Multiscale compressed sensing method for ROI coding

In order to obtain more information about the target from UAV images through limited wireless channel capacity, a multiscale ROI compressed sensing method was proposed. The 9/7 biorthogonal wavelet transform was carried out on the image to concentrate the energy to the low frequency domain. The compressed sensing method was executed in the wavelet domain by assigning higher measurement rate to low frequency component of the whole image, and lower ones for the high frequencies component of the ROI, which would decrease with the rising of the frequencies. Finally, the smoothed projected Landweber (SPL) method was used for reconstruction. Experiments demonstrate that the proposed method can improve the PSNR by 0.6-1.2 dB compared with the state of art methods.