This demo implements the algorithm introduced in the paper
A Wavelet Perspective on Variational Perceptually-Inspired Color Enhancement
published in 2014 by E Provenzi and V Caselles.
This algorithm performs color and contrast enhancement in the wavelet domain and this demo allows you to test it on your own images.
Note that image larger than 1200x1200 are resized to avoid too long computation time and/or mermory consumption.
Instructions
First upload an image using the file selector in the top left part of this page.
The in the color enhancement section, you may set your parameters and then click on Apply to see the algorithm effect on the uploaded picture.
The computation may take several seconds.
Once the image is processed, it will be displayed and you can use the arrow keys to navigate between the original image,
the processed image or see differences between these images. These different pictures are also available with the View selector.
The Stretch dynamic selector allows to expand the image dynamic after algorithm processing.
The stretching can be done either on the luminance channel or oon each RGB channel independently.
Parameters
The best way to understand then is with any doubt to read the paper, but i provide some hint to build intuition on their effects.
α: For each channel, it controls the dispersion arround the average value.
Set to zero, this parameter has no effect, while set to 1 the image will be washed out.
w: It controls the contrast enhancement the bigger the value the greater the effect.
K: It controls the magnitude of details enhanced. Small values (1) will only enhance the strongest edges while
greater values will also enhance small details, noise included.
γ: This parameter is not described in the article, it implements a gamma functionnal for the φ variable used in the article.
Γ: This parameter is also not described in the article, it allows for applying a global gamma correction
on the input image before processing. This gamma correction is reverted after processing on the output data.
Generally values arround 0.5 provide very nice results.