Changes DIPlib 3.3.0

Changes to DIPlib

New functionality

• Added dip::MakeRegionsConvex2D(), which works for both labelled and binary images.

• Added Perimeter() (as an alias for Length()), Reverse(), Rotate(), Scale(), and Translate() as member functions to dip::Polygon.

• dip::MomentAccumulator has a new method PlainSecondOrder(), which returns the plain old second order central moments, compared to the moment of inertia tensor returned by SecondOrder().

• Added dip::Image::Mask(), to apply a mask to an image.

• Added dip::DimensionArray<>::reverse() and dip::PixelSize::Reverse() to reverse the elements of an array.

• Added dip::Image::ReverseDimensions(), which permutes dimensions to reverse their order.

• dip::Label has a new string argument, mode, which can be set to "largest" to return a labeled image that only contains the largest connected component in the input image.

• Added dip::PixelSize::UnitLength(), dip::PixelSize::UnitSize(), dip::PixelSize::ForcePhysical(), and dip::PixelSize::SameUnits() to simplify finding the right units for measurement results.

• Added operators to multiply and divide two dip::Vertex objects. They apply an element-wise product.

• dip::ChainCode::Length has an optional parameter to include the image boundary pixels in the measurement.

• The "Perimeter" feature can be configured: setting the "include boundary pixels" parameter to a non-zero value causes the perimeter measurement to include pixels at the image boundary.

• dip::TriangleThreshold() and dip::BackgroundThreshold() have a new parameter sigma, which defaults to 4 to preserve the old behavior.

Changed functionality

• The deterministic initialization for dip::GaussianMixtureModel() is more robust, making the initial Gaussians overlap instead of setting their sigma to 1.

• dip::ConvexHull no longer holds a dip::Polygon as a private member, instead it uses dip::Polygon as a base class. This makes all of the dip::Polygon functionality directly available on the convex hull. dip::ConvexHull::Polygon() now just returns a base class reference.

• dip::Units (and by extension dip::PhysicalQuantity and dip::PixelSize) now uses the Greek letter μ (U+03BC), instead of the legacy symbol µ (U+00B5) for micron (when DIPlib is built with Unicode enabled). Both symbols are recognized when parsing a string, but the strings generated are now different.

• dip::BackgroundThreshold() now determines the half width at half height with sub-sample precision, and takes the smoothing of the histogram into account, such that the amount of smoothing should have little influence in the computed threshold.

• The “Mu” and “GreyMu” features no longer use the pixel sizes if the units for different image dimensions differ. No uses of the tensor make sense if the units for the various dimensions don’t match, in this case it is more useful to report the tensor in pixel units.

Bug fixes

• dip::DrawPolygon2D(), when drawing filled polygons, would skip the bottom row in the polygon. The algorithm is a bit more sophisticated now to properly handle these bottom rows. This also takes care of some rounding errors that could be seen for polygons with very short edges.

• dip::ResampleAt with a map input argument and using "cubic" interpolation could overflow, yielding unsightly artifacts. See issue #107.

• Better error messages for some forms of dip::Image::View::At()

• dip::ImageDisplay() didn’t pay attention to the dim1 and dim2 parameters if the image was 2D.

• dip::DefineROI() was incorrect if the input and output images were the same object.

• dip::GaussianMixtureModel() could produce NaN for amplitude, those components now have a zero amplitude.

• The “SolidArea” feature didn’t take the pixel size into account. This also caused the “Roundness” feature to report wrong values. The “SurfaceArea” feature didn’t properly scale the result by the pixel size. Several composed features (computed from other features) didn’t produce correct values for anisotropic pixels (“P2A”, “Roundness”, and “PodczeckShapes”).

• dip::div_round() was incorrect, which caused dip::DirectedPathOpening() to not use certain directions.

• dip::DirectionalStatisticsAccumulator::StandardDeviation() returned NaN if the mean was identical to zero.

Changes to DIPimage

Changed functionality

• threshold(.., 'triangle') now accepts a parameter to control how much smoothing to apply.

(See also changes to DIPlib.)

Bug fixes

None, but see bugfixes to DIPlib.

Changes to PyDIP

New functionality

• Added dip.MakeRegionsConvex2D().

• Added Perimeter() (as an alias for Length()), Reverse(), Rotate(), Scale(), and Translate() as methods to dip.Polygon.

• dip.Polygon and dip.ChainCode now both have __len__, __getitem__ and __iter__ methods, meaning that they can be treated like lists or other iterables. Previously, the values of dip.Polygon had to be extracted by conversion to a NumPy array, and the values of dip.ChainCode by copying to a list when accessing its codes property.

• The structure returned by dip.Moments() has a new component plainSecondOrder, which contains the plain old second order central moments, compared to the moment of inertia tensor contained in secondOrder.

• Added Mask() and ReverseDimensions() as methods to dip.Image.

• Added dip.ReverseDimensions(), which reverses the indexing order for images for the remainder of the session. Has repercussion on how the dip.Image buffer is exposed, buffer protocol objects are converted to a dip.Image, files are read and written, and how DIPviewer displays images. It is intended to make indexing into a dip.Image match the indexing into the corresponding NumPy array, which should make it easier to mix calls to DIPlib and scikit-image in the same program. Note that this also causes positive angles to be counter-clockwise instead of clockwise.

• Added the @ and @= operators for dip.Image objects. These apply matrix multiplication if both operands are non-scalar images. That is, the vector or matrix at each pixel is multiplied by the vector or matrix at the corresponding pixel in the other operand. The two image’s tensor dimensions must be compatible. If one of the operands is a scalar image, the normal element-wise multiplication is applied. One of the operands can be a single pixel (a list of numbers).

• Added dip.MeasurementTool.Configure to allow measurement features to be configured.

• Added AspectRatio(), SameUnits(), UnitLength(), UnitSize(), ForcePhysical(), and ApproximatelyEquals() as methods to dip.PixelSize.

Changed functionality

• Operators overloaded for dip.Image objects can use lists of numbers as a second argument, which is interpreted as a 0D tensor image (column vector). This makes img / img[0] possible.

• When PyDIPjavaio fails to load, the error is no longer displayed immediately. Instead, it is shown when dip.ImageRead() fails. The error message is also a bit more helpful. See issue #106.

• The __repr__ string for many classes has changed to be more consistent and informative.

• The * and *= operators have changed meaning, they now always apply element-wise multiplication, their previous behavior is now obtained with the new @ and @= operators. NOTE! This breaks backwards compatibility. To keep old code working that depends on image matrix multiplications, you can do

  import diplib as dip
  dip.Image.__mul__ = dip.Image.__matmul__
  dip.Image.__rmul__ = dip.Image.__rmatmul__
  dip.Image.__imul__ = dip.Image.__imatmul__
  

  But we recommend instead that you update the code to use the right operators.

(See also changes to DIPlib.)

Bug fixes

• __len__() now properly returns 0 for an empty (raw) image. This makes if img fail for a raw image.

• Fixed a few issues with indexing into image, allowing using a list of coordinates, and allowing assigning a scalar value into a multi-channel image.

• dip.SubpixelMaxima() and dip.SubpixelMinima(), when a mask image was given, didn’t work correctly.

• The **= operator did the computations correctly, but then assigned None to the variable instead of the result of the operation. The ** operator didn’t work with a single pixel (a number or a list) on the left-hand side.

(See also bugfixes to DIPlib.)

Changes to DIPviewer

New functionality

• Added option to change axis labels.

Bug fixes

• Automatically close all PyDIPviewer windows at program exit to avoid segfault.

Changes to DIPjavaio

None.