Citation:

We have decided to make the library available to the research community free of charge. If you use VistaCoderTools in your research, we kindly ask that you reference this website: 

J. Malo, J. Gutiérrez, G. Camps, I. Epifanio, G. Gómez and M. Simón. "VistaCoderTools: a DCT image coding toolbox for Matlab" , http://www.uv.es/vista/vistavalencia/software/software.html

and the paper(s) associated to each algorithm.

Installation:

• Download the file VistaCoderTools1.0.zip (0.6 MBytes)
• Decompress at your computer and set the Matlab path accordingly
• Look at the help of the functions below for instructions on how to use each algorithm.

• MODEL 1: block DCT transform coding with quantization based on JPEG-like linear CSF vision model. It does not include masking relations at all. [as in Wallace91]

• MODEL 2: block DCT transform coding with quantization based on a simple (point-wise) non-linear masking model. It includes auto-masking but it does not include masking relations among coefficients. [ Malo95 , Malo99 , Malo00 ]

• MODEL 3: block DCT transform coding with quantization based on simultaneous diagonalization of covariance matrix and (fixed) perceptual metric matrix (first approximation to  include masking relations among DCT coefficients). [ Epifanio03 ]

• MODEL 4: block DCT transform plus non-linear divisive normalization transform and uniform quantization. This is the proper way to take frequency selectivity and the masking relations into account in the quantization process. [Malo06]

• MODEL 5 : block DCT transform plus CSF inspired constant insensitivity SVM coefficient selection (RKi-1).  SVM based on a rough linear vision model.  [as in Robinson03]

• MODEL 6 : block DCT transform coding plus CSF adaptive insensitivity SVM coefficient selection. SVM based on an accurate linear vision model. [Gomez05]

• MODEL 7 : block DCT transform coding plus divisive normalization and constant insensitivity SVM coefficient selection.   SVM trained in a vision model domain that takes into account frequency selectivity and masking relations among coefficients. [Camps08]

• Matlab functions for achromatic encoding-decoding:

• vista_encoder_achrom.m 

Applies one of the above models to generate file(s) at the desired target entropy(ies), measured in bits / pixel (i.e. file size / number of pixels)

• vista_decoder_achrom.m

Decodes the files generated by vista_encoder_achrom.m

• vista_coder_tools_demo_achrom.m
  

Shows an example of the encoding-decoding process

Basic features (Color Images*):

* Patent pending, Color features are included in VistaCoderTools2.0 (not available on-line)

• MODEL 1 : block DCT transform coding with quantization based on JPEG -like linear CSF vision model in each chromatic channel. It does not include masking relations at all. [inspired in Wallace91]

• MODEL 2 : block DCT transform plus chromatic divisive normalization in each color channel and constant insensitivity SVM coefficient selection. SVM trained in a (space and color) vision model domain that takes into account frequency seletivity and masking relations among coefficients. [Simon08]

• Matlab functions for chromatic encoding-decoding:

• vista_encoder_color.m 

Applies one of the above color models to generate file(s) at the desired target entropy(ies), measured in bits / pixel (i.e. file size / number of pixels)

• vista_decoder_color.m

Decodes the files generated by vista_encoder_color.m

• vista_coder_tools_demo_color.m

Shows an example of the encoding-decoding process

Achromatic images at 0.55 bits/pix
	Original (8 bits/pix)
JPEG-like quantization	Simple masking model [Malo95, Malo99, Malo00]	General masking model Malo06
SVM + crude linear model Robinson03	SVM + accurate linear model Gomez05	General masking+SVM Gutiérrez 07, Camps08

Color images at 0.95 bits/pix

Original	JPEG-like quantization	General chrom.masking+SVM Simon08