This patch applies color correction matrix (CCM) in debayering if the
CCM is specified. Not using CCM must still be supported for performance
reasons.
The CCM is applied as follows:
[r1 g1 b1] [r]
[r2 g2 b2] * [g]
[r3 g3 b3] [b]
The CCM matrix (the left side of the multiplication) is constant during
single frame processing, while the input pixel (the right side) changes.
Because each of the color channels is only 8-bit in software ISP, we can
make 9 lookup tables with 256 input values for multiplications of each
of the r_i, g_i, b_i values. This way we don't have to multiply each
pixel, we can use table lookups and additions instead. Gamma (which is
non-linear and thus cannot be a part of the 9 lookup tables values) is
applied on the final values rounded to integers using another lookup
table.
Because the changing part is the pixel value with three color elements,
only three dynamic table lookups are needed. We use three lookup tables
to represent the multiplied matrix values, each of the tables
corresponding to the given matrix column and pixel color.
We use int16_t to store the precomputed multiplications. This seems to
be noticeably (>10%) faster than `float' for the price of slightly less
accuracy and it covers the range of values that sane CCMs produce. The
selection and structure of data is performance critical, for example
using bytes would add significant (>10%) speedup but would be too short
to cover the value range.
The color lookup tables can be represented either as unions,
accommodating tables for both the CCM and non-CCM cases, or as separate
tables for each of the cases, leaving the tables for the other case
unused. The latter is selected as a matter of preference.
The tables are copied (as before), which is not elegant but also not a
big problem. There are patches posted that use shared buffers for
parameters passing in software ISP (see software ISP TODO #5) and they
can be adjusted for the new parameter format.
Color gains from white balance are supposed not to be a part of the
specified CCM. They are applied on it using matrix multiplication,
which is simple and in correspondence with future additions in the form
of matrix multiplication, like saturation adjustment.
With this patch, the reported per-frame slowdown when applying CCM is
about 45% on Debix Model A and about 75% on TI AM69 SK.
Using std::clamp in debayering adds some performance penalty (a few
percent). The clamping is necessary to eliminate out of range values
possibly produced by the CCM. If it could be avoided by adjusting the
precomputed tables some way then performance could be improved a bit.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
Constructing the color mapping tables is related to stats rather than
debayering, where they are applied. Let's move the corresponding code
to stats processing.
The same applies to the auxiliary gamma table. As the gamma value is
currently fixed and used in a single place, with the temporary exception
mentioned below, there is no need to share it anywhere anymore.
It's necessary to initialize SoftwareIsp::debayerParams_ to default
values. These initial values are used for the first two frames, before
they are changed based on determined stats. To avoid sharing the gamma
value constant in artificial ways, we use 0.5 directly in the
initialization. This all is not a particularly elegant thing to do,
such a code belongs conceptually to the similar code in stats
processing, but doing better is left for larger refactoring.
This is a preliminary step towards building this functionality on top of
libipa/algorithm.h, which should follow.
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Reviewed-by: Andrei Konovalov <andrey.konovalov.ynk@gmail.com>
Reviewed-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Source files in libcamera start by a comment block header, which
includes the file name and a one-line description of the file contents.
While the latter is useful to get a quick overview of the file contents
at a glance, the former is mostly a source of inconvenience. The name in
the comments can easily get out of sync with the file name when files
are renamed, and copy & paste during development have often lead to
incorrect names being used to start with.
Readers of the source code are expected to know which file they're
looking it. Drop the file name from the header comment block.
The change was generated with the following script:
----------------------------------------
dirs="include/libcamera src test utils"
declare -rA patterns=(
['c']=' \* '
['cpp']=' \* '
['h']=' \* '
['py']='# '
['sh']='# '
)
for ext in ${!patterns[@]} ; do
files=$(for dir in $dirs ; do find $dir -name "*.${ext}" ; done)
pattern=${patterns[${ext}]}
for file in $files ; do
name=$(basename ${file})
sed -i "s/^\(${pattern}\)${name} - /\1/" "$file"
done
done
----------------------------------------
This misses several files that are out of sync with the comment block
header. Those will be addressed separately and manually.
Signed-off-by: Laurent Pinchart <laurent.pinchart@ideasonboard.com>
Reviewed-by: Daniel Scally <dan.scally@ideasonboard.com>
Black may not be represented as 0 pixel value for given hardware, it may
be higher. If this is not compensated then various problems may occur
such as low contrast or suboptimal exposure.
The black pixel value can be either retrieved from a tuning file for the
given hardware, or automatically on the fly. The former is the right
and correct method, while the latter can be used when a tuning file is
not available for the given hardware. Since there is currently no
support for tuning files in software ISP, the automatic, hardware
independent way, is always used. Support for tuning files should be
added in future but it will require more work than this patch.
The patch looks at the image histogram and assumes that black starts
when pixel values start occurring on the left. A certain amount of the
darkest pixels is ignored; it doesn't matter whether they represent
various kinds of noise or are real, they are better to omit in any case
to make the image looking better. It also doesn't matter whether the
darkest pixels occur around the supposed black level or are spread
between 0 and the black level, the difference is not important.
An arbitrary threshold of 2% darkest pixels is applied; there is no
magic about that value.
The patch assumes that the black values for different colors are the
same and doesn't attempt any other non-primitive enhancements. It
cannot completely replace tuning files and simplicity, while providing
visible benefit, is its goal. Anything more sophisticated is left for
future patches.
A possible cheap enhancement, if needed, could be setting exposure +
gain to minimum values temporarily, before setting the black level. In
theory, the black level should be fixed but it may not be reached in all
images. For this reason, the patch updates black level only if the
observed value is lower than the current one; it should be never
increased.
The purpose of the patch is to compensate for hardware properties.
General image contrast enhancements are out of scope of this patch.
Stats are still gathered as an uncorrected histogram, to avoid any
confusion and to represent the raw image data. Exposure must be
determined after the black level correction -- it has no influence on
the sub-black area and must be correct after applying the black level
correction. The granularity of the histogram is increased from 16 to 64
to provide a better precision (there is no theory behind either of those
numbers).
Reviewed-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Signed-off-by: Hans de Goede <hdegoede@redhat.com>
Signed-off-by: Milan Zamazal <mzamazal@redhat.com>
Signed-off-by: Kieran Bingham <kieran.bingham@ideasonboard.com>
