mindtct appears to completely ignore the pixels-per-mm input parameter
(ippmm). When processing AES4000 images, the binarized image is
completely mangled and a lot of ridge information is lost.
Resizing the AES4000's small images results in a huge imaging performance
gain.
We use imagemagick for the resizing, as it's resizing code resamples the
image too (smoothing it out), which further improves performance.
The windows driver takes one sample at the previous register settings
and then changes some registers before resamping (and getting much better
images). The exact changes that it makes seem to vary, perhaps based on
the histogram.
Anyway, this is an approximation of the settings used for the 2nd sample
which should help matching results.
My UareU4000 device isn't working properly -- only seeing an all-white
image and no finger-on detection, however I think this code should work
for working devices.
When we include the await_finger_off procedure (i.e. set mode 12), the
current power-up code is unreliable - no 56aa interrupt arrives sometimes.
Comparing the F0 data and hwstat at this point between a working and a
non-working run yields no differences.
Work around this issue by retrying power-on 3 times.
While testing the new code on my MS keyboard, I found some strange
conditions where occasionally, the device will stop providing images
headers until it is replugged.
Detect and work around this when it happens.
This should improve driver stability somewhat. Powerup problems seem to
happen when we receive 3 unrelated interrupts while waiting for the
power-on interrupt, so let's assume that the device can only buffer 3
interrupts, and when the buffer is full, it discards new ones.
When we detect a possible buffer overflow, ask the caller to retry the
operation.
Also, saw the interrupt of death a few times, so add in a warning when
this happens. Haven't seen it since handling interrupt overflows though.
Add concept of "discovered prints" and flesh out the API. A discovered print
was found based on file name but has not been checked for file integrity,
and of course there is no guarantee that it is usable when you try to
use it later (might have been deleted, etc).
The .fprint directory structure was changed - instead of using mnemonics
like "rind" we use their numeric value in the fp_finger enum as the filename.
Drivers can now offer a discover function, and are expected to offer the
device devtype during discovery, without accessing the USB device. Hmm.
Maybe we'll have to change that later...
Drivers now have an ID number. These will be assigned by me and documented
on the wiki. 0 cannot be used.
Drivers now define a devtype for each device they initialise. This is to
cope with the situation where a driver can support varying devices where
their print data is incompatible (i.e. image scaling is totally changed).
This doesn't apply to any existing supported devices.
Print data no longer includes driver name, and includes driver ID instead.
Paths to saved print data now include driver ID and devtype, and no longer
include driver name.
APIs exposed for converting a print_data into a blob which can then
be loaded back again later. Useful for systems who don't want to use
my simple storage system (which is only aimed at a single user).
File format is now defined and will be documented on the wiki. The header
is larger as we can no longer rely on directory paths in all scenarios.
Print data compat check now checks devtype and driver ID.
The basic model is that image drivers declare a fp_img_driver structure
rather than a fp_driver struct. fp_img_driver will contain primitive imaging
operations such as 'scan finger and return image'. The imgdev layer will
generically implement the primitive fp_driver operations, and the imgdev
layer will fix up the enroll/verify/etc pointers at driver registration
time.
Removed const from all fp_driver declarations, as these are now modified
dynamically in the case of imaging drivers.
Prints can now be saved to disk (but you currently must classify which
finger they are) and you can load them later.
Added 2 simple example programs to demonstrate this.
Convert enrollment function to return a signed integer, which is negative
on error or corresponds into fp_enroll_result otherwise.
Now we can treat a 'FAIL' condition differently from an actual error: fail
means that enrollment didn't complete because the data was nonsense or
whatever (e.g. scanned a different finger for each stage?).
Updated upekts accordingly.
This completes an enrollment implementation for the upekts driver.
It's based on code from libthinkfinger, but as there seemed to be many
unknown details there, I spent some time analysing the bus messages in
more detail and have effectively rewritten almost all of the borrowed code.
Note that libthinkfinger is GPL licensed, so that reflects on this code too.
At release time I will ask them if they are OK for me to relicense it.
upekts will need to know when the first enrollment stage is attempted for
the first time, so add an __enroll_stage counter which actually indicates
the next stage to enroll. -1 is a special value and it means 0 is next *and*
it is the initial attemt.
Added more debug output to the enroll handler.
Added new fp_enroll_status codes for too short or uncentered scans.
Changed the print_data allocator to consider the device rather than the
driver, this feels more natural. Added missing return value.
Make fp_enroll_status codes start at 1. 0 can now be used as a
special/temporary value by the drivers. Also check that we aren't exceeding
the number of enroll stages.
Also add a missing exit() call to the verify example and update for the
above.
Daniel Drake