<target_file> is the file name of an image for which detection probability should be computed (e.g. distorted image) and <mask_file> is the file name of an image that is a reference (e.g. original, not distorted). The order in which both images are passed to the vdp is important and may affect the results.
This command can handle various image formats (for the list of recognized formats, execute pfsin --help). It also recognizes low-dynamic-range files (gamma corrected) and adds necessary switches to transform from gamma corrected luma values to luminance (assuming sRGB color space for input images). The maximum luminance of the display is by default set to 80cd/m^2. It can be modified with --multiply-lum switch (see man page of vdpcmp for more information). To learn more about gamma correction, see Charles Poynton's web page on Gamma (http://www.poynton.com/notes/colour_and_gamma/GammaFAQ.html).
Note that HDR VDP compares only luminance for a pair of images so differences in color are not predicted. The D65 white point is used to compute luminance from color images.
A viewing conditions (screen size, resolution, and viewing distance) may be specified using -vc option. The parameter <visual-conditions> is the name of the file with the viewing conditions. Such file should be located in the directory $HOME/.viewing-conditions and should have the following structure:
display-width=0.375 display-height=0.300 display-x-resolution=1280 display-y-resolution=1024 min-distance=0.5 max-distance=0.5
Both distance and display width / height are given in meters. Resolution is given in pixels. If no -vc option is given, the values from the example above are used.
-s option writes summary information to <file.cvs>. Summary information contains the ratios of pixels that contain distortion of the probability greater than 75% (first number) and 95% (second number). The numbers are separated by a comma. Value 0 denotes no pixels, 1 all pixels.
VDP has a large number options, which can significantly affect the results. The default options were tuned for natural images and sine-grating noise. If the predicted probability of detection for the default options is higher / lower than you would expect, you may need to tune some options for your set of images. The primary parameters for tuning are:
See also manual page of vdpcmp for the full list of options.
This command is a frontend that combines several low level commands: pfsin to load hdr image, vdpcmp to compute probability map of visual differences, vdpvis to visualize results, and pfsout to write low dynamic range image.
Compare 'memorial.hdr' and 'memorial_sinnoise.hdr', write the result to 'vdpout.png'.