Saliency-guided wireless transmission of still images using SoftCast
Hadi Hadizadeh · 2016
A novel transmission scheme called SoftCast has recently shown a great potential for wireless transmission of visual signals such as images and videos. In this scheme, the visual input signal is processed by a set of linear transformations to produce a sequence of decorrelated transform coefficients. After that, the produced numbers are scaled by separate scaling factors within a power-distortion optimization (PDO) procedure, and the scaled coefficients are then directly transmitted over the channel. The scaling factors are determined by the expected energy of the coefficients such that coefficients with higher expected energy get more transmission power to get through the noisy channel with lower distortion as compared to coefficients with lower expected energy. Given that, due to the visual attention mechanism of the human brain, some parts of an image are more visually important (salient) than others, it is reasonable to allocate higher transmission power to more salient regions than others to achieve a high subjective quality. Hence, in this paper we propose to use the visual saliency information of an image for transmission power allocation in the PDO process of SoftCast. Using an eye-tracking-weighted peak-signal-to-noise ratio (EWPSNR) measure of subjective quality, our experimental results indicate that at the same channel signal-to-noise ratio (CSNR), the proposed method outperforms the original SoftCast scheme with an average EWPSNR of about 2 dB.