@TECHREPORT\{IMM1997-01138,
    author       = "R. Larsen and J. D. Hansen and B. K. Ersb{\o}ll and K. Conradsen",
    title        = "Temporal interpolation in Meteosat images",
    year         = "1997",
    keywords     = "Optical flow, temporal interpolation, Meteosat",
    number       = "",
    series       = "",
    institution  = "Dept. of Mathematical Modelling, Technical University of Denmark, {DTU}",
    address      = "Richard Petersens Plads, Building 321, {DK-}2800 Kgs. Lyngby",
    type         = "",
    url          = "http://www2.compute.dtu.dk/pubdb/pubs/1138-full.html",
    abstract     = "The geostationary weather satellite Meteosat supplies us with a visual and an
infrared image of the earth every 30 minutes. However, due to transmission error
s some images
may be missing. European {TV} weather reports are often supported
by such infrared image sequences.
The cloud movements in such animated films are perceived as being jerky due to t
he
low temporal sampling rate in general and missing images in particular.
In order to perform a satisfactory temporal interpolation
we estimate and use the optical flow corresponding to every image in the sequenc
e.
The estimation of the optical flow is based on
images sequences where the clouds are segmented from the land/water that might a
lso be
visible in the images. Because the pixel values measured correspond directly to
temperature
and because clouds (normally) are colder than land/water we use an estimated lan
d temperature
map to perform a threshold between clouds and land/water. The temperature maps
are estimated using  observations from the image sequence itself
at cloud free pixels and ground temperature measurements from a series of meteor
ological
observation stations in Europe.  The temporal interpolation of the images is bas
ed on a path
of each pixel determined by the estimated optical flow.
The performance of the algorithm is illustrated by the interpolation of
a sequence of Meteosat infrared images."
}