Hey Chris, I wonder if you would be interested in seeing more color space
conversions being added to the colorconv.py module, or if you are deliberately keeping it simple? For instance, I would like to add conversions to and from CIE LAB, and to allow alternative white points.
+1 on more color conversions!
I'm also interested in seeing conversions to and from human cone LMS spaces available somewhere for SciPy users. I'm not sure whether or not this kind of thing would be appropriate for scikits.image.
LMS is definitely interesting. We also have some code around to reproduce other human-like color spaces (e.g. opponent space from Itti & Walther, see Walther's PhD Thesis, Appendix A.2 http://bit.ly/deN6iU), as well as all the color spaces used in Geusebroek & van de Sande (as in http://bit.ly/bKg9nBor http://bit.ly/bYLF21). Would you be interested in these? I'll be happy to add them too. Please feel free to contribute any color space you'd like, possibly using Cython if you need speed. Thanks. Cheers, N
I thought I should first post a short message about this before starting anything. I'm aware that the SciPy image sandbox used to contain more conversions, probably in a slightly confusing way (e.g.http://projects.scipy.org/scipy/browser/trunk/Lib/sandbox/image/ color.py?rev=1698<http://projects.scipy.org/scipy/browser/trunk/Lib/sandbox/image/color.py?rev...>). Anyway, please let me know what you think.
Thanks, Chris
-- Nicolas Pinto Ph.D. Candidate, Brain & Computer Sciences Massachusetts Institute of Technology, USA http://web.mit.edu/pinto
On Sat, Jul 3, 2010 at 12:11 AM, Nicolas Pinto <nicolas.pinto@gmail.com>wrote:
Hey Chris,
I wonder if you would be interested in seeing more color space
conversions being added to the colorconv.py module, or if you are deliberately keeping it simple? For instance, I would like to add conversions to and from CIE LAB, and to allow alternative white points.
+1 on more color conversions!
I'm also interested in seeing conversions to and from human cone LMS spaces available somewhere for SciPy users. I'm not sure whether or not this kind of thing would be appropriate for scikits.image.
LMS is definitely interesting. We also have some code around to reproduce other human-like color spaces (e.g. opponent space from Itti & Walther, see Walther's PhD Thesis, Appendix A.2 http://bit.ly/deN6iU), as well as all the color spaces used in Geusebroek & van de Sande (as in http://bit.ly/bKg9nB or http://bit.ly/bYLF21). Would you be interested in these? I'll be happy to add them too.
Please feel free to contribute any color space you'd like, possibly using Cython if you need speed.
Thanks.
Cheers,
N
I thought I should first post a short message about this before starting anything. I'm aware that the SciPy image sandbox used to contain more conversions, probably in a slightly confusing way (e.g.http://projects.scipy.org/scipy/browser/trunk/Lib/sandbox/image/ color.py?rev=1698<http://projects.scipy.org/scipy/browser/trunk/Lib/sandbox/image/color.py?rev...>). Anyway, please let me know what you think.
When I looked at those I reused only the conversions that were unambiguously defined and did not need to make too many assumptions. We discussed it here: http://groups.google.com/group/scikits-image/browse_thread/thread/4b4e69fd0dbce9cc/7db21c67738edeed?lnk=gst&q=color+spaces#7db21c67738edeed
More color spaces is good, but it's important to reference everything very well and also decide on the scaling for float/uint etc. Cheers, Ralf
On 3 July 2010 07:13, Ralf Gommers <ralf.gommers@googlemail.com> wrote:
More color spaces is good, but it's important to reference everything very well and also decide on the scaling for float/uint etc.
This is a good point, one that Chris and I also discussed yesterday. Here's a proposed guideline: 1. Functions should allow the following types of input images, unless explicitly documented otherwise: float, double uint8 uint16 2. Functions should output the same type of output image, unless explicitly documented otherwise. 3. Floating point images must be [0, 1]. Uint8 images are [0, 255] and Uint16 images [0, 65535]. Does that sound reasonable? We can write a test decorator to ensure that these requirements are upheld. Regards Stéfan
2010/7/3 Stéfan van der Walt <stefan@sun.ac.za>
On 3 July 2010 07:13, Ralf Gommers <ralf.gommers@googlemail.com> wrote:
More color spaces is good, but it's important to reference everything very well and also decide on the scaling for float/uint etc.
This is a good point, one that Chris and I also discussed yesterday. Here's a proposed guideline:
1. Functions should allow the following types of input images, unless explicitly documented otherwise:
float, double uint8 uint16
2. Functions should output the same type of output image, unless explicitly documented otherwise.
3. Floating point images must be [0, 1]. Uint8 images are [0, 255] and Uint16 images [0, 65535].
Does that sound reasonable? We can write a test decorator to ensure that these requirements are upheld.
Clear guidelines and as little magic behind the scenes sounds good to me. Some separate type conversion / scaling utilities may be useful to provide.
Cheers, Ralf
i would prefer the following types: uint8 int32 float double purely for the fact that a sobel filter must return signed values, and it doesnt make sense to upcast a uint8 to a float. On 07/03/2010 10:47 AM, St�fan van der Walt wrote:
On 3 July 2010 07:13, Ralf Gommers <ralf.gommers@googlemail.com> wrote:
More color spaces is good, but it's important to reference everything very well and also decide on the scaling for float/uint etc.
This is a good point, one that Chris and I also discussed yesterday. Here's a proposed guideline:
1. Functions should allow the following types of input images, unless explicitly documented otherwise:
float, double uint8 uint16
2. Functions should output the same type of output image, unless explicitly documented otherwise.
3. Floating point images must be [0, 1]. Uint8 images are [0, 255] and Uint16 images [0, 65535].
Does that sound reasonable? We can write a test decorator to ensure that these requirements are upheld.
Regards St�fan
participants (4)
-
Nicolas Pinto -
Ralf Gommers -
S. Chris Colbert -
Stéfan van der Walt