Low Memory Implementation of Saliency Map Using Strip-Based Method

摘要

Works in the area of visual saliency are expanding rapidly where visual salience is beginning to find importance in many multimedia and object detection applications. The core of the visual saliency models is the saliency map where information from various features such as intensity, colour, and orientation are encoded onto a single master map. However, the required amount of memory to hold the related maps in the computation of the saliency map is large. This could be seen as a potential complication in hardware constrained environment. In this paper, a low memory implementation of a saliency map using strip-based method is proposed. Simulation results showed that the strip-based method is able to provide a reasonable saliency map while keeping the memory requirements low.