Single image super-resolution using sparse representations with structure constraints

Julio Cesar FERREIRA1,2,4    Olivier LE MEUR 2   Christine GUILLEMOT 1   Eduardo A. B. DA SILVA 3   Gilberto A. CARRIJO 4

1INRIA, France
2University of Rennes 1, France
3Federal University of Rio de Janeiro (UFRJ), Brazil
4Federal University of Uberlandia (UFU), Brazil

ICIP (2014)

Back

Abstract

This paper describes a new single-image super-resolution algorithm based on sparse representations with image structure constraints. A structure tensor based regularization is introduced in the sparse approximation in order to improve the sharpness of edges. The new formulation allows reducing the ringing artefacts which can be observed around edges reconstructed by existing methods. The proposed method, named Sharper Edges based Adaptive Sparse Domain Selection (SE-ASDS), achieves much better results than many state-of-the-art algorithms, showing significant improvements in terms of PSNR (average of 29.63, previously 29.19), SSIM (average of 0.8559, previously 0.8471) and visual quality perception.

Paper and presentation

Paper [pdf] Poster [pdf]

Materials











LR              Nearest                  Dong et al.                 Our Results     

The training dataset is composed of 5 images and testing dataset is composed of 10 pictures. You can download these dataset below.

[Testing Dataset]

[Training Dataset]

Some Results

The PSNR (dB) and SSIM results (luminance components) of super-resolved HR images. 



	Images
	Daubechies et al.
	Dai et al.
	Yang et al.
	Marquina et al.
	Dong et al.
	Our results




	Girl
	32.93 0.8102  
	31.94 0.7704
	32.51 0.7912
	31.21 0.7878
	33.54 0.8242
	33.56 0.8252




	Parrots
    28.78 0.8845
    27.71 0.8682
    27.98 0.8665
    27.59 0.8856
    30.00 0.9093
    30.29 0.9136




	Butterfly
    25.16 0.8336
    25.19 0.8623
    23.73 0.7942
    26.60 0.9036
    27.34 0.9047
    28.48 0.9236




	Leaves
    24.59 0.8310
    24.34 0.8372
    24.35 0.8170
    24.58 0.8878
    26.80 0.9058
    27.69 0.9261




	Parthenon
    26.32 0.7135
    25.87 0.6791
    24.08 0.6305
    25.89 0.7163
    26.83 0.7349
    27.05 0.7446




	Flower
    28.16 0.8120
    27.50 0.7800
    27.76 0.7929
    27.38 0.8111
    29.19 0.8480
    29.29 0.8511




	Hat
    29.92 0.8438
    29.68 0.8389
    29.65 0.8362
    29.19 0.8569
    30.93 0.8707
    31.51 0.8805




	Raccoon
    28.80 0.7549
    27.96 0.6904
    28.49 0.7273
    27.53 0.7076
    29.24 0.7677
    29.27 0.7686




	Bike
    23.48 0.7438
    23.31 0.7219
    23.20 0.7188
    23.61 0.7567
    24.62 0.7962
    24.97 0.8098




	Plants
    31.87 0.8792
    31.45 0.8617
    31.48 0.8698
    31.28 0.8784
    33.47 0.9095
    34.17 0.9163




	Average
    28.03 0.8115
    27.49 0.7910
    27.69 0.7954
    27.49 0.8190
    29.19 0.8471
    29.63 0.8559

Citation

Ferreira et al. (2014). Single image super-resolution using sparse representations with structure constraints, ICIP 2014 . 
@inproceedings{ferreira14single,
    title = {{Single image super-resolution using sparse representations with structure 
		constraints}},
    author = {Ferreira, Julio Cesar and Le Meur, Olivier and Guillemot, Christine and 
		Da Silva, Eduardo Antonio Barros and Carrijo, Gilberto Arantes},
    keywords = {super-resolution, sparse representations, structure tensors},
    language = {English},
    booktitle = {{IEEE International Conference on Image Processing}},
    address = {Paris, France},
    audience = {international },
    year = {2014},
    month = Oct,
}

Related