fast conversion of IplImage to Numpy array

Question

The newer OpenCV documentation here says you can convert an IplImage to a Numpy array just like this:

arr = numpy.asarray( im )

but that doesn't work for my needs, because it apparently doesn't support math:

x = arr/0.01
TypeError: unsupported operand type(s) for /: 'cv2.cv.iplimage' and 'float'

If I try to specify data type, I can't even get that far:

arr = numpy.asarray( im, dtype=num.float32 )
TypeError: float() argument must be a string or a number

So I'm using the code provided in the older documentation here. Basically, it does this:

arr = numpy.fromstring( im.tostring(), dtype=numpy.float32 )

But the tostring call is really slow, perhaps because it's copying the data? I need this conversion to be really fast and not copy any buffers it doesn't need to. I don't think the data are inherently incompatible; I'm creating my IplImage with cv.fromarray in the first place, which is extremely fast and accepted by the OpenCV functions.

Is there a way I can make the newer asarray method work for me, or else can I get direct access to the data pointer in the IplImage in a way that numpy.fromstring will accept it? I'm using OpenCV 2.3.1 prepackaged for Ubuntu Precise.

OpenCV is under heavy development, also the Python bindings. You should try OpenCV 2.4.2 or 2.4.3rc first. — ypnos
– ypnos, Commented Oct 27, 2012 at 21:50
The question is valid. There is already some legacy Python code for OpenCV cv bindgins. They are still supported as of OpenCV 2.4.2. Converting them to cv2 may not always be an option. — sastanin
– sastanin, Commented Dec 28, 2012 at 18:15

Don Cruickshank · Accepted Answer · 2013-04-28 15:02:43Z

9

Fun Fact:
Say you call:

import cv2.cv as cv    #Just a formality!

Capture = cv.CaptureFromCAM(0)
Img = cv.QueryFrame(Capture)

The object Img is an ipimage, and numpy.asarray(Img) is erratic at best. However! Img[:,:] is a cvmat type, and numpy.asarray(Img[:,:]) works fantastically, and more important: quickly!

This is by far the fastest way I've found to grab a frame and make it an ndarray for numpy processing.

edited Apr 28, 2013 at 15:02

Don Cruickshank

6,0186 gold badges53 silver badges50 bronze badges

answered Apr 28, 2013 at 14:32

chris r.

1061 silver badge2 bronze badges

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

1 Comment

jab Over a year ago

Awesome, I will try that. It won't be soon, unfortunately.

Andrey Kamaev · Accepted Answer · 2012-10-28 10:07:09Z

0

That page does not say about IplImage. It says about CvMat which is different.

Anyway you'd better use wrappers from newer cv2 namespace. It natively uses numpy arrays instead of own image containers. Also the whole cv module is considered deprecated and will be completely dropped in the nearest major release.

answered Oct 28, 2012 at 10:07

Andrey Kamaev

30.2k6 gold badges100 silver badges89 bronze badges

2 Comments

jab Over a year ago

You mean I should use cv2.fromarray() or something like that, and then numpy.asarray() will magically work? Can you provide a link to docs?

jab Over a year ago

Also, how the *#@$ was anybody supposed to know this? I've been using OpenCV off and on since version 0.1, and I always run away screaming. It's like a horror movie -- I'm afraid there will be a monster behind the door, but I open it anyway.

Collectives™ on Stack Overflow

fast conversion of IplImage to Numpy array

2 Answers 2

1 Comment

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

1 Comment

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related