Cluster

Project summary:

Visual saliency has been studied for a long time for the research progresses in computer vision. Detecting and then segmenting the salient part in images helps a lot in both mid-level image processing, e.g., object extraction, and some high-level vision tasks, e.g., scene parsing, automatic image cropping, object tracking, object importance evaluation, and image and video interestingness. In our work, we focus on the subarea of salient object detection, which aims at detecting the most instinct whole object from natural images. Nevertheless, most of the current works use bottom-up saliency models to generate their salient maps, lacking of the concept of “whole object”, while the top-down saliency models have better control on “object”, but they are based on supervised learning. In our work, we make use of the global optimization tools—differential evolution algorithm to attain the object level constraints without learning.

Figure 1. xxxx.

Research highlight:

Our work will provide object level consistency for the salient object detection task without using supervised learning.

Experimental results:

To be added.

Reference:

To be added