Version 0.16¶

The following functions are deprecated and will be removed in 0.18: skimage.measure.compare_mse, skimage.measure.compare_nrmse, skimage.measure.compare_pnsr, skimage.measure.compare_ssim Their functionality still exists, but under the new skimage.metrics submodule under different names.
Additionally, three new functions have been added to skimage.metrics: skimage.metrics.variation_of_information skimage.metrics.adapted_rand_error skimage.metrics.contingency_table
A new example of plotting these evaluation metrics has been added to the docs.

Version 0.15¶

skimage.feature.canny now uses a more accurate Gaussian filter internally; output values will be different from 0.14.
skimage.filters.threshold_niblack and skimage.filters.threshold_sauvola now accept a tuple as window_size besides integers.

skimage.filters.gaussian_filter has been removed. Use skimage.filters.gaussian instead.
skimage.filters.gabor_filter has been removed. Use skimage.filters.gabor instead.
The old syntax support for skimage.transform.integrate has been removed.
The normalise parameter of skimage.feature.hog was removed due to incorrect behavior: it only applied a square root instead of a true normalization. If you wish to duplicate the old behavior, set transform_sqrt=True.
skimage.measure.structural_similarity has been removed. Use skimage.measure.compare_ssim instead.
In skimage.measure.compare_ssim, the dynamic_range has been removed in favor of ‘data_range.
In skimage.restoration.denoise_bilateral, the sigma_range kwarg has been removed in favor of sigma_color.
skimage.measure.marching_cubes has been removed in favor of skimage.measure.marching_cubes_lewiner.
ntiles_* parameters have been removed from skimage.exposure.equalize_adapthist. Use kernel_size instead.
skimage.restoration.nl_means_denoising has been removed in favor of skimage.restoration.denoise_nl_means.
skimage.measure.LineModel has been removed in favor of skimage.measure.LineModelND.
In skimage.feature.hog visualise has been changed to visualize.
freeimage plugin of skimage.io has been removed.

skimage.filter has been removed. Use skimage.filters instead.
skimage.filters.canny has been removed. canny is available only from skimage.feature now.
Deprecated filters hsobel, vsobel, hscharr, vscharr, hprewitt, vprewitt, roberts_positive_diagonal, roberts_negative_diagonal have been removed from skimage.filters.edges.
The sigma parameter of skimage.filters.gaussian and the selem parameter of skimage.filters.median have been made optional, with default values.
The clip_negative parameter of skimage.util.dtype_limits is now set to None by default, equivalent to True, the former value. In version 0.15, will be set to False.
The circle parameter of skimage.transform.radon and skimage.transform.iradon are now set to None by default, equivalent to False, the former value. In version 0.15, will be set to True.
Parameters ntiles_x, ntiles_y have been removed from skimage.exposure.equalize_adapthist.
The freeimage io plugin is no longer supported, and will use imageio instead. We will completely remove the freeimage plugin in Version 0.14.

equalize_adapthist now takes a kernel_size keyword argument, replacing the ntiles_* arguments.
The functions blob_dog, blob_log and blob_doh now return float arrays instead of integer arrays.
transform.integrate now takes lists of tuples instead of integers to define the window over which to integrate.
reverse_map parameter in skimage.transform.warp has been removed.
enforce_connectivity in skimage.segmentation.slic defaults to True.
skimage.measure.fit.BaseModel._params, skimage.transform.ProjectiveTransform._matrix, skimage.transform.PolynomialTransform._params, skimage.transform.PiecewiseAffineTransform.affines_* attributes have been removed.
skimage.filters.denoise_* have moved to skimage.restoration.denoise_*.
skimage.data.lena has been removed.

The skimage.filter subpackage has been renamed to skimage.filters.
Some edge detectors returned values greater than 1–their results are now appropriately scaled with a factor of sqrt(2).

No longer wrap imread output in an Image class
Change default value of sigma parameter in skimage.segmentation.slic to 0
hough_circle now returns a stack of arrays that are the same size as the input image. Set the full_output flag to True for the old behavior.
The following functions were deprecated over two releases: skimage.filter.denoise_tv_chambolle, skimage.morphology.is_local_maximum, skimage.transform.hough, skimage.transform.probabilistic_hough,skimage.transform.hough_peaks. Their functionality still exists, but under different names.