KeCoDe [Kernel Image Coding and Denoising]
Authors
V. Laparra, J. Gutiérrez, J. Muñoz, G. Camps-Valls, and J. Malo
Download
Full Matlab Package


Code for the paper
KeCoDe: Kernel Image Coding and Denoising Toolbox
V. Laparra, J. Gutiérrez, J. Muñoz, G. Camps-Valls, and J. Malo
Journal of Statistical Software, submitted
, 2014.


Abstract
Image coding and denoising are well-established fields of research in which kernel methods have entered only recently. Kernel methods in general and support vector machines (SVM) in particular have yielded very competitive state-of-the-art approaches. However, the advantages of these methods are only obtained when they are properly designed to capture the statistics of natural images and the perceptually relevant features. We have developed a Matlab toolbox that implements two adaptive versions of SVM specially well suited for coding and denoising. In these applications, the key is the right choice of (1) the coefficient-dependent e-insensitivity related to the perceptual meaning of the image representation domain, (2) the kernel design related to the statistical and perceptual relations between image coefficients, and (3) the coefficient-dependent penalization factor related to the energy of the image coefficients and their perceptual relevance.

The presented Kernel-based Image Coding and Denoising (KeCoDe) toolbox wraps the novel SVM approaches and includes related coding and denosing methods for comparison purposes.


Coding Results

Achromatic Images (0.55 bits/pix)

jpeg
simple_masking
general_masking
JPEG-like coding [Wallace91]
(CSF+uniform quantizer)
[Malo95, Malo99, Malo00]
(Simplified Masking)
[Malo06]
(General Masking)
RKI1
SVM_CSF
General_masking_SVM
[Robinson03]
(SVM+Simplified CSF)
[Gómez05]
(SVM+accurate CSF)
[Camps08]
(General Masking+SVM)


Color Images (0.95 bits/pix)

original
jpeg
general_masking_SVM
Original (24 bits/pix)
JPEG [Gutiérrez12]
General Masking + SVM


Image Coding schemes included in KeCode
  • JPEG-like coding: linear CSF + uniform quantizer.
  • Non-uniform adaptive quantizer based on simplified masking models [Malo95, Malo99, Malo00].
  • Non-uniform adaptive quantizer based on general masking models [Malo06].
  • SVM DCT coefficient selection using simplified CSF [Robinson03].
  • SVM DCT coefficient selection using accurate CSF [Gómez05].
  • SVM coefficient selection in divisive normalized domain [Camps08].
  • SVM coefficient selection in divisive normalized domain with accurate color contrast definition [Gutiérrez12].
Restoration Results

Gaussian Noise Blur + Gaussian Noise JPEG Noise Salt and Pepper
noise
deblur
jpeg
salt
PSNR=25         SSIM=0.83
SCIELab=0.72   DN=0.18
PSNR=24.6       SSIM=0.61
SCIELab=0.57   DN=0.19
PSNR=25         SSIM=0.72
SCIELab=1.22   DN=0.25
PSNR=25.3       SSIM=0.83
SCIELab=0.37   DN=0.19


Image Restoration schemes included in KeCode
  • Wavelet and Kernel based denoising methods
  • SVM regression with Mutual Information Kernels (includes relations among coefficients) [Laparra10]
  • Hard Thresholding [Donoho95]
  • Soft Thresholding [Donoho95]
  • Bayesian approach assuming Gaussian marginal PDFs [Figueiredo01] 
  • Regularization in local frequency domains
  • Adaptive regularization functional based on perceptual divisive normalization (includes relations among coefficients) [Gutiérrez06]
  • L2 regularization functional [Tychonov77].
  • CSF-based regularization functional [Andrews77].
  • Adaptive Auto-Regressive regularization functionals [Banham97].

Regularization (denoising)


Regularization (deblurring+denoising)


Regularization (removing JPEG noise)


Regularization (removing Salt and Pepper)




Wavelet and Kernel-based methods (Denoising)


Wavelet and Kernel-based methods (Removing JPEG noise)



Related Papers
  • [Malo06] J. Malo, I. Epifanio, R. Navarro, and E. Simoncelli. Non-linear image representation for efficient perceptual coding. IEEE Trans. Im. Proc., 15(1):68–80, 2006.
  • [Gómez05] G. Gómez, G. Camps-Valls, J. Gutiérrez, and J. Malo. Perceptual adaptive insensitivity for support vector machine image coding. IEEE Transactions on Neural Networks, 16(6):1574–1581, 2005.
  • [Camps08] G. Camps-Valls, J. Gutiérrez, G. Gómez, and J. Malo. On the suitable domain for SVM training in image coding. Journal of Machine Learning Research, 9:49–66, 2008.
  • [Gutiérrez12] J. Gutiérrez, M.J. Luque, G. Camps-Valls and J. Malo. A Color Contrast De nition for Perceptually based Color Image Coding. Recent patents on Signal Processing. 2(1):33-55, 2012
  • [Gutiérrez06] J. Gutiérrez, F. Ferri, and J. Malo. Regularization operators for natural images based on nonlinear perception models. IEEE Tr. Im. Proc., 15(1):189–200, 2006.
  • [Laparra10] V. Laparra, J. Gutiérrez, G. Camps-Valls, and J. Malo. Image denoising with kernels based on natural image relations. Journal of Machine Learning Research, 11:873–903, 2010.

Copyright & Disclaimer
This software is licensed under the FreeBSD license (also known as Simplified BSD license).

Copyright (c) 2006-2013  Valero Laparra,
Juan Gutierrez, Jordi Muñoz, Gustau Camps-Valls, and Jesus Malo {jordi.munoz, juan.gutierrez, valero.laparra, gustavo.camps, jesus.malo}@uv.es
All rights reserved.

Redistribution and use in source and binary forms, with or without modification, are permitted provided that the following conditions are met:

- Redistributions of source code must retain the above copyright notice,  this list of conditions and the following disclaimer.
- Redistributions in binary form must reproduce the above copyright notice, this list of conditions and the following disclaimer in the documentation and/or other materials provided with the distribution.

THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES  OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT,  INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY  WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.