Using openCV, resize multiple the same size of images at once in Python

Question

One image can be resized using opencv on python3

import cv2 
res_image = cv2.resize(image, dsize=(50, 100))

Also, multiple images can be resized through for syntax.

for image in images:
  res_image = cv2.resize(image, dsize=(50, 100))

But I want to resize the the same size of images(array) at once instead of using for loop.

 >>> images.shape 
 (32, 64, 64, 3)

Please let me know if you have any solve to this problem. (even if don't use opencv)

thank you.

Christoph Rackwitz · Accepted Answer · 2023-08-10 16:06:26Z

2

cv.resize() apparently supports an arbitrary number of channels. Since it doesn't mix data between channels, you could use this to your advantage.

Note that cv::Mat() currently only supports upto 512 channels. That is relevant here because numpy arrays are mapped to cv::Mat for OpenCV to handle the data.

Approach:

transpose from NHWC to HWCN
reshape to (H, W, C*N)
resize
undo reshape
undo transpose

images = ...
(N,H,W,C) = images.shape
assert N*C <= 512

instack = images.transpose((1,2,3,0)).reshape((H,W,C*N))

outstack = cv.resize(instack, ...)

out_images = outstack.reshape((H,W,C,N)).transpose((3,0,1,2))

Beyond that number of channels, you will have to use a loop.

edited Aug 10, 2023 at 16:06

answered Dec 5, 2020 at 18:33

Christoph Rackwitz

16.4k5 gold badges42 silver badges56 bronze badges

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

Comments

Jubin Ben · Accepted Answer · 2022-08-24 06:32:55Z

0

The following code could do the trick.

def resize_batch(img_batch, resize_width, resize_height):
    """
    :params
        image: np.array(), shape -> (batch, width, height, channels)
        resize_width: The resize width dimension. 
        resize_height: The resize height dimension. 

    :returns
        array of shape -> (batch, resized_width, resized_height, channels)
    """
    batch, original_width, original_height, channel = img_batch.shape
    
    rd_ch = img_batch[:,:,:,0]
    gr_ch = img_batch[:,:,:,1]
    bl_ch = img_batch[:,:,:,2]
    
    resized_images = np.zeros((batch, resize_width, resize_height, channel), dtype=np.uint8)
    
    x_scale = original_width/resize_width
    y_scale = original_height/resize_height
    
    resize_idx = np.zeros((resize_width, resize_height))
    resize_index_x = np.ceil(np.arange(0, original_width, x_scale)).astype(int)
    resize_index_y = np.ceil(np.arange(0, original_height, y_scale)).astype(int)
    resize_index_x[np.where(resize_index_x == original_width)]  -= 1
    resize_index_y[np.where(resize_index_y == original_height)] -= 1
    
    resized_images[:,:,:,0] = rd_ch[:,resize_index_x,:][:,:,resize_index_y]
    resized_images[:,:,:,1] = gr_ch[:,resize_index_x,:][:,:,resize_index_y]
    resized_images[:,:,:,2] = bl_ch[:,resize_index_x,:][:,:,resize_index_y]
    
    return resized_images

edited Aug 24, 2022 at 6:32

answered Aug 24, 2022 at 5:58

Jubin Ben

293 bronze badges

2 Comments

Christoph Rackwitz Over a year ago

what does that do? nearest neighbor sampling? you're throwing around index arrays. I think that'll be slow. also... crosspost: stackoverflow.com/questions/64576675/…

Jubin Ben Over a year ago

The code above is just for resizing all the images in a batch. Yes, I understand it's slower than using cv2.resize() in a loop. The above code was just a part of an experiment. I was trying to make my own resize function for a particular batch of images. I thought using NumPy arrays could make the resize operation faster. I thought my above code could give someone an understanding of how to create their own resize function, or maybe someone could suggest me a better approach to resize all images in a batch more quickly.

Collectives™ on Stack Overflow

Using openCV, resize multiple the same size of images at once in Python

2 Answers 2

Comments

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

2 Answers 2

Comments

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related