Class to hold values from image expression parser
Intended use:
Public interface
Prerequisite
Etymology
ImageExprParse is the class used to parse an image expression command.
Synopsis
ImageExprParse is used by the parser of image expression statements. The parser is written in Bison and Flex in files ImageExprGram.y and.l. The statements in there use the routines in this file to act upon a reduced rule.
The main function (and the only function to be used by a user) is the static function ImageExprParse::command which parses an expression command. It returns a LatticeExprNode object containing the expression represented as a tree. The object can be used as a Lattice(Expr)<T> in other operations.
The syntax of the command is similar to that of expressions in C++. E.g.
LatticeExprNode min(const LatticeExprNode &left, const LatticeExprNode &right)
LatticeExprNode sin(const LatticeExprNode &expr)
Numerical 1-argument functions.
The following items can be used in an expression:
-
Binary operators +, -, *, /, % (modulo), and ^ (power).
-
Unary operators + and -.
-
Comparison operators ==, >, >=, <, <=, and !=.
-
Logical operators &&, ||, and !.
-
Constant single and double precision values.
No exponent or exponent "e" results in single precision (Float), while "d" results in double precision (Double).
-
The imaginary part of a complex value can be given by the suffix "i". A full complex number can be given by addition. E.g. "3+4i". The complex is single (Complex) or double (DComplex) precision depending on the constituting parts.
-
The special constants pi and e can be given as a double precision value by means of the functions pi() and e().
-
Boolean constants T and F can be given.
-
A lot of functions are available. They are the same as the ones supported by class LatticeExprNode.
-
Explicit conversion functions float, double, complex and dcomplex are available. Conversions are automatically done where needed, but for performance reasons it may sometimes be better to do explicit conversions. See also below in the first example.
-
An image can to be given using its file name. The file name can contain environment variables and user home directories using the standard UNIX syntax $ENVVAR and ~username. There are 3 ways to specify a file name:
-
When the name contains no other special characters than $, ~, and. it can be given as such.
-
Backslashes can be used to escape individual special characters.
-
The full name can be enclosed in quotes (single or double) to escape the entire name. Adjacent quoted parts are combined to one name, which can be used to use quotes in the file name.
Note that escaping has to be used too for the file name T or F (otherwise it is the boolean constant). E.g. image.data
"~noordam/data/image.data"
"~/image.data"
"$HOME/image.data"
$HOME\/image.data
"ab'c"'d"e' results in ab'cd"e
-
An image can also be given by means of the
$n notation, where n is the sequence number in the tempLattices argument given to the command function. Note that the sequence numbers start counting at 1 (to be compliant with glish indexing).
It can, for instance, be used to use non-persistent lattices in an expression.
When the expression is parsed, it is checked if the images and lattices involved have conforming shapes and coordinates. Note, however, that some functions (e.g. mean) reduce an image to a scalar. Such an image can have a different shape and coordinates.
The data types of the images and constants involved can be different. The data type of a subexpression is the common data type (e.g. Float and Double result in Double; Complex and Double result in DComplex). Automatic implicit conversions are done where needed. However, for performance reasons it may sometimes be better to convert explicitly. See below in the first example.
The expression evaluator (which is not part of the parser) evaluates the expression in chunks to avoid having to keep large temporary results. A scalar subexpression is evaluated only once to avoid unnecessary (possibly expensive) calculations.
Some examples:
img1 + min(float(pi()), mean(img2)) - Suppose img1 and img2 are images with single precision data. They do not need to have conforming shapes and coordinates, because only the mean of img2 is used.
Note that pi is explicitly converted to single precision, because pi() results in a Double. If that was not done, the expression result would be a Double with the effect that all data of img1 had to be converted to Double.
min(img1, (min(img1)+max(img1))/2) - This example shows that there are 2 min functions. One with a single argument returning the minimum value of that image. The other with 2 arguments returning a lattice containing img1 data clipped at the value of the 2nd argument.
Example
LatticeExpr<Double> expr ("a + sin(b)");
ArrayLattice<Double> arr(expr.shape());
arr.copyData (expr);
Line 1 creates a LatticeExpr object for the given expression. Note that a and b are names of lattice files (e.g. PagedImage).
Line 2 creates an ArrayLattice with the same shape as the expression (which is the shape of lattice a (and b)).
Line 3 copies the result of the expression to the ArrayLattice.
Motivation
It is necessary to be able to give an image expression command in ASCII. This can be used in glish to operate on lattices/images.
Definition at line 205 of file ImageExprParse.h.
