You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
Is your feature request related to a problem? Please describe.
In some situations, like identifying bright embryos within seed xray images, it would be great to have a radial version of analyze.distribution, such that the output would describe the count of nonzero pixels (or average value?) in bins that specified distance from the center of a labeled object.
Describe the solution you'd like
Either a "radial" category for the direction parameter in the existing analyze.distribution function or a separate analyze.distribution_radial function that does what is described above. The output would be similar, a histogram for visualization and outputs for either count or average value in bins. Because labeled objects will not be perfectly round, there should probably also be a cutoff option for where the distribution stops to account for strange things at the edges. Also, the count would probably need to be a percentage, since bins will be uneven (there will be more pixels in farther bins).
The text was updated successfully, but these errors were encountered:
Is your feature request related to a problem? Please describe.
In some situations, like identifying bright embryos within seed xray images, it would be great to have a radial version of analyze.distribution, such that the output would describe the count of nonzero pixels (or average value?) in bins that specified distance from the center of a labeled object.
Describe the solution you'd like
Either a "radial" category for the direction parameter in the existing analyze.distribution function or a separate analyze.distribution_radial function that does what is described above. The output would be similar, a histogram for visualization and outputs for either count or average value in bins. Because labeled objects will not be perfectly round, there should probably also be a cutoff option for where the distribution stops to account for strange things at the edges. Also, the count would probably need to be a percentage, since bins will be uneven (there will be more pixels in farther bins).
The text was updated successfully, but these errors were encountered: