Feature Detection in OpenCV Python Bindings

Question

I've combed the web looking for a way to get the OpenCV 2.3.1a feature extraction/descriptor bindings to spit out any flavor of image features/descriptors(STAR/SURF/ORB/SIFT/FAST). I am well aware that OpenCV has a method called "goodFeaturesToTrack. This doesn't help me as there are no feature descriptors (which is what I really need). I have followed the documentation as listed here:

http://opencv.itseez.com/modules/features2d/doc/feature_detection_and_description.html

Nothing seems to work. I've tried all of the flavors of descriptors/features. I've tried using single and multiple channel images (i.e. color and black and white) and multiple image formats (8bit and 32f). I have worked with the current distribution and building the bindings from the source repo. Most of the methods result in a "unknown is not a numpy array" error. Here is an example:

SimpleCV:1>import cv2
SimpleCV:2>img = Image("aerospace.jpg")
SimpleCV:3>bwimg = img._getGrayscaleBitmap()
SimpleCV:4>bwimg
SimpleCV:4><iplimage(nChannels=1 width=600 height=400 widthStep=600 )>
SimpleCV:5>surfer = cv2.SURF(0.5,4,2,False,False)
SimpleCV:6>points = surfer.detect(bwimg,None)
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)

/Library/Python/2.6/site-packages/SimpleCV-1.2-py2.6.egg/SimpleCV/Shell/Shell.pyc in <module>()
-

TypeError: <unknown> is not a numpy array
SimpleCV:7>

It is worth noting that I am using SimpleCV to load the image, but the method _getGrayscaleBitmap() returns the gray 8bit IPL image used by OpenCV. I am sure this works as I use it with hundred of other OpenCV methods without incidence.

So can anyone point me to a WORKING example of this code on the web. I have combed through dozens of examples and found nothing that works.

For what it is worth I submitted this ticket to Willow Garage, but I would love to find a solution ASAP. code.ros.org/trac/opencv/ticket/1582 — kscottz
– kscottz, Commented Feb 3, 2012 at 16:45

Jan Erik Solem · Accepted Answer · 2012-02-03 17:22:14Z

10

Kat, this works for me:

s = cv2.SURF()
mask = uint8(ones(gray.shape))
keypoints = s.detect(gray,mask)

I can plot the key points and all. To get the descriptors you can try this

k,d = s.detect(gray,mask,False)
d = d.reshape((-1,128))
print d.shape, len(k)

d should have the same length at the list of key points.

I have this example in the OpenCV chapter here: http://www.maths.lth.se/matematiklth/personal/solem/book.html

answered Feb 3, 2012 at 17:22

Jan Erik Solem

1811 silver badge2 bronze badges

Sign up to request clarification or add additional context in comments.

4 Comments

kscottz Over a year ago

Still getting the error. Can you do me a favor and tell me the type of the gray image? it looks like a numpy array. Do the cvMat and IPL Image types not work?

kscottz Over a year ago

Okay from SimpleCV this looks like it will work: gray = uint8(np.array(cv.GetMat(img._getGrayscaleBitmap())).transpose()) It will take me a day or so to do a proof of concept.

Jan Erik Solem Over a year ago

Yeah, it is a uint8 numpy array. The cv2 python bindings need some work, most of the features don't seem to be implemented yet.

xamox Over a year ago

I also came across this problem, although it could be the bindings may have been busted: tech.groups.yahoo.com/group/OpenCV/message/85320 I was about to reference your python book using VLFeat (although the python wrappers are outdated). :)

Andy Rosenblum · Accepted Answer · 2012-08-15 19:46:07Z

0

Looks like you have a PIL image. Try converting to a numpy image: npImage = np.array(img)

answered Aug 15, 2012 at 19:46

Andy Rosenblum

2431 gold badge4 silver badges10 bronze badges

Collectives™ on Stack Overflow

Feature Detection in OpenCV Python Bindings

2 Answers 2

4 Comments

Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

4 Comments

Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related