Saliency Detection for Compressive Sensing Measurements

Compressive sensing can obtain high-quality image reconstruction at a lower sampling rate. Using image saliency for compressive sensing measurement and reconstruction can effectively improve the image quality of reconstruction. For this reason, we propose a method of saliency detection using compressive sensing global measurement values. This method uses the characteristics of the Hadamard matrix to reconstruct the original image with low resolution, and then uses the low resolution image for saliency detection. This method takes advantage of the feature that saliency detection does not require high-resolution images. At the same time, low-resolution images are also conducive to neural network reconstruction. Experimental results prove that low-resolution images can indeed obtain better saliency detection results. In order to further improve the saliency detection results, we also propose a saliency detection method with an adaptive measurement matrix. Experiments show that our method can obtain better reconstruction quality and saliency maps with a set of trained compressive sensing measurement matrices.