Findclouds Manual 28
Evaluation
"Algorithm CDOC":
Next section of evaluation is "Algorithm CDOC" with
images "Difference", "Difference HCF", "Thick Clouds",
"Thin Clouds", "Clear Sky Library" and the result
"Evaluation CDOC". For this analysis the checkbox
“Algorithm CDOC” has to be activated.
This evaluation implements the algorithm which is
described by "A method for Cloud Detection and
Opacity Classification based on ground based sky
imagery"
[3]
(capitalisation shows our acronym CDOC).
The algorithm uses images from a clear sky library (a
collection of cloudless images). It determines clouds by
creating the red/blue ratio of the current image and the
best matching clear sky image and evaluating the
difference of them. The simple "Difference" provides
the thick clouds and the "Difference HCF" provides the
thin clouds. For ''Difference HCF'' the mean values of
blue sky pixel in current image and clear sky image will be calculated and the images will be adjusted for
minimum difference of clear sky mean values. That way allows to determine small differences caused by thin
clouds or haze.
The
"Difference"
and
"Difference HCF"
shows the differences of the red/blue ratio of the original image and
the clear sky image. In reality these are grey scale values. They are colourised to make them more
demonstrative. The colour bar between these images shows the range of values and scales the colours. It is
valid for both images (Difference, Difference HCF), so the same colour will show the same value and the
images are directly comparable.
The slider on right-hand side of "Difference HCF" adjusts the level of clear sky pixel. Pixel less than this value
will be estimated as clear sky in HCF calculation.
The
"Clear Sky Library"
shows the image, that will be used to calculate the
differences.
A problem of the CDOC method results out of the fact, that bright parts of the Clear
Sky Image (the area near the sun) can not be evaluated, because there is no
evaluable red/blue ratio left by overexposed pixel.
To handle this problem, there is the possibility to mask the bright parts of the Clear
Sky Image. This reduces the area of evaluation, but ensures, that the remaining
part will be evaluated in a correct way. Otherwise the evaluation will tell too less
cloudiness, because the invalid area normally will be recognised as blue sky.
For adjustment of the "clear sky brightness mask" there is a slider on right-hand
side of the image to adjust the level.
The examples at the right side show the unmasked evaluation (1) resulting in a
cloudiness of 0.96 for a totally clouded image. This difference results out of the Clear
Sky Image. By using the brightness mask (2), the calculated cloudiness results in
1.00. So there is a better result, although a smaller part of the image was evaluated.
Anyway this difference will be reduced substantial by use of underexposed images,
because underexposure reduces the overexposed area.
The
"Thick Clouds"
shows pixel of 'difference ratio' with values larger than "Level
Thick Clouds" as thick cloud pixel. The slider on right hand side of the image allows
to adjust the level of thick clouds and make the result visually match better to original
sky image.
The
"Thin Clouds"
shows pixel of 'difference HCF ratio' with values larger than
"Level Thin Clouds" as thin cloud pixel. The slider on right hand side of the image
allows to adjust level of thin clouds and make the result visually match better to
original sky image.
The
"Evaluation CDOC"
shows the final evaluation of the CDOC algorithm - the
calculated cloudiness values. The cloudiness values are based on counting the
pixels of this image.
The analysis produces the values 'Thick Clouds' and 'Thin Clouds'.
The 'Cloudiness CDOC' is the sum of these two values to make a comparison to
whole Cloudiness BRBG more easy.
1
2
