convert rgb image from numpy array to HSV (opencv)

Question

When I'm converting an image from RGB to HSV, if the image come straight from opencv, everything is alright:

img = cv2.imread(path)
img = cv2.cvtColor(img, cv2.COLOR_BGR2HSV)

However, if this image come from a numpy array of shape (nb_of_images, 224, 224, 3) there is some complications.

Here is my import function:

def import_images(path_list):
    path_len = len(path_list)
    images = numpy.zeros((path_len, 224, 224, 3), dtype = numpy.float64)

    for pos in range(path_len):
        testimg = cv2.imread(path_list[pos])

        if(testimg is not None):
            testimg = cv2.cvtColor(testimg, cv2.COLOR_BGR2RGB)
            testimg = cv2.resize(testimg, (224, 224))
            images[pos, :, :, :] = testimg
    return images

And now, here is my trouble:

images = import_images(["./test/dog.jpg"])
img = images[0, :, :, :]
img = cv2.cvtColor(img, cv2.COLOR_RGB2HSV)

The console give the following error:

cv2.error: /io/opencv/modules/imgproc/src/color.cpp:11073: error: (-215) depth == 0 || depth == 2 || depth == 5 in function cvtColor

I tried to change the image type:

img.astype(numpy.float32)

but the console give the same error

What am I missing ?

--edit--

I'm using python 3.5

numpy (1.14.2)

opencv-python (3.4.0.12)

May be [this](https://stackoverflow.com/questions/7587490/converting-numpy-array-to-opencv-array) helps you! — Rick M., Apr 18 '18 at 14:49
@RickM. Maybe that last answer, the rest of it is using ancient deprecated opencv API. — Dan Mašek, Apr 18 '18 at 14:52
@DanMašek Difficult to pin point since OP hasn't mentioned the API version. @Mel Abgrall There could be an issue in this line as well `images[pos, :, :, :] = testimg` — Rick M., Apr 18 '18 at 14:53
@RickM. made an edit with versions. What kind of issue can cause images[pos, :, :, :] = testimg ? — Mael Abgrall, Apr 18 '18 at 14:58
@RickM. Not that difficult, given that OP used exclusively the new API. — Dan Mašek, Apr 18 '18 at 15:03
@DanMašek this was the source of the issue thanks ! I'll post an answer with the correct code — Mael Abgrall, Apr 18 '18 at 15:13

Dan Mašek · Accepted Answer · 2018-04-18T15:22:37.983

2

The problem is with the datatype of the elements in images. Right now it's np.float64.

Let's look at the assert in the C++ source code

CV_Assert( depth == CV_8U || depth == CV_16U || depth == CV_32F );

Translated to numpy, this means the data type of elements has to be np.uint8, np.uint16, or np.float32 for cvtColor to work at all. There are other more specific checks for some of the color conversions as well.

As you mention that 32 bit floats are sufficient for your use-case, you can do

images = numpy.zeros((path_len, 224, 224, 3), dtype = numpy.float32)

edited Apr 18 '18 at 15:22

answered Apr 18 '18 at 15:13

Dan Mašek

17,852
6
57
85

1

revised this line in function import_images: images = numpy.zeros((path_len, 224, 224, 3), dtype = numpy.float32) – Mael Abgrall Apr 18 '18 at 15:20

convert rgb image from numpy array to HSV (opencv)

1 Answers1