missing kernel or block size in equalize_adapthist
<https://lh3.googleusercontent.com/-0jVbtyT6ArM/VWNtiWHECgI/AAAAAAAABOg/AKpRcr4fKHU/s1600/nuclei_equalize_adapthist.jpg> Hello, I am using equalize_adapthist after color deconvolution and some morphological operations to enhance local contrast of nuclei in histological images. There are different results when applying equalize_adapthist when using small patches and larger size whole images (left small patch, right large size whole image). The CLAHE implementation in ImageJ has a 'block size' parameter and the implementation in Pixinsight has a 'Kernel size' parameter to get results depending on the local context. Is it possible to 'simulate' this behaviour with scikit-image? Best regards, Kai
Hi Kai, Do the `n_tiles*` arguments not meet your needs? By defining the number of tiles in X and Y, you are equivalently setting a block size. http://scikit-image.org/docs/stable/api/skimage.exposure.html#equalize-adapt... Regards, Steve On Monday, May 25, 2015 at 1:49:53 PM UTC-5, Kai Wiechen wrote:
I am using equalize_adapthist after color deconvolution and some morphological operations to enhance local contrast of nuclei in histological images. There are different results when applying equalize_adapthist when using small patches and larger size whole images (left small patch, right large size whole image). The CLAHE implementation in ImageJ has a 'block size' parameter and the implementation in Pixinsight has a 'Kernel size' parameter to get results depending on the local context. Is it possible to 'simulate' this behaviour with scikit-image?
Best regards,
Kai
Hi Steve, but these parameters are limited to the range 1..16. The small test patches are 150x150 pixels, the complete images have 1920x1448 pixels. I have applied equalize_adapthist here with default parameters (n_tiles=8). Best regards, Kai On Tuesday, May 26, 2015 at 12:54:34 AM UTC+2, Steven Silvester wrote:
Hi Kai,
Do the `n_tiles*` arguments not meet your needs? By defining the number of tiles in X and Y, you are equivalently setting a block size.
http://scikit-image.org/docs/stable/api/skimage.exposure.html#equalize-adapt...
Regards,
Steve
On Monday, May 25, 2015 at 1:49:53 PM UTC-5, Kai Wiechen wrote:
I am using equalize_adapthist after color deconvolution and some morphological operations to enhance local contrast of nuclei in histological images. There are different results when applying equalize_adapthist when using small patches and larger size whole images (left small patch, right large size whole image). The CLAHE implementation in ImageJ has a 'block size' parameter and the implementation in Pixinsight has a 'Kernel size' parameter to get results depending on the local context. Is it possible to 'simulate' this behaviour with scikit-image?
Best regards,
Kai
Hi Kai, Is the issue that you want finer grain control of the block size? If so, that would require a substantial rewrite. Please feel free to open an issue on Github if that is the case. Regards, Steve
On May 26, 2015, at 1:58 AM, Kai Wiechen <kwiechen1@gmail.com> wrote:
Hi Steve,
but these parameters are limited to the range 1..16. The small test patches are 150x150 pixels, the complete images have 1920x1448 pixels. I have applied equalize_adapthist here with default parameters (n_tiles=8).
Best regards,
Kai
On Tuesday, May 26, 2015 at 12:54:34 AM UTC+2, Steven Silvester wrote: Hi Kai,
Do the `n_tiles*` arguments not meet your needs? By defining the number of tiles in X and Y, you are equivalently setting a block size.
http://scikit-image.org/docs/stable/api/skimage.exposure.html#equalize-adapt... <http://scikit-image.org/docs/stable/api/skimage.exposure.html#equalize-adapthist>
Regards,
Steve
On Monday, May 25, 2015 at 1:49:53 PM UTC-5, Kai Wiechen wrote: <https://lh3.googleusercontent.com/-0jVbtyT6ArM/VWNtiWHECgI/AAAAAAAABOg/AKpRcr4fKHU/s1600/nuclei_equalize_adapthist.jpg>Hello,
I am using equalize_adapthist after color deconvolution and some morphological operations to enhance local contrast of nuclei in histological images. There are different results when applying equalize_adapthist when using small patches and larger size whole images (left small patch, right large size whole image). The CLAHE implementation in ImageJ has a 'block size' parameter and the implementation in Pixinsight has a 'Kernel size' parameter to get results depending on the local context. Is it possible to 'simulate' this behaviour with scikit-image?
Best regards,
Kai
-- 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/2elntOClGOY/unsubscribe <https://groups.google.com/d/topic/scikit-image/2elntOClGOY/unsubscribe>. To unsubscribe from this group and all its topics, send an email to scikit-image+unsubscribe@googlegroups.com <mailto:scikit-image+unsubscribe@googlegroups.com>. For more options, visit https://groups.google.com/d/optout <https://groups.google.com/d/optout>.
Hi Steve, yes, I would like to have finer control of block/kernel size to obtain similar results if I apply this to small test patches or the larger complete image. Is there any reason to limit ntiles_x/y to 16? Regards, Kai Am Mittwoch, 27. Mai 2015 03:17:01 UTC+2 schrieb Steven Silvester:
Hi Kai,
Is the issue that you want finer grain control of the block size? If so, that would require a substantial rewrite.
Please feel free to open an issue on Github if that is the case.
Regards,
Steve
On May 26, 2015, at 1:58 AM, Kai Wiechen <kwie...@gmail.com <javascript:>> wrote:
Hi Steve,
but these parameters are limited to the range 1..16. The small test patches are 150x150 pixels, the complete images have 1920x1448 pixels. I have applied equalize_adapthist here with default parameters (n_tiles=8).
Best regards,
Kai
On Tuesday, May 26, 2015 at 12:54:34 AM UTC+2, Steven Silvester wrote:
Hi Kai,
Do the `n_tiles*` arguments not meet your needs? By defining the number of tiles in X and Y, you are equivalently setting a block size.
http://scikit-image.org/docs/stable/api/skimage.exposure.html#equalize-adapt...
Regards,
Steve
On Monday, May 25, 2015 at 1:49:53 PM UTC-5, Kai Wiechen wrote:
I am using equalize_adapthist after color deconvolution and some morphological operations to enhance local contrast of nuclei in histological images. There are different results when applying equalize_adapthist when using small patches and larger size whole images (left small patch, right large size whole image). The CLAHE implementation in ImageJ has a 'block size' parameter and the implementation in Pixinsight has a 'Kernel size' parameter to get results depending on the local context. Is it possible to 'simulate' this behaviour with scikit-image?
Best regards,
Kai
-- 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/2elntOClGOY/unsubscribe. To unsubscribe from this group and all its topics, send an email to scikit-image...@googlegroups.com <javascript:>. For more options, visit https://groups.google.com/d/optout.
Hi Kai, The limits were set based on the original implementation, but there is no reason they need to stay at 16. Using an arbitrary size tile is what would take a substantial rewrite. Looking at your numbers, it looks like a 13x10 block size will get you ~150x150 tile size. At the full 16x16, you'd have 120x90 pixels. Regards, Steve On Thu, May 28, 2015 at 1:35 PM, Kai Wiechen <kwiechen1@gmail.com> wrote:
Hi Steve,
yes, I would like to have finer control of block/kernel size to obtain similar results if I apply this to small test patches or the larger complete image. Is there any reason to limit ntiles_x/y to 16?
Regards,
Kai
Am Mittwoch, 27. Mai 2015 03:17:01 UTC+2 schrieb Steven Silvester:
Hi Kai,
Is the issue that you want finer grain control of the block size? If so, that would require a substantial rewrite.
Please feel free to open an issue on Github if that is the case.
Regards,
Steve
On May 26, 2015, at 1:58 AM, Kai Wiechen <kwie...@gmail.com> wrote:
Hi Steve,
but these parameters are limited to the range 1..16. The small test patches are 150x150 pixels, the complete images have 1920x1448 pixels. I have applied equalize_adapthist here with default parameters (n_tiles=8).
Best regards,
Kai
On Tuesday, May 26, 2015 at 12:54:34 AM UTC+2, Steven Silvester wrote:
Hi Kai,
Do the `n_tiles*` arguments not meet your needs? By defining the number of tiles in X and Y, you are equivalently setting a block size.
http://scikit-image.org/docs/stable/api/skimage.exposure.html#equalize-adapt...
Regards,
Steve
On Monday, May 25, 2015 at 1:49:53 PM UTC-5, Kai Wiechen wrote:
I am using equalize_adapthist after color deconvolution and some morphological operations to enhance local contrast of nuclei in histological images. There are different results when applying equalize_adapthist when using small patches and larger size whole images (left small patch, right large size whole image). The CLAHE implementation in ImageJ has a 'block size' parameter and the implementation in Pixinsight has a 'Kernel size' parameter to get results depending on the local context. Is it possible to 'simulate' this behaviour with scikit-image?
Best regards,
Kai
-- 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/2elntOClGOY/unsubscribe. To unsubscribe from this group and all its topics, send an email to scikit-image...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
--
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/2elntOClGOY/unsubscribe. To unsubscribe from this group and all its topics, send an email to scikit-image+unsubscribe@googlegroups.com. For more options, visit https://groups.google.com/d/optout.
Hi Kai, We are having a discussion now about how to fix/improve the adapthist block handling: https://github.com/scikit-image/scikit-image/issues/1541. Regards, Steve On Monday, May 25, 2015 at 1:49:53 PM UTC-5, Kai Wiechen wrote:
I am using equalize_adapthist after color deconvolution and some morphological operations to enhance local contrast of nuclei in histological images. There are different results when applying equalize_adapthist when using small patches and larger size whole images (left small patch, right large size whole image). The CLAHE implementation in ImageJ has a 'block size' parameter and the implementation in Pixinsight has a 'Kernel size' parameter to get results depending on the local context. Is it possible to 'simulate' this behaviour with scikit-image?
Best regards,
Kai
participants (2)
-
Kai Wiechen -
Steven Silvester