More precisely, it allows the decoder to try `SOF1` images. There are
still some sub-kind of this kind of JPEG that we don't support. In a
nutshell `SOF1` images allow more Huffman and quantization tables, 12
bits precision and arithmetic encoding. This patch only brings support
for the "more tables" part.
Please note that `SOF2` images are also allowed to have more tables, so
we gave the decoder the ability to handle these in the same time.
Instead of testing all possible code to find the good symbol, we use a
lookup table to directly find the expected symbol. This method is used
by most Huffman decoder (gzip or libjpeg-turbo).
In order to use the correct key when peeking a constant number of bits
from the stream, we generate duplicates in the table. As an example, for
the code 110, all entries with that pattern 110***** will be set to
110's symbol. So, when you read this code plus garbage from following
codes, you still find the correct symbol.
Advantages of encapsulation are really obvious here:
- Put related code together
- Prevent external functions to modify the object
- Abstract the implementation
No functional changes intended.
We used to limit this number to two and name them with their usual
usage. The specification is however broader, and you can find files that
use more tables as in the following link:
20050519/tests/A_2_1-BF-01.htm
No need to store them in `i32`, the JPEG norm specifies that they are
not bigger than 16 bits for extended JPEGs. So, in this patch, we
replace `i32` with `i16`. It almost divides memory usage by two :^)
This patch adds support for properly read images with four components,
basically CMYK or YCCK. However, we still lack color spaces
transformations for this type of image. So, it just postpones failure.
As mentioned in F.2.1.5 - Inverse DCT (IDCT), the decoder needs to
perform a level shift by adding 128. This used to be done in
`ycbcr_to_rgb` after the conversion. Now, we do it in `inverse_dct` in
order to ensure that the task is done unconditionally.
Consequences of this are that we are no longer required to explicitly
do it for RGB images and also, the `ycbcr_to_rgb` function is exactly
like the specification.
When reading the stream, interpreted as a normal value 0xF0 means skip
15 values and assign the 16th to 0. On the other hand, the marker ZRL
- which has the value 0xF0, means skip 16 values. For baseline JPEGs,
ZRL doesn't need to be interpreted differently as writing the 16th value
has no consequence. This is no longer the case with refining scans.
That's why this patch implement correctly ZRL.
We used to skip over zero coefficient by modifying the loop counter. It
is unfortunately impossible to perform this with SOF2 images as only
coefficients with a zero-history should be skipped.
This induces no behavior change for the user of the function.
This commit is nonsense for anything else than SOF2 images with spectral
approximation. For this particular case, skips like EOB or ZRL only
apply to coefficients with a zero-history. This commit prepares the code
to handle this behavior.
This function has probably been added when we weren't as good with error
propagations as we are now. We can safely remove it and let future
calls to `read` fail if the file is corrupted.
This can be tested with the following bytes (already used in 9191829a):
ffd8ffc000000800080ef701101200ffda00030100
Introduced in 2c98eff, support for non-interleaved scans was not working
for frames with a number of MCU per line or column that is odd. Indeed,
the decoder assumed that they have scans that include a fabricated MCU
like scans with multiple components.
This patch makes the decoder handle images with a number of MCU per line
or column that is odd. To do so, as in the current decoder state we do
not know if components are interleaved at allocation time, we skip over
falsely-created macroblocks when filling them. As stated in 2c98eff,
this is probably not a good solution and a whole refactor will be
welcome.
It also comes with a test that open a square image with a side of 600px,
meaning 75 MCUs.
