Changes DIPlib 3.6.1

Changes to DIPlib

New functionality

• Added a "antisym reflect" (dip::BoundaryCondition::ANTISYMMETRIC_REFLECT) boundary condition. It is similar to "asym mirror", but ensures the derivative is constant at the image boundary. See Issue #214.

• Added dip::Image::Pixel::ExpandTensor().

• TIFF files can now be read and written using ZSTD compression, if configured during build, using the zstd library on the system (i.e. is not part of the DIPlib source distribution, and is not enabled by default).

• Added dip::Dice, which does the opposite of dip::Tile.

Changed functionality

• The "label" color map produced by dip::ColorMapLut() and used by dip::ApplyColorMap() now has 60 unique colors, up from 19. Only the first 6 colors are the same as before. This also affects the labeled image display in DIPviewer, and in the Python and MATLAB packages.

• The stream insertion operator (<<) for dip::Image now also notes if the image is protected, if the data is external, if it has an ExternalInterface attached, and if the data is contiguous.

• dip::Image::HasContiguousData() and dip::Image::HasNormalStrides() no longer check if the image is forged or not. These functions only check the sizes and strides of the image, which can be set before the image is forged.

• dip::ExternalInterface has a new virtual member function Name() that derived classes can overload to give themselves a name.

• dip::Canny() has a new boundaryCondition argument, that is passed through when computing the gradient.

• dip::Tile() has relaxed requirements for input image sizes. It is now possible to tile images of different sizes as long as the images on the same row have the same height, and images on the same column have the same width.

• dip::Framework::VariadicScanLineFilter is more efficient when the input images are scalar. And because it applies the same operation to every scalar, its function is identical when treating the tensor dimension as a spatial dimension, making the images scalar. This is accomplished by using the dip::Framework::ScanOption::TensorAsSpatialDim option in dip::Framework::Scan()/dip::Framework::ScanMonadic()/dip::Framework::ScanDyadic(). With these changes, element-wise arithmetic, bitwise and comparison operators should be a bit faster now for non-scalar images. To enforce proper use, we have:
  • Added a fourth template parameter Scalar to dip::Framework::VariadicScanLineFilter, which defaults to false to avoid changing the behavior of existing code. But instantiating this template with Scalar == false will generate a deprecation warning. When set to true input images must be scalar.
  • Added dip::Framework::NewScalarMonadicScanLineFilter(), dip::Framework::NewScalarDyadicScanLineFilter(), dip::Framework::NewScalarTriadicScanLineFilter() and dip::Framework::NewScalarTetradicScanLineFilter(), replacing dip::Framework::NewMonadicScanLineFilter(), dip::Framework::NewDyadicScanLineFilter(), dip::Framework::NewTriadicScanLineFilter() and dip::Framework::NewTetradicScanLineFilter(), which are now deprecated.
• dip::FindHoughCircles() and its sub-function dip::PointDistanceDistribution() have a new argument options, through which one can pass the option "normalized" to avoid a bias towards larger circles.

Bug fixes

• dip::Image::Mask used multiplication for masking, which doesn’t work to mask out NaN or Infinity values.

• dip::NormalizedConvolution and dip::NormalizedDifferentialConvolution didn’t handle NaN or infinity input values correctly if they were masked out. This now works correctly for the case of binary mask image. For non-binary masks, the user is now warned by the documentation to remove such values from the image before the convolution. See issue #211.

• The low-level B-spline interpolation function had a bug that could sometimes cause the program to crash. See issue #212.

• dip::Rotation() has a bug if the "periodic" boundary condition is given. It now throws an exception if this boundary condition is used, rather than potentially crashing the program.

• The numeric function dip::modulo() didn’t handle all cases correctly.

• operator%() and the unary operator-() for dip::Image::Pixel inputs threw a “data type not supported” exception for single-precision float pixels.

• dip::ReadPixelWithBoundaryCondition() didn’t implement the boundary conditions exactly the same way as all other functions in this library, it now produces the exact same values for the modes that it supports.

• Projection functions that compute sums (dip::Sum(), dip::Mean(), dip::SumAbs(), dip::MeanAbs(), dip::SumSquare(), dip::MeanSquare(), dip::SumSquareModulus(), dip::MeanSquareModulus()) or products (dip::Product(), dip::GeometricMean()) now do the computation using double precision internally, before casting to the output data type (typically single-precision float). This should lead to better precision for large images.

• dip::ImageSliceIterator::SetCoordinate(), Set() and Reset() didn’t update the internal data pointer, meaning that the slice returned by the iterator after calling any of these functions didn’t match the coordinate, and could point to out-of-bounds data.

• dip::RadonTransformCircles() now sets the connectivity parameter to WatershedMaxima() to 0, instead of 1. This should significanlty improve the quality of the local maxima detected. See discussion #227.

• dip::FindHoughCircles() did not pass the range parameter to the dip::PointDistanceDistribution() sub-function. See issue #228.

Updated dependencies

• Updated LibTIFF to version 4.7.1.

Build changes

• Suppress some warnings when building libjpeg with GCC

Changes to DIPimage

New functionality

• Added the overload dip_image/nnz.

Changed functionality

• Added the 'exclude_out_of_bounds_values' option to diphist.

• In the dipzoom mode of the dipshow window, double-clicking no longer zooms to 100%. Instead, the second click in the double-click repeats the action of the first click. Use shift-click to zoom to 100%.

• The diplink command no longer silently ignores erroneous window handles in its input arguments. It now also accepts string inputs.

• dip_image/tensorfun with the options 'isempty', 'islogical', 'isreal', 'ndims' or 'prodofsize' now returns an array with one element per tensor element, instead of an array matching the tensor sizes. This makes these options logically compatible with the other options ('max', 'mean', etc.).

(See also changes to DIPlib.)

Bug fixes

• When linking one or more windows, their zoom level is now adjusted to the controlling window.

• The dipzoom drag interactions in dipshow windows now also propagate to linked windows.

• All functions should now work correctly when given modern-style strings ("") as input. Previously, all functions implemented in M-code assumed string inputs were the old-style strings (actually char arrays, ''), and would give (sometimes strange) errors with modern strings. Functions implemented in MEX-files always accepted both types of string.

(See also bugfixes to DIPlib.)

Changes to PyDIP

New functionality

• Added bindings for three functions that manipulate the image’s data segment: dip.Image.ForceNormalStrides(), dip.Image.ForceContiguousData() and dip.Image.Separate().

• Added dip.Dice().

• Added bindings for a few specific iterators: dip.BresenhamLineIterator, dip.ImageSliceIterator and dip.ImageTensorIterator() (which is a function that returns a dip.ImageSliceIterator object).

Changed functionality

• Regular indexing (such as img[10:40:2, :]), which creates a new image that shares data with the original image, now has the output image protected. This allows the user to write into the sub-image with confidence. See issue #204.

• dip.PhysicalQuantity has a new alias dip.PQ, making it easier to write code that works with physical quantities and pixel sizes.

• For DIPlib functions that take a dip::Range as input, the Python bindings take a slice object. Now they can also take a tuple, defined in the same way as the slice() constructor except that None cannot be given. For example, instead of slice(5, 10) you can now write (5, 10). Note that (5,) is the same as (0, 5), because that is how slice() is defined. slice(None, None) is the same as slice(0, -1) which now can be written as (0, -1), or simply ().

(See also changes to DIPlib.)

Bug fixes

• dip.Rotation2D() and dip.Rotation3D() use the "add zeros" boundary condition by default, like the C++ functions do.

(See also bugfixes to DIPlib.)

Changes to DIPviewer

New functionality

• Added the dip::viewer::ViewingOptions::Mapping::Modulo scaling option (shown as “MOD” in the UI). It computes the scaled pixel value modulo 256, but skipping 0 (i.e. values 0-255 are unchanged, 256 maps to 1, etc.). This is useful with the “labels” color map: if there are more than 255 objects, the color map will repeat correctly. It is also useful to display distance mappings: turn on this mode, then scale the intensities by dragging down from the top of the histogram on the right-hand side of the window.

• Added a series of getter and setter functions to dip::viewer::SliceViewer to simplify modifying the image display programatically.

Changed functionality

• Programmatic linking now follows the same rules as manual linking with respect to dimensionality.

Bug fixes

• When choosing the labeled color map, a rounding error in the mapping from grayvalues to color map indices became apparent, noticeable in a few objects getting the same color as the object with the previous ID.

• Offsets are no longer overwritten when updating the image being displayed.

• When updating the image, the mapping range is only recalculated automatically if it has not been manually adjusted.

Changes to DIPjavaio

None.

Changes to the dipview tool

• The dipview and dipviewjava command-line tools will now first attempt to read a file as if it were a multi-page TIFF file where the pages compose a 3D image. If this fails, they proceeds as they did previously, reading only the first page.