Simple Conversion between RealTypes

[maarzt]

When I need to convert between two RealTypes a usally us a lambda expression like this:

Converter<RealType, RealType> converter = (i, o) -> o.setReal(i.getRealDouble());

But this converter is not optimal in term of performance (To convert ByteType -> IntType you would not need do cast to double), a precision (UnsignedLongType can be poorly represented as double).

This PR adds three methods to fix this:

• Getting converter between two RealTypes, which is optimal in terms of performance and precision, is as easy as:

Converter<UnsignedByteType, LongType> converter = 
     RealTypeConverter.getConverter(new UnsignedByteType(), new LongType());

• Converting the type of a RandomAccessibleInterval:

RandomAccessibleInterval<LongType> converter = 
     RealTypeConverter.convert(image, new LongType());

• Copy pixels between to RandomAccessibleIntervals:

CopyPixels.fromTo(imageA, imageB);

[maarzt]

I'm not sure about the names of the static methods, or to which class the should belong: RealConverters.getConverter, RealConverters.convert and especially CopyPixels.fromTo.

And I think about adding more functionality: Converters that bound the value to minimum and maximum. And converters allow to scale the values.

[tpietzsch]

@StephanPreibisch @axtimwalde I just noticed (because this PR touches it), that Util.round() behaviour differs from Math.round() for negative numbers. E.g., Math.round() rounds -0.5 to 0, while Util.round() rounds -0.5 to -1.

Do you remember what the reason was to do it like this?
And if there was no specific reason, should we change it to just do Math.round()?

[tpietzsch]

I'm not sure about the names of the static methods, or to which class the should belong: RealConverters.getConverter, RealConverters.convert and especially CopyPixels.fromTo.

I like RealTypeConverters.getConverter and RealTypeConverters.convert.

CopyPixels.fromTo is not good.
The name hides the fact that it converts between types.
Also, placed in a unrelated CopyPixels class, it is weird that it is limited to RealType.
I would like RealTypeConverters.copyConverted(...) better.

A general copy-pixels method would be nice though.
(There is ImgUtil.copy(...) but it's very specific. Maybe could be deprecated when something more general is available).
I would change signature to the source being only RandomAccessible (not Interval), so that it can be used to fill pixels of an output interval from a potentially unbounded source.

Then I could imagine to have a CopyPixels class with more methods where the converted copy would fit in. E.g.

public final class CopyPixels
{
	public static < T extends Type< T > > void copy(
			RandomAccessible< T > source,
			RandomAccessibleInterval< T > destination )
	{...}

	public static void copyReal(
			RandomAccessible< ? extends RealType< ? > > source,
			RandomAccessibleInterval< ? extends RealType< ? > > destination )
	{...}
}

And I think about adding more functionality: Converters that bound the value to minimum and maximum. And converters allow to scale the values.

I think this is a good idea, but could probably be done later in a separate PR.

[tpietzsch]

With these classes being public could you add javadoc, please.
Can be very minimal, e.g.,

/**
 * Converts via {@link RealType#getRealDouble}.
 */

is better than nothing.

[maarzt]

done

[tpietzsch]

A general copy-pixels method would be nice though.
Also related: #237

[maarzt]

@tpietzsch I did the changes you requested:

• I moved the copy method to RealTypeConverters.copyFromTo(RandomAccessible source, RandomAccesibleInterval destination).
• First argument of the copy method is now RandomAccessible.
• I additionally moved all the private methods in RealTypeConverters to RealTypeConverterInternals. This is to better separate the public methods, from the many private methods. It also provides more space in the RealTypeConverters class, to add more functionality like scaling and bounded converters.

[tpietzsch]

@maarzt I merged this accidentaly, now force-pushed to master to undo that.
So the state of this PR doesn't reflect reality unfortunately

[tpietzsch]

I'll review and then merge locally...

[tpietzsch]

ok, merged now...