Changes DIPlib 3.5.2

Changes to DIPlib

New functionality

• dip::ImageWriteTiff() now can write 3D images to a TIFF file as a series of slices. See PR #182.

• Additions to the dip::Graph class:

  • Added dip::Graph::EdgeVertex() for convenience. graph.EdgeVertex( edge, which ) is the same as what previously was written graph.Edges()[ edge ].vertices[ which ].

  • Added dip::Graph::IsValidEdge().

  • Added dip::Graph::UpdateEdgeWeights<>(), an overload that takes a function as input; this function is applied to the two vertex weights for each edge, and should return an edge weight.

  • dip::MinimumSpanningForest() is now a free function. The dip::Graph::MinimumSpanningForest() class function still exists for backwards-compatibility, it calls the free function.

• Added dip::DirectedGraph, a directed version of dip::Graph.

• Added dip::GraphCut(), a function that computes the minimum cut of a dip::DirectedGraph. This is a segmentation algorithm that splits the graph into two sections based on two marker vertices (nodes).

• Added dip::GraphCut(), a function that computes the grap-cut segmentation of an image.

• Added dip::Label() with a dip::Graph and a dip::DirectedGraph as input. It finds connected components in the graph.

• Added dip::Histogram::Configuration::Mode::IS_COMPLETE, which prevents a configuration from being modified when computing a histogram. It is set by dip::Histogram::Configuration::Complete(). This option is dangerous to set manually! It is intended for re-using the same configuration across multiple histogram computations.

• Added dip::Image::ReinterpretCastBinToUint8() and dip::Image::ReinterpretCastUint8ToBin() for the specific case of casting bin to uint8 and back. They don’t provide any functionality that dip::Image::ReinterpretCast() didn’t already provide, but they are much safer to use.

Changed functionality

• dip::AlignedAllocInterface now aligns each of the scanlines (rows of the image), not just the first one. This means it can be used to create images that are compatible with some image display widgets (Windows bitmap, Qt image, etc.).

• dip::LowestCommonAncestorSolver is no longer in the public API. This class contained code used in the Exact Stochastic Watershed (dip::StochasticWatershed with seeds set to "exact" or nIterations set to 0).

Bug fixes

• dip::Log2 computed the natural logarithm instead of the base-2 logarithm. See PR #168.

• dip::StructureTensorAnalysis3D() would try to read a non-existing tensor element when requesting the ‘l3’ output (producing an obscure error message).

• dip::GaussFT() could, under some circumstances, try to reforge the output image to an intermediate, extended size. This would throw an exception if the image was protected, but was always a bad thing to do. See issue #170.

• The Fourier Transform, when using the default PocketFFT, used a plan cache that was not thread safe. When calling any function using a Fourier Transform from multiple threads, a race condition could occur. We’ve added a mutex to the function that maintains the cache to avoid this.

• In 3.5.0, the original version of dip::GetImageChainCodes() was deprecated in favor of a new one that takes a std::vector< dip::LabelType > as input, as opposed to a dip::UnsignedArray. This caused code that called the function with an initializer array (dip::GetImageChainCodes( image, { 1 } )) to become ambiguous. A new overload that takes an initializer list as input fixes this ambiguity.

• operator+() for dip::Image (i.e. the function called for +img) created a new data segment and copied the data over when the input was binary. In that case it now creates a new uint8 image that points to the same data segment of the binary image, as intended.

• dip::EdgeObjectsRemove() produced an empty output image when the input and output images were the same object.

Build changes

• The documentation building target was renamed to “doc” (as “apidoc”). This target builds much more documentation than just the DIPlib API documentation (the DIPimage and PyDIP user manuals, build instructions, etc.).

Changes to DIPimage

New functionality

• The function documentation now has links to the online DIPlib documentation. Instead of only naming the DIPlib functions used, the function names are now hyperlinks.

Changed functionality

• The 'FileWriteWarning' setting now also applies to the warning regarding conversion to uint8 when imwrite delegates to MATLAB’s built-in image writing capability. A warning is no longer produced if the image already was uint8.

(See also changes to DIPlib.)

Bug fixes

• The dip_image constructor (and consequently some functions such as newim) again accept some data type aliases that existed in DIPimage 2.x ('bin8', 'int', 'uint', 'float', 'complex'). These are not terribly useful, but there’s no reason not to accept them.

• When imread delegated to MATLAB’s built-in image file reading capability, it failed to tag CMYK images as such.

(See also bugfixes to DIPlib.)

Build and installation changes

• Ported and improved the old DIPimage test suite, which is not (yet?) run automatically with make test because it requires DIPimage to be installed.

Changes to PyDIP

New functionality

• Added the function dip.Doc(), which will open the documentation for the given function or class in the default web browser.

• Automatically copying the brief section of the documentation to each function as the Python doc string. We finally have some documentation in Python!

• Overloaded len() for dip.Measurement. It’s equal to NumberOfObjects() (the number of rows), and thus produces the same value it does when casting the measurement objects to a NumPy array (len(measurement) is the same as len(np.asarray(measurement))).

• Overloaded len() and iter() for dip.Measurement.IteratorFeature and dip.Measurement.IteratorObject, and added keys(), values() and items() class methods. Now both behave more like Python dicts. But note that all three of these methods will output a list with a copy of the values, it’s not a view as with dicts.

Changed functionality

• Converted several tuple output arguments to namedtuple. Note of these changes should affect existing code:

  • The function dip.MaximumAndMinimum() now returns a namedtuple of type MinMaxValues. The two values can now (and preferably) be accessed using the dot notation: mm.maximum instead of mm[1]. See issue #184.

  • The functions dip.MandersColocalizationCoefficients() and dip.CostesColocalizationCoefficients() now return a namedtuple of type ColocalizationCoefficients.

  • The functions dip.ChainCode.BoundingBox() and dip.Polygon.BoundingBox() now return a namedtuple of type BoundingBoxInteger and BoundingBoxFloat respectively, which contain two namedtuples of type VertexInteger or VertexFloat.

  • Other dip.Polygon functions such as dip.Polygon.Centroid() now return a namedtuple of type VertexFloat.

  • dip.ChainCode.start is now a namedtuple of type VertexInteger.

• The types SubpixelLocationResult, RadonCircleParameters, RegressionParameters, GaussianParameters, FeatureInformation, ValueInformation, EllipseParameters, FeretValues, RadiusValues, QuartilesResult, StatisticsValues, CovarianceValues, MomentValues and SpatialOverlapMetrics, all used only as outputs to functions, and all simply emulating a C++ struct, are no longer special types in the diplib.PyDIP_bin namespace, but namedtuples. They new types behave identically, but can additionally be unpacked, for example: _, q1, _, q3, _ = dip.Quartiles(img).

• dip::Measurement::IteratorFeature was bound in Python as dip.Measurement.MeasurementFeature, and dip::Measurement::IteratorObject as dip.Measurement.MeasurementObject. The names of these classes now match the C++ name, to make dip.Doc() useful with these classes. But the names don’t make much sense in Python because these objects don’t work as iterators like they do in C++.

(See also changes to DIPlib.)

Bug fixes

None, but see bugfixes to DIPlib.

Changes to DIPviewer

Bug fixes

• The Java interface used Native.loadLibrary(), which was deprecated. It now uses Native.load() instead.

Changes to DIPjavaio

None.