[IPOL discuss] demo RGBprocess (RGB cube visualization)

[José Luis Lisani]

Nicolas,
> A few comments, from the (naive) user point of view:
>
> # input
>
> * All the proposed images are labelled "<something>  image". The
>    "<image>" part is obvious, I thing"<something>" is enough (Balls,
>    Horses, ...)
> * It's not "buda" (the city), it's not "buddha" (the Buddhism
>    founder), it's a "monk".
> * It's not a "lake", it's "kinkaku-ji" (or the "golden pavilion").
> * It's not a "star", it's a "starfish"

Thank you for the remarks,
in fact the names are provisional (the "monk" image is in fact image 
376020.jpg in Berkeley train dataset).
I plan to modify the input.html template so we can include the credits 
of the images.

> # params/subimage
>
> * I'd like to have both on the same page, ie, on the "params" page,
>    either click a button to run the algorithm on the whole image, or
>    click two points on the image to select a rectangle.

I need your help on that point.
I just followed the example in

  http://mw.cmla.ens-cachan.fr/megawave/demo/test_subimage/

where subimages are selected in two steps.
How can I click on two points of the image and send this information in 
a single html form?

> * If I select the first corner on a red background, I can't see the
>    red cross drawn at this position. A solution is to draw this cross
>    with 2 colors (white and black, for example), like the ascii example
>    hereafter
>             B
>             B
>             W
>        BBBWWWWWBBB
>             W
>             B
>             B
OK
> * The bilinear interpolation used to scale up tiny images is
>    disturbing. Why is this interpolation used? If you need more pixels,
>    can't you extend the selected rectangle instead of zooming it?
Bilinear interpolation is used because it preserves the order between 
the gray levels of the
image, thus preserving their "geometry".

Zoom is used to get a denser set of color points in the RGB cube, so we 
see better the distribution
of colors in the selected subimage.
> # results
>
> * "view 1" ... "view 21" are not meaningful titles. Maybe the camera
>    position, or rotation angles, can be better.
I don't think this information is really useful for the user. It is 
enough for him to know
that these are different views of the color cube.
> * "RGB views: original, filtered, density view" could be a subtitle
>    (<h2>...</h2>).
OK
> * We need a subtitle for the first part of the results
>    (original/filtered), and no separation (<hr>) between the two parts
>    of the results.
OK
> * I feel that one rotation axis is enough to understand the 3D layout,
>    ie views 6-21; but I may not understand correctly what is to be
>    seen.
Views 1-5 show that the color clusters are (approximately) contained in 
planes of constant hue (planes that
contain the black-white axis of the cube), so it is important to display 
them.
> * Some users may want to get the 3D data. You could produce a 3D data
>    file (VRML[1], X3D[2], PLY[3], ...). Later, when the technology
>    improves, we may embed this 3D data into the web page with
>    HTML5 and WebGL.
>
> [1]http://en.wikipedia.org/wiki/VRML
> [2]http://en.wikipedia.org/wiki/X3D
> [3]http://en.wikipedia.org/wiki/PLY_(file_format)
I will use the PLY file generator in Julie Digne code, as suggested by 
Pascal Monasse
> # misc
>
> * What vabout seeing the color distribution in other coordinate
>    systems, ie HSL instead of RGB for example. Is it interesting?
In our paper (ref. [1] in 
https://edit.ipol.im/edit/algo/blm_color_dimensional_filtering/ ) we 
justify the use
of RGB representation instead of other color model.


José Luis