def test_find_haar_features():
img = Factory.Image(testimage)
img1 = img.copy()
face = HaarCascade("face.xml") # old HaarCascade
f = img.find_haar_features(face)
f2 = img1.find_haar_features("face_cv2.xml") # new cv2 HaarCascade
assert len(f) > 0
assert len(f2) > 0
f.draw()
f2.draw()
f[0].width
f[0].height
f[0].length
f[0].area
results = [img, img1]
name_stem = "test_find_haar_features"
perform_diff(results, name_stem)
# incorrect cascade name
f3 = img.find_haar_features(cascade="incorrect_cascade.xml")
assert_equals(f3, None)
# incorrect cascade object
f4 = img.find_haar_features(cascade=img1)
assert_equals(f4, None)
# Empty image
img2 = Factory.Image((100, 100))
f5 = img2.find_haar_features("face_cv2.xml")
assert_equals(f5, None)
def test_haarcascade():
img = Factory.Image(testimage)
faces = img.find_haar_features(FACECASCADE)
if faces:
faces.draw()
img.save(testoutput)
else:
assert False
cascade = HaarCascade(FACECASCADE, "face_cascade")
assert_equals(cascade.get_name(), "face_cascade")
fhandle = os.path.join(DATA_DIR, "HaarCascades", "face.xml")
assert_equals(cascade.get_fhandle(), fhandle)
cascade.set_name("eye_cascade")
assert_equals(cascade.get_name(), "eye_cascade")
new_fhandle = os.path.join(DATA_DIR, "HaarCascades", "eye.xml")
cascade.load(new_fhandle)
assert_equals(cascade.get_fhandle(), new_fhandle)
emptycascade = HaarCascade()
def find_haar_features(img,
cascade,
scale_factor=1.2,
min_neighbors=2,
use_canny=cv2.cv.CV_HAAR_DO_CANNY_PRUNING,
min_size=(20, 20),
max_size=(1000, 1000)):
"""
**SUMMARY**
A Haar like feature cascase is a really robust way of finding the
location of a known object. This technique works really well for a few
specific applications like face, pedestrian, and vehicle detection. It
is worth noting that this approach **IS NOT A MAGIC BULLET** . Creating
a cascade file requires a large number of images that have been sorted
by a human.vIf you want to find Haar Features (useful for face
detection among other purposes) this will return Haar feature objects
in a FeatureSet.
For more information, consult the cv2.CascadeClassifier documentation.
To see what features are available run img.list_haar_features() or you
can provide your own haarcascade file if you have one available.
Note that the cascade parameter can be either a filename, or a
HaarCascade loaded with cv2.CascadeClassifier(),
or a SimpleCV HaarCascade object.
**PARAMETERS**
* *cascade* - The Haar Cascade file, this can be either the path to a
cascade file or a HaarCascased SimpleCV object that has already been
loaded.
* *scale_factor* - The scaling factor for subsequent rounds of the
Haar cascade (default 1.2) in terms of a percentage
(i.e. 1.2 = 20% increase in size)
* *min_neighbors* - The minimum number of rectangles that makes up an
object. Ususally detected faces are clustered around the face, this
is the number of detections in a cluster that we need for detection.
Higher values here should reduce false positives and decrease false
negatives.
* *use-canny* - Whether or not to use Canny pruning to reject areas
with too many edges (default yes, set to 0 to disable)
* *min_size* - Minimum window size. By default, it is set to the size
of samples the classifier has been trained on ((20,20) for face
detection)
* *max_size* - Maximum window size. By default, it is set to the size
of samples the classifier has been trained on ((1000,1000) for face
detection)
**RETURNS**
A feature set of HaarFeatures
**EXAMPLE**
>>> faces = HaarCascade(
... "./SimpleCV/data/Features/HaarCascades/face.xml",
... "myFaces")
>>> cam = Camera()
>>> while True:
>>> f = cam.get_image().find_haar_features(faces)
>>> if f is not None:
>>> f.show()
**NOTES**
OpenCV Docs:
- http://opencv.willowgarage.com/documentation/python/
objdetect_cascade_classification.html
Wikipedia:
- http://en.wikipedia.org/wiki/Viola-Jones_object_detection_framework
- http://en.wikipedia.org/wiki/Haar-like_features
The video on this pages shows how Haar features and cascades work to
located faces:
- http://dismagazine.com/dystopia/evolved-lifestyles/8115/
anti-surveillance-how-to-hide-from-machines/
"""
if isinstance(cascade, basestring):
cascade = HaarCascade(cascade)
if not cascade.get_cascade():
return None
elif isinstance(cascade, HaarCascade):
pass
else:
logger.warning('Could not initialize HaarCascade. '
'Enter Valid cascade value.')
return None
haar_classify = cv2.CascadeClassifier(cascade.get_fhandle())
objects = haar_classify.detectMultiScale(img.to_gray(),
scaleFactor=scale_factor,
minNeighbors=min_neighbors,
minSize=min_size,
flags=use_canny)
if objects is not None and len(objects) != 0:
return FeatureSet(
[Factory.HaarFeature(img, o, cascade, True) for o in objects])
return None
def test_haarcascade():
img = Factory.Image(testimage)
faces = img.find_haar_features(FACECASCADE)
if faces:
faces.draw()
img.save(testoutput)
else:
assert False
cascade = HaarCascade(FACECASCADE, "face_cascade")
assert_equals(cascade.get_name(), "face_cascade")
fhandle = os.path.join(DATA_DIR, "HaarCascades", "face.xml")
assert_equals(cascade.get_fhandle(), fhandle)
cascade.set_name("eye_cascade")
assert_equals(cascade.get_name(), "eye_cascade")
new_fhandle = os.path.join(DATA_DIR, "HaarCascades", "eye.xml")
cascade.load(new_fhandle)
assert_equals(cascade.get_fhandle(), new_fhandle)
emptycascade = HaarCascade()
def find_haar_features(img, cascade, scale_factor=1.2, min_neighbors=2,
use_canny=cv2.cv.CV_HAAR_DO_CANNY_PRUNING,
min_size=(20, 20), max_size=(1000, 1000)):
"""
**SUMMARY**
A Haar like feature cascase is a really robust way of finding the
location of a known object. This technique works really well for a few
specific applications like face, pedestrian, and vehicle detection. It
is worth noting that this approach **IS NOT A MAGIC BULLET** . Creating
a cascade file requires a large number of images that have been sorted
by a human.vIf you want to find Haar Features (useful for face
detection among other purposes) this will return Haar feature objects
in a FeatureSet.
For more information, consult the cv2.CascadeClassifier documentation.
To see what features are available run img.list_haar_features() or you
can provide your own haarcascade file if you have one available.
Note that the cascade parameter can be either a filename, or a
HaarCascade loaded with cv2.CascadeClassifier(),
or a SimpleCV HaarCascade object.
**PARAMETERS**
* *cascade* - The Haar Cascade file, this can be either the path to a
cascade file or a HaarCascased SimpleCV object that has already been
loaded.
* *scale_factor* - The scaling factor for subsequent rounds of the
Haar cascade (default 1.2) in terms of a percentage
(i.e. 1.2 = 20% increase in size)
* *min_neighbors* - The minimum number of rectangles that makes up an
object. Ususally detected faces are clustered around the face, this
is the number of detections in a cluster that we need for detection.
Higher values here should reduce false positives and decrease false
negatives.
* *use-canny* - Whether or not to use Canny pruning to reject areas
with too many edges (default yes, set to 0 to disable)
* *min_size* - Minimum window size. By default, it is set to the size
of samples the classifier has been trained on ((20,20) for face
detection)
* *max_size* - Maximum window size. By default, it is set to the size
of samples the classifier has been trained on ((1000,1000) for face
detection)
**RETURNS**
A feature set of HaarFeatures
**EXAMPLE**
>>> faces = HaarCascade(
... "./SimpleCV/data/Features/HaarCascades/face.xml",
... "myFaces")
>>> cam = Camera()
>>> while True:
>>> f = cam.get_image().find_haar_features(faces)
>>> if f is not None:
>>> f.show()
**NOTES**
OpenCV Docs:
- http://opencv.willowgarage.com/documentation/python/
objdetect_cascade_classification.html
Wikipedia:
- http://en.wikipedia.org/wiki/Viola-Jones_object_detection_framework
- http://en.wikipedia.org/wiki/Haar-like_features
The video on this pages shows how Haar features and cascades work to
located faces:
- http://dismagazine.com/dystopia/evolved-lifestyles/8115/
anti-surveillance-how-to-hide-from-machines/
"""
if isinstance(cascade, basestring):
cascade = HaarCascade(cascade)
if not cascade.get_cascade():
return None
elif isinstance(cascade, HaarCascade):
pass
else:
logger.warning('Could not initialize HaarCascade. '
'Enter Valid cascade value.')
return None
haar_classify = cv2.CascadeClassifier(cascade.get_fhandle())
objects = haar_classify.detectMultiScale(
img.to_gray(), scaleFactor=scale_factor,
minNeighbors=min_neighbors, minSize=min_size,
flags=use_canny)
if objects is not None and len(objects) != 0:
return FeatureSet(
[Factory.HaarFeature(img, o, cascade, True) for o in objects])
return None
