Build

Clone both ccv and py-ccv:

mkdir ccv_build
cd ccv_build
git clone https://github.com/liuliu/ccv
git clone https://github.com/gpip/py-ccv

Patch ccv so it can be compiled as a dynamic lib, and compile it:

cd ccv/lib
patch -p2 < ../../py-ccv/dynlib.patch
./configure
make libccv.so

Build and install the Python wrapper (ARCHFLAGS was used on a OSX build, adjust for your platform):

cd ../../py-ccv
ARCHFLAGS='-arch x86_64' INCDIR=../ccv/lib LIBDIR=../ccv/lib python setup.py install

Build only this wrapper

If you already have libccv.so then you might want to install directly via pip:

LDFLAGS="-L$(pwd)/ccv/lib" CFLAGS="-I$(pwd)/ccv/lib" pip install ccv

Remember to adjust the paths according to where libccv.so and ccv.h are installed in your system.

Face Detection Usage

(DYLD_LIBRARY_PATH was used on OSX, adjust it for your platform)

Face detection using SCD

$ DYLD_LIBRARY_PATH=../ccv/lib python -m ccv.face_detect -c ../ccv/samples/face.sqlite3 img/lena.png
img/lena.png Feature(x1=229, y1=216, x2=381, y2=368, confidence=5.014610767364502)

Face detection using BBF

$ DYLD_LIBRARY_PATH=../ccv/lib python -m ccv.face_detect --bbf -c ../ccv/samples/face img/lena.png
img/lena.png Feature(x1=230, y1=211, x2=384, y2=365, confidence=0.4947386682033539)

Help

$ DYLD_LIBRARY_PATH=../ccv/lib python -m ccv.face_detect.py --help
Usage: face_detect.py [options] filename...

Options:
  -h, --help            show this help message and exit
  --bbf                 Use BBF detector
  --scd                 Use SCD detector
  -c CASCADE, --cascade=CASCADE
                        Path to cascade to read
  --quiet

Using face_detect as a library

from ccv import face_detect

names = ['img/lena.png']
result = face_detect.main('scd', '../ccv/samples/face.sqlite3', False, *names)

# face_detect.main is a generator which yields tuples of
# (<filename>, [<list of face_detect.Feature>])
for name, rects in result:
    print name, rects

Visualization

$ convert img/lena.png -fill none -stroke blue -strokewidth 3 -draw "rectangle 229,216 381,368" result.png

Using the library

import sys
from ccv import ccv_read, ccv_write, sobel, lib

# Read file passed.
inp = ccv_read(sys.argv[1])
# Apply Sobel.
res = sobel(inp, lib.CCV_8U)
# Save the result as "sobel.jpg"
ccv_write(res, "sobel.jpg")

Pointers returned by the higher level wrapper, ccv, are automatically freed.

Functions available from import ccv

ccv.ccv_read(filename, ttype=None)

Read an image from a filename and return an internal ccv_dense_matrix_t*. ttype can be specified to typically convert the input image to grayscale: ttype=ccv.lib.CCV_IO_GRAY | ccv.lib.CCV_IO_ANY_FILE.

---

ccv.ccv_write(im, outname)

Write an image (returned from ccv_read or other compatible functions) to a filename specified by outname.

---

ccv.ccv_slice(im, x, y, rows, cols)

Slice an input matrix given offsets x and y, and number of rows and columns. A new matrix is returned.

ccv_slice can be used to effectively crop a region of an input image and return a cropped image.

---

ccv.sobel(im, ttype, dx=1, dy=1)

Apply the Sobel operator to the input and return a new matrix. dx and dy dictate the window size for the operator, and is optimized for values of 1 and 3. ttype defined the type of the output matrix, which for typical images should be set to ccv.lib.CCV_8U.

---

ccv.gradient(im, dx=1, dy=1)

Compute the gradient at each pixel and return, respectively, theta and magnitude matrices of type ccv.lib.CCV_32F. Typically, these matrices will be converted via ccv.visualize if you want to visually inspect them.

---

ccv.visualize(mat)

Convert an input matrix into a matrix within visual range.

---

ccv.prepare_scd_cascade(inp)

Read SCD classifier cascade from inp and return a ccv_scd_classifier_cascade_t*.

---

ccv.prepare_bbf_cascade(inp)

Read BBF classifier cascade from inp and return a ccv_bbf_classifier_cascade_t*.

---

ccv.scd_detect_objects(filename, cascade)

Use SURF-Cascade object detection by reading an image from filename and using the cascade obtained from ccv.prepare_scd_cascade. A list of ccv.Feature is returned.

---

ccv.bbf_detect_objects(filename, cascade)

Use Brightness Binary Feature object detection by reading an image from filename and using the cascade obtained from ccv.prepare_bbf_cascade. A list of ccv.Feature is returned.

---

ccv.Feature

A namedtuple with parameters x1, y1, x2, y2, and confidence.