Detect faces and features in images to help cropping them
Switch branches/tags
Nothing to show
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Permalink
Failed to load latest commit information.
ccv
opencv
.gitignore
MIT-LICENSE
README.md
compare.rb

README.md

Face'n'feature detection

Detect faces and features in images to help cropping them.

There are two versions of this tool:

  • One is a simple wrapper around the Haar & LBP Cascade classifiers of OpenCV
  • The other is based on a SURF-Cascade Detection from ccv.

The ccv version detects more faces, but the OpenCV version is about 3 times faster and also detects features. Choose the version that fits you best!

Install

For OpenCV:

cd opencv
apt-get install build-essential cmake libopencv-dev
cmake -DCMAKE_BUILD_TYPE=Release .
make

For ccv:

cd ccv
apt-get install build-essential libjpeg-dev libpng12-dev
make ccv
make

Usage

fnf-detect <image>

It will output one face/feature per line, with:

<x> <y> <w> <h> <type>
<x> is the X coordinate (from the left)
<y> is the Y coordinate (from the top)
<w> is the width
<h> is the height
<type> is face, profile or feature

Note : the ccv version can only detect faces.

Examples

$ ./fnf-detect image.jpg
302 302 1559 708 face

It can be a lot faster to make a thumbnail of an image and extract faces and features from it, and then extrapolate the coordinates. It's just a small bit less accurate:

$ convert image.jpg -resize '25%' thumbnail.jpg

$ identify image.jpg
image.jpg JPEG 3648x2736 3648x2736+0+0 8-bit DirectClass 2.199MB 0.000u 0:00.000

$ identify thumbnail.jpg
thumbnail.jpg JPEG 912x684 912x684+0+0 8-bit DirectClass 172KB 0.000u 0:00.000

$ time ./fnf-detect image.jpg
2251 1013 352 352 face
./fnf-detect image.jpg  1,39s user 0,03s system 454% cpu 0,312 total

$ time ./fnf-detect thumbnail.jpg
563 253 88 88 face
./fnf-detect thumbnail.jpg  0,16s user 0,03s system 189% cpu 0,101 total

$ echo $((563 * 4)) $((253 * 4)) $((88 * 4)) $((88 * 4))
2252 1012 352 352

Inspiration

Credits

Copyright af83 (c) 2015

Released under the MIT license