Title :
Targeted L1L2: Naturalness-constrained image recovery from random projections
Author :
Freeman, Greg J. ; Caramanis, Constantine ; Bovik, Alan C.
Author_Institution :
Electr. & Comput. Eng., Univ. of Texas at Austin, Austin, TX, USA
Abstract :
We devise an elastic net convex program that generates natural-looking compressive sensed images by mimicking features indicative of good quality images. We use the shape parameter of divisively normalized wavelet coefficients to iteratively solve a convex program and target the distribution of each wavelet band. Using this method we are able to create better quality images than other methods as gauged by perceptually relevant quality indices (SSIM).
Keywords :
compressed sensing; convex programming; L1L2; SSIM; elastic net convex program; good quality images; natural-looking compressive sensed images; naturalness-constrained image recovery; normalized wavelet coefficients; perceptually relevant quality indices; random projections; Compressed sensing; Image coding; Image quality; Image reconstruction; Matching pursuit algorithms; Shape; Wavelet domain; Compressed sensing; image quality;
Conference_Titel :
Image Analysis and Interpretation (SSIAI), 2014 IEEE Southwest Symposium on
Conference_Location :
San Diego, CA
DOI :
10.1109/SSIAI.2014.6806038
