency has gained much interest where motion data estimated from an image sequence are utilized In this paper, we propose the saliency generation method that enhances the spatial saliency based on the combination of spatial and motion saliencies without the consideration of motion classification. Further, an affinity model is integrated for the purpose of connecting close-by pixels with different colors and obtaining a similar saliency. In experiment, we performed the proposed method on seven image sequence sets. Our saliency map is compared with Zhai's method for evaluating the saliency improvement. Further, from the objective performance evaluation, we validated that the saliency value increases by +41 per a pixel over the spatial saliency on the average.