Image Visual Saliency Feature Extraction Based on Multi-Scale Tensor Space

In view of the traditional saliency detection method gets imprecise and vague region boundary, so that the detected object is not connected, the paper proposes image visual saliency feature extraction based on multi-scale tensor space. The method introduces the tensor space, using multiple low-level image features to construct the tensor space, after reducing dimension the image space structure and correlation features are preserved, making the detected object connective, which is beneficial to feature extraction and target detection, at last the features uncertainty weights are calculated for total saliency feature fusion. The experimental results show that the algorithm of feature extraction proposed in this paper is closer to the real object and achieves better results.