HELP FLOAT_ARRAYPROCS                           David Young
                                                November 1992
                                                revised Oct 2001

LIB *FLOAT_ARRAYPROCS provides a suite of procedures for carrying out
some common operations (and one or two less common ones) on arrays. The
arrays must be packed arrays of single precision floating point values,
(except where stated) as created by *NEWSFLOATARRAY or -array_of_real-
(from*FORTRAN_DEC).

Below, ARR as an argument means that an array is required. ARR_FALSE
means that either an array is required, or <false> or another object may
be supplied instead. When the argument is called ARR_FALSE any array
given will have its contents overwritten, and then be returned as the
result of the procedure; if <false> is given, a new array will be
created and filled; if some other object is given, it will be used as a
tag in a call to oldsfloatarray (see *NEWSFLOATARRAY) which will cause a
recyclable array to be returned.

All the array arguments in one call of one of these routines must have
the same bounds, and arrays must not be offset in their arrayvectors.

         CONTENTS - (Use <ENTER> g to access required sections)

 -- Operations on individual pixels
 -- Neighbourhood operations

-- Operations on individual pixels ------------------------------------

For these routines, ARR_FALSE can be the same as any other array
argument, which will be overwritten.

float_arraycopy(ARR, ARR_FALSE) -> ARR_OUT
    Copy the values in the array. (See *ARRAYSAMPLE for how to copy
    values from a region of an array.)

byte2float(ARR, ARR_FALSE) -> ARR_OUT
    Converts a byte array into a float array containing the same
    numerical values.

float2byte(MIN1, MAX1, ARR, MIN2, MAX2, ARR_FALSE) -> ARR_OUT
    Converts a float array ARR to a byte array. Values are scaled
    linearly, with MIN1 in ARR mapping to MIN2 in the output, and
    likewise MAX1 and MAX2.

float_arrayabs(ARR, ARR_FALSE) -> ARR_OUT
    Takes the absolute value of each element.

float_arraysqr(ARR, ARR_FALSE) -> ARR_OUT
    Squares each element.

float_arraysqrt(ARR, ARR_FALSE) -> ARR_OUT
    Takes the square root of each element.

float_arraylogistic(ARR, ARR_FALSE) -> ARR_OUT
    Applies the logistic function 1/(1+exp(-x)) to each element.

float_arraysum(ARR1, ARR2, ARR_FALSE) -> ARR_OUT
    Adds ARR1 and ARR2 element by element.

float_arraydiff(ARR1, ARR2, ARR_FALSE) -> ARR_OUT
    Subtracts each element of ARR2 from the corresponding element of
    ARR1.

float_arraymult(ARR1, ARR2, ARR_FALSE) -> ARR_OUT
    Multiplies ARR1 and ARR2 element by element.

float_complexmult(ARR1R, ARR1I, ARR2R, ARR2I, ARR_FALSE_R, ARR_FALSE_I)
        -> (ARR_OUT_R, ARR_OUT_I)
    Multiplies the complex pair (ARR1R, ARR1I) and the complex pair
    (ARR2R, ARR2I) element by element.

float_arraydiv(ARR1, ARR2, ARR_FALSE) -> ARR_OUT
    Divides ARR1 by ARR2 element by element. No element of ARR2 may be
    zero.

float_arraycomb(K1, ARR1, K2, ARR2, ARR_FALSE) -> ARR_OUT
    Linear array combination: multiplies each element of ARR1 by K1,
    each element of ARR2 by K2, adds these products and stores the
    result in ARROUT.

float_arrayhypot(ARR1, ARR2, ARR_FALSE) -> ARR_OUT
    Stores sqrt(ARR1(I)**2 + ARR2(I)**2) in ARR_OUT(I) (see MAN *HYPOT).
    (Here I can stand for multiple indices in the POP-11 array.)

float_arrayatan2(ARR1, ARR2, ARR_FALSE) -> ARR_OUT
    Stores atan2(ARR1(I), ARR2(I)) in ARR_OUT(I) (see MAN *ATAN2).
    Note that the arguments are the opposite way round to the POP-11
    arctan2 function.

float_threshold(V1, THRESH, V2, ARR, ARR_FALSE) -> ARR_OUT
    Each element that is less than THRESH is set to V1, others are set
    to V2.  If V1 is <false> then elements less than THRESH are set to
    the original data value; if V2 is <false> then elements greater than
    or equal to THRESH are set to the original data value.

float_threshold2(V1, THRESH1, V2, THRESH2, V3, ARR1, ARR_FALSE) -> ARR_OUT
    Each element that is less than or equal to THRESH1 is set to V1;
    each element that is greater than or equal to THRESH2 is set to V3;
    all the others are set to V2. If any of V1, V2 and V3 are <false>,
    the original data is retained wherever they would have been used.

float_arraymean(ARR) -> NUMBER
    Returns the mean of the elements of ARR.

float_mean_sd(ARR) -> (MEAN, SD)
    Returns the meand and SD of the values of the elements of ARR.

float_arraymean_mask(ARR, MASK, WHEREZERO) -> NUMBER
    MASK is an array the same size as ARR; WHEREZERO is a boolean.
    If WHEREZERO is <true>, returns the mean of those elements of ARR
    for which there is a zero in the corresponding element of MASK.
    If WHEREZERO is <false>, returns the means of those elements of ARR
    for which the corresponding element of MASK is non-zero.

float_arrayhist(ARR, MN, MX, NBINS) -> HIST
    Returns the histogram of values in ARR. A value V in ARR is counted
    if MN <= V < MX. There are NBINS equally-spaced bins; HIST is a
    vector of length NBINS containing the counts. On exit, HIST(I)
    contains the number of values V in ARR for which (I-1)*BINSIZE + MN
    <= V < I*BINSIZE + MN where BINSIZE = (MX-MN)/NBINS.
        For example, if ARR happens to contain integral values from 0 to
    255, then float_arrayhist(ARR, 0, 256, 256) will return a result
    such that HIST(I) contains the number of values in ARR equal to I-1.
        Instead of NBINS, the last argument may be an *INTVEC or an
    array whose arrayvector is an intvec. In this case, the structure
    will receive the results and be returned; its length will be the
    number of bins.

float_setconst(NUMBER, ARR)
    Sets each element of ARR to NUMBER.

float_multconst(NUMBER, ARR, ARR_FALSE) -> ARR_OUT
    Multiplies each element of ARR by NUMBER.

float_addconst(NUMBER, ARR, ARR_FALSE) -> ARR_OUT
    Adds NUMBER to each element of ARR.

float_multconst_mask(NUMBER, ARR, MASK, WHEREZERO, ARR_FALSE) -> ARR_OUT;
    Multiplies by NUMBER every element of ARR that has a corresponding
    zero (non-zero) value in MASK if WHEREZERO is <true> (<false>).
    Other elements are simply copied into ARR_OUT.

float_addconst_mask(NUMBER, ARR, MASK, WHEREZERO, ARR_FALSE) -> ARR_OUT;
    Adds NUMBER to every element of ARR that has a corresponding zero
    (non-zero) value in MASK if WHEREZERO is <true> (<false>). Other
    elements are simply copied into ARR_OUT.

float_arraywtdav_mask(ALPHA1, ALPHA2, ARR1, ARR2, MASK, ARR_FALSE)
                    -> ARR_OUT
    If A1 is an element of ARR1 and A2 is an element of ARR2, then the
    corresponding element of ARR_OUT is set to ALPHA*A1 + (1-ALPHA)*A2,
    ALPHA is ALPHA1 for any element where the corresponding value of
    MASK is zero, and ALPHA is ALPHA2 otherwise. This is useful to
    averaging images selectively.

-- Neighbourhood operations -------------------------------------------

The output array, if supplied, must not the same as any other array
argument.

float_dilate_nonzero(NUMBER, ARR1, ARR_FALSE) -> ARR_OUT
    Dilates non-zero areas using a 3x3 structuring element. Dilated
    regions are set to NUMBER.

float_erode_nonzero(NUMBER, ARR1, ARR_FALSE) -> ARR_OUT
    Erodes non-zero areas using a 3x3 structuring element. Non-zero
    regions are set to NUMBER.

--- $popvision/help/float_arrayprocs
--- Copyright University of Sussex 2001. All rights reserved.
