Mailman 3 October 2013 - scikit-image

Image deconvolution submission
by François Orieux Oct. 25, 2013

Oct. 25, 2013

Hi, I plan to submit image deconvolution algorithms to skimage. Before the pull request and peer review of the code I contact the mailing for a first reaction and comments. François Orieux -- IAP, CNRS-UPMC - Paris, France +33 1 73 77 55 11 http://research.orieux.fr

3 3

Re: Clarity
by Stéfan van der Walt Oct. 25, 2013

Oct. 25, 2013

On Fri, Oct 25, 2013 at 4:35 AM, Matthew Trentacoste <mt(a)matttrent.com> wrote: > Some of the best of breed work on this topic is by Sylvain Paris, a > researcher also at Adobe: > http://people.csail.mit.edu/sparis/publi/2011/siggraph/ That looks really neat! A little bit tongue in cheek: http://nbviewer.ipython.org/urls/gist.github.com/stefanv/7158075/raw/041808… Stéfan

1 0

Re: import error on linux
by Payal Gupta Oct. 25, 2013

Oct. 25, 2013

hello everyone... my problem solved... thanks to all.... :) :) On Friday, September 27, 2013 12:52:35 PM UTC+5:30, Payal Gupta wrote: > > hello.... > > i am using skimage library on raspberry pi which have linux platform but > i cant import skimage module on this. > > what do i do? > anybody know this prob. > reply >

3 2

Re: Clarity
by Matthew Trentacoste Oct. 25, 2013

Oct. 25, 2013

Doing the computation per RGB channel would introduce chromatic shifts. My guess is that chromatic scaling is probably uniform (either across channels or adjusting the saturation of the image in a non-spatially variant manner). Some of the best of breed work on this topic is by Sylvain Paris, a researcher also at Adobe: http://people.csail.mit.edu/sparis/publi/2011/siggraph/ You can use it to boost certain spatial scales to increase local contrast and reconstruct an image without artifacts. On Thu, Oct 24, 2013 at 6:45 PM, Juan Nunez-Iglesias <jni.soma(a)gmail.com>wrote: > Thanks Josh, Jerome, > > Actually what I'm really after is a good, *local* foreground/background > correction. The local bit is what attracted me to Clarity even though it's > not really what I'm after. > > Particularly of interest is something that will be robust to artifacts in > the images. > > Juan. > > On Fri, Oct 25, 2013 at 10:13 AM, Josh Warner <silvertrumpet999(a)gmail.com>wrote: > >> I think that's on the right track. >> >> But a naive contrast mask affects only the luminance. In the first >> example the colors are also very punched up; just looking this seems like >> more than a simple global contrast mask. >> >> My guess? They're applying the equivalent of a contrast mask, but to each >> RGB channel separately. This is "wrong" from an image processing >> perspective, but in the world of Instagram it seems people are more >> interested in filters than reality. >> >> I wouldn't be against this for advertising purposes, though I'd never do >> it to my own images. >> >> On Thursday, October 24, 2013 2:03:22 AM UTC-5, Jerome Kieffer wrote: >> >>> On Thu, 24 Oct 2013 17:27:55 +1100 >>> Juan Nunez-Iglesias <jni....(a)gmail.com> wrote: >>> >>> > Hi guys, >>> > >>> > Does anyone here have any ideas about what the Camera+ Clarity filter >>> is >>> > doing? Some examples of the filter >>> > here<https://secure.flickr.**com/photos/pierrewikberg/**6983939966/<https://secure.flickr.com/photos/pierrewikberg/6983939966/>>and >>> >>> > here <http://obamapacman.com/2011/**03/enhance-iphone-camera-** >>> clarity-video/<http://obamapacman.com/2011/03/enhance-iphone-camera-clarity-video/>>. >>> >>> > >>> > Might be a nice addition to skimage.exposure... =) >>> >>> Looks like a contrast mask: >>> http://www.gimp.org/tutorials/**ContrastMask/<http://www.gimp.org/tutorials/ContrastMask/> >>> >>> Here is an old implementation, starting from a JPEG: >>> def contrastMask(self, outfile): >>> """Ceci est un filtre de debouchage de photographies, aussi >>> appelé masque de contraste, >>> il permet de rattrapper une photo trop contrasté, un contre >>> jour, ... >>> Écrit par Jérôme Kieffer, avec l'aide de la liste python@aful, >>> en particulier A. Fayolles et F. Mantegazza avril 2006 >>> necessite numpy et PIL. >>> >>> @param: the name of the output file (JPEG) >>> @return: filtered Photo instance >>> """ >>> >>> try: >>> import numpy >>> # import scipy.signal as signal >>> except: >>> logger.error("This filter needs the numpy library available >>> on https://sourceforge.net/**projects/numpy/files/<https://sourceforge.net/projects/numpy/files/>") >>> >>> return >>> >>> t0 = time.time() >>> dimX, dimY = self.pil.size >>> >>> ImageFile.MAXBLOCK = dimX * dimY >>> img_array = numpy.fromstring(self.pil.**tostring(), >>> dtype="UInt8").astype("**float32") >>> img_array.shape = (dimY, dimX, 3) >>> red, green, blue = img_array[:, :, 0], img_array[:, :, 1], >>> img_array[:, :, 2] >>> #nota: this is faster than desat2=(ar.max(axis=2)+ar.min(**axis=2))/2 >>> >>> desat_array = (numpy.minimum(numpy.minimum(**red, green), blue) >>> + numpy.maximum(numpy.maximum(**red, green), blue)) / 2.0 >>> inv_desat = 255. - desat_array >>> blured_inv_desat = self._gaussian.blur(inv_desat, config.**ContrastMaskGaussianSize) >>> >>> bisi = numpy.round(blured_inv_desat).**astype("uint8") >>> k = Image.fromarray(bisi, "L").convert("RGB") >>> S = ImageChops.screen(self.pil, k) >>> M = ImageChops.multiply(self.pil, k) >>> F = ImageChops.add(ImageChops.**multiply(self.pil, S), >>> ImageChops.multiply(**ImageChops.invert(self.pil), M)) >>> exitJpeg = op.join(config.**DefaultRepository, outfile) >>> F.save(exitJpeg, quality=80, progressive=True, Optimize=True) >>> try: >>> os.chmod(exitJpeg, config.DefaultFileMode) >>> except IOError: >>> logger.error("Unable to chmod %s" % outfile) >>> exifJpeg = Exif(exitJpeg) >>> exifJpeg.read() >>> self.exif.copy(exifJpeg) >>> exifJpeg.comment = self.exif.comment >>> >>> -- >>> Jerome Kieffer <goo...(a)terre-adelie.org> >>> >> -- >> You received this message because you are subscribed to the Google Groups >> "scikit-image" group. >> To unsubscribe from this group and stop receiving emails from it, send an >> email to scikit-image+unsubscribe(a)googlegroups.com. >> For more options, visit https://groups.google.com/groups/opt_out. >> > > -- > You received this message because you are subscribed to the Google Groups > "scikit-image" group. > To unsubscribe from this group and stop receiving emails from it, send an > email to scikit-image+unsubscribe(a)googlegroups.com. > For more options, visit https://groups.google.com/groups/opt_out. > -- [ matthew m trentacoste ] [ ] [ mt(a)matttrent.com ] [ http://matttrent.com ] [ twitter: @matttrent ] [ +1.415.326.3226 ]

1 0

Re: Clarity
by Jerome Kieffer Oct. 25, 2013

Oct. 25, 2013

On Thu, 24 Oct 2013 17:27:55 +1100 Juan Nunez-Iglesias <jni.soma(a)gmail.com> wrote: > Hi guys, > > Does anyone here have any ideas about what the Camera+ Clarity filter is > doing? Some examples of the filter > here<https://secure.flickr.com/photos/pierrewikberg/6983939966/>and > here <http://obamapacman.com/2011/03/enhance-iphone-camera-clarity-video/>. > > Might be a nice addition to skimage.exposure... =) Looks like a contrast mask: http://www.gimp.org/tutorials/ContrastMask/ Here is an old implementation, starting from a JPEG: def contrastMask(self, outfile): """Ceci est un filtre de debouchage de photographies, aussi appelé masque de contraste, il permet de rattrapper une photo trop contrasté, un contre jour, ... Écrit par Jérôme Kieffer, avec l'aide de la liste python@aful, en particulier A. Fayolles et F. Mantegazza avril 2006 necessite numpy et PIL. @param: the name of the output file (JPEG) @return: filtered Photo instance """ try: import numpy # import scipy.signal as signal except: logger.error("This filter needs the numpy library available on https://sourceforge.net/projects/numpy/files/") return t0 = time.time() dimX, dimY = self.pil.size ImageFile.MAXBLOCK = dimX * dimY img_array = numpy.fromstring(self.pil.tostring(), dtype="UInt8").astype("float32") img_array.shape = (dimY, dimX, 3) red, green, blue = img_array[:, :, 0], img_array[:, :, 1], img_array[:, :, 2] #nota: this is faster than desat2=(ar.max(axis=2)+ar.min(axis=2))/2 desat_array = (numpy.minimum(numpy.minimum(red, green), blue) + numpy.maximum(numpy.maximum(red, green), blue)) / 2.0 inv_desat = 255. - desat_array blured_inv_desat = self._gaussian.blur(inv_desat, config.ContrastMaskGaussianSize) bisi = numpy.round(blured_inv_desat).astype("uint8") k = Image.fromarray(bisi, "L").convert("RGB") S = ImageChops.screen(self.pil, k) M = ImageChops.multiply(self.pil, k) F = ImageChops.add(ImageChops.multiply(self.pil, S), ImageChops.multiply(ImageChops.invert(self.pil), M)) exitJpeg = op.join(config.DefaultRepository, outfile) F.save(exitJpeg, quality=80, progressive=True, Optimize=True) try: os.chmod(exitJpeg, config.DefaultFileMode) except IOError: logger.error("Unable to chmod %s" % outfile) exifJpeg = Exif(exitJpeg) exifJpeg.read() self.exif.copy(exifJpeg) exifJpeg.comment = self.exif.comment -- Jerome Kieffer <google(a)terre-adelie.org>

3 2

Clarity
by Juan Nunez-Iglesias Oct. 24, 2013

Oct. 24, 2013

Hi guys, Does anyone here have any ideas about what the Camera+ Clarity filter is doing? Some examples of the filter here<https://secure.flickr.com/photos/pierrewikberg/6983939966/>and here <http://obamapacman.com/2011/03/enhance-iphone-camera-clarity-video/>. Might be a nice addition to skimage.exposure... =) Juan.

1 0

Re: CIELUV Transformations
by Isaac Gerg Oct. 23, 2013

Oct. 23, 2013

Good stuff -- thank you for sharing. On Wed, Oct 23, 2013 at 11:56 AM, Jaime Fernández del Río < jaime.frio(a)gmail.com> wrote: > I have a coworker that is a member of one of CIE's divisions, so I asked > him the LAB vs LUV question, and here's his reply: > > "Yes, there is a simple historical explanation - LUV was championed by the > display industry while LAB was the favorite of "surface" industries (print, > materials manufacturing, ...). LUV has the beneficial property of having a > saturation predictor and of being related to the luminance-linear u'v' > chromaticity space. Let me know if you'd like more detail. > > Today, I am not aware of LUV being used anywhere - LAB has won the battle > between the two and is used even for displays. Either that, or, where > higher performance is needed in terms of perceptual prediction accuracy, > CIECAM02 (a color appearance model) is used. It is significantly more > complex, but predicts more visual phenomena and allows for taking viewing > conditions better into account. > > So, I'd say use LAB if what it does is good enough and CIECAM02 if it > fails (e.g., for gamut mapping LAB has serious hue non-uniformity)." > > So, who's up for coding CIECAM02 color transformations, whatever that may > be? ;-) > > Jaime > > -- > (\__/) > ( O.o) > ( > <) Este es Conejo. Copia a Conejo en tu firma y ayúdale en sus planes > de dominación mundial. > > -- > You received this message because you are subscribed to a topic in the > Google Groups "scikit-image" group. > To unsubscribe from this topic, visit > https://groups.google.com/d/topic/scikit-image/DIRaSXJoEes/unsubscribe. > To unsubscribe from this group and all of its topics, send an email to > scikit-image+unsubscribe(a)googlegroups.com. > For more options, visit https://groups.google.com/groups/opt_out. >

1 0

Re: CIELUV Transformations
by Jaime Fernández del Río Oct. 23, 2013

Oct. 23, 2013

I have a coworker that is a member of one of CIE's divisions, so I asked him the LAB vs LUV question, and here's his reply: "Yes, there is a simple historical explanation - LUV was championed by the display industry while LAB was the favorite of "surface" industries (print, materials manufacturing, ...). LUV has the beneficial property of having a saturation predictor and of being related to the luminance-linear u'v' chromaticity space. Let me know if you'd like more detail. Today, I am not aware of LUV being used anywhere - LAB has won the battle between the two and is used even for displays. Either that, or, where higher performance is needed in terms of perceptual prediction accuracy, CIECAM02 (a color appearance model) is used. It is significantly more complex, but predicts more visual phenomena and allows for taking viewing conditions better into account. So, I'd say use LAB if what it does is good enough and CIECAM02 if it fails (e.g., for gamut mapping LAB has serious hue non-uniformity)." So, who's up for coding CIECAM02 color transformations, whatever that may be? ;-) Jaime -- (\__/) ( O.o) ( > <) Este es Conejo. Copia a Conejo en tu firma y ayúdale en sus planes de dominación mundial.

1 0

Re: CIELUV Transformations
by Isaac Gerg Oct. 23, 2013

Oct. 23, 2013

Unfortunately, I do not know. You could check out Principles of Color Technology by Berns and see if he talks more about it in his book. Additionally, you could email burns and ask him. I dont have the time at the moment to track this down. Isaac

1 0

Re: CIELUV Transformations
by Stéfan van der Walt Oct. 23, 2013

Oct. 23, 2013

On Wed, Oct 23, 2013 at 3:25 PM, Isaac Gerg <isaac.gerg(a)gergltd.com> wrote: > As I understand it, they simply could not agree on a single standard and > felt both must be approved. That is as much as i can get out of the > literature. But were arguments made on both sides that were so compelling, or did they have different trade-offs? I'm just trying to get a grip on the situation. Stéfan

1 0