Fast geodesic saliency map for Infrared Imagery

In this work, a saliency map calculation method for Infrared Imagery has been proposed and the resulting map is used for segmentation of the salient objects in the image. The proposed method exploits the geodesic distance concept and computational complexity is much less than the similar methods. Moreover, the proposed work is amenable to be parallelized, suitable for multi-core or FPGA implementation. This method uses the assumption that the boundaries of an image belong to the background. Basically, the geodesic distance of each pixel to the image boundary is calculated and this distance is a measure of the distance to the background model and proportional to the foreground model. The proposed method is tested in our IR dataset and the results reveal that this method has similar performance compared to its counterpart, which uses an approximate geodesic distance metric as this method does. In addition, the proposed work has also been tested in RGB images and results are illustrated.