[pHash-support] Initial Testing

Evan Klinger eklinger at phash.org
Sat Jan 30 08:57:28 PST 2010


How much resize, cropping and trimming are you talking about?
We are aware of the performance issue and we are investigating ways to
speed it up but at this time there is no getting around reading the
video frame by frame.
The video hash accepts any format that FFmpeg does, and it does
support flv files. You may be missing the codec for FFmpeg. I played
an FLV file last night using mplayer without any trouble.
You can also try playing around with the video hash parameters to see
if it helps when matching. Try adjusting the adaptive threshold lower
in the source. You can also tweak the sample rate. Currently it is
sampling every .5 second. Setting it to sample every second may help.
Please let us know your findings.

Thank you

On Sat, Jan 30, 2010 at 1:49 AM, David C <d140frv at hotmail.co.uk> wrote:
> Hi,
> These are the results of the initial testing from my developer:
> I uploaded different types of codec, trimmed & resize version of 1 video on
> the server. Upon testing, found out that video hash is not accurate yet. Its
> working fine if the 2 video compared have same length/duration and
> resolution. Buf if the video is resize, crop or trimmed, the phash get 0 in
> similarity :(  which is not good and the process is too long coz its process
> frame by frame. Video hash method doesn't accept flv files also. I've used
> to test the existing test_video on the source file coz all functions for
> video hash testing are their. So I conclude that it's not best to use the
> video hash method. If you've doubt with my testing, you may ask someone to
> thoroughly test the video hash. :)
> I am just confirming it is the variable length DCT.
> Thanks,
> David
> ________________________________
> Got a cool Hotmail story? Tell us now
> _______________________________________________
> pHash-support mailing list
> pHash-support at lists.phash.org
> http://lists.phash.org/listinfo.cgi/phash-support-phash.org

More information about the pHash-support mailing list