On far end saliency detection of images by compressive sensing

Abstract

Compressed Sensing (CS) is an extremely popular technique used in magnetic resonance imaging, wireless body area network, cognitive radio etc. In this work, CS is used for saliency detection (SD) of images. SD is the technique of highlighting the information of the images which are relevant to human vision. It finds usage in medical imaging, telemetry, advertisement etc. Often SD is performed at a place far away from where the image is captured. So CS can be used to compressively sample the image and thus reduce the amount of image data to be transmitted. But CS reconstruction brings about degradation in the image quality. In this work, CS is combined with SD and the effect of CS on the final saliency map is obtained and compared with the ground truth obtained by using SD without CS. The effect is studied in terms of accuracy, precision and recall of the final saliency maps. This paper proves that though CS tries to degrade the saliency maps, the savings in transmission energy, data rate far outweighs the degradation. Moreover, the degradation effect can be minimized by choosing the sampling rate carefully.