cnvlv(3G)cnvlv(3G)NAME
cnvlv - modify the operation of lrectwrite and rectcopy to convolve pixel
data
FORTRAN 77 SPECIFICATION
subroutine cnvlv()PARAMETERS
convop the convolution mode: general, separable or off
border a token specifying the source for the input data border
xksize the size of the kernel width in pixels in the x dimension
yksize the size of the kernel width in pixels in the y dimension
kernel an array of kernel elements
bias a value usually between [-1.0,1.0] to be added to the result of
the convolution.
DESCRIPTION
RealityEngine supports a sequence of of floating point image processing
operations. The operations are performed in the following order: integer
to floating point conversion, scale, bias, convolution, floating point to
integer conversion, and finally lookup table mapping.
The pixels are converted from integers in the range specified by
pixmode(PM_INPUT_TYPE), to the floating point range [0.,1.] if the input
type was unsigned or (-1.,1] if the input type was signed. The pixels are
multiplied by a color scale and then a color bias is added. The scale
and bias values are specified using the pixeltransfer() function. The
resulting pixels are convolved, if convolution has been enabled by the
convolve() function. The resulting pixels are converted back to integers
where the range [0.0,1.0] maps to [0,4095] if pixeltransfer(PT_MAP_COLOR)
is disabled or [0,lookup_table_size), if table lookup mapping is enabled.
The result is written to the framebuffer.
cnvlv specifies the parameters defining the convolution to be performed
by lrectwrite or rectcopy.
convop may be CV_GENERAL, CV_SEPARABLE, or CV_OFF, which is the default.
CV_GENERAL specifies that a full 3x3, 5x5, or 7x7 convolution should be
performed. CV_SEPARABLE specifies that the convolution should be
performed as two one dimensional convolutions: first a horizontal
convolution, followed by a vertical convolution. Separable convolutions
can be considerably faster than general convolutions as the number of
arithmetic operations is reduced.
border may be:
Page 1
cnvlv(3G)cnvlv(3G)
CV_REDUCE to specify that the input border data will be supplied by
the programmer so that the output data will be less than that input.
If the input data is MxN and the kernel size KxK, the output data
will be (M - K + 1)x(N - K + 1).
xksize and yksize may be 3, 5 or 7 and must be equal to each other.
kernel is a xksize x yksize array of floats in row major order if convop
is CV_GENERAL. If the convop is CV_SEPARABLE, the kernel should be an
array of xksize + yksize floats; first the coefficients for the
horizontal pass, followed by the coefficients for the vertical pass. The
coefficients should be chosen so the output range will be between
[0.,1.].
If convolution is enabled, all of the features of pixmode will be ignored
except for PM_INPUT_FORMAT, PM_OUTPUT_FORMAT, PM_INPUT_TYPE,
PM_OUTPUT_TYPE, PM_OFFSET, and PM_STRIDE.
If the pixel input format has multiple components, they will be convolved
independently.
SEE ALSONOTES
cnvlv is supported only on RealityEngine systems with release 5.0.1 or
greater.
Page 2
