bprodz's profile - activity

@berak I've added them in an edit to the question using the 'add attachment' button, is that what you meant?

I'm trying to stitch 2 images together by using template matching find 3 sets of points which I pass to cv2.getAffineTransform() get a warp matrix which I pass to cv2.warpAffine() into to align my images.

However when I join my images the majority of my affine'd image isn't shown. I've tried using different techniques to select points, changed the order or arguments etc. but I can only ever get a thin slither of the affine'd image to be shown.

Could somebody tell me whether my approach is a valid one and suggest where I might be making an error? Thanks in advance.

This is the final result that I get. Here are the original images (1, 2) and the code that I use:

EDIT: Here are the images in jpeg format: , , .

import cv2
import numpy as np


def showimage(image, name="No name given"):
    cv2.imshow(name, image)
    cv2.waitKey(0)
    cv2.destroyAllWindows()
    return

image_a = cv2.imread('image_a.png')
image_b = cv2.imread('image_b.png')


def get_roi(image):
    roi = cv2.selectROI(image) # spacebar to confirm selection
    cv2.waitKey(0)
    cv2.destroyAllWindows()
    crop = image_a[int(roi[1]):int(roi[1]+roi[3]), int(roi[0]):int(roi[0]+roi[2])]
    return crop
temp_1 = get_roi(image_a)
temp_2 = get_roi(image_a)
temp_3 = get_roi(image_a)

def find_template(template, search_image_a, search_image_b):
    ccnorm_im_a = cv2.matchTemplate(search_image_a, template, cv2.TM_CCORR_NORMED)
    template_loc_a = np.where(ccnorm_im_a == ccnorm_im_a.max())

    ccnorm_im_b = cv2.matchTemplate(search_image_b, template, cv2.TM_CCORR_NORMED)
    template_loc_b = np.where(ccnorm_im_b == ccnorm_im_b.max())
    return template_loc_a, template_loc_b


coord_a1, coord_b1 = find_template(temp_1, image_a, image_b)
coord_a2, coord_b2 = find_template(temp_2, image_a, image_b)
coord_a3, coord_b3 = find_template(temp_3, image_a, image_b)

def unnest_list(coords_list):
    coords_list = [a[0] for a in coords_list]
    return coords_list

coord_a1 = unnest_list(coord_a1)
coord_b1 = unnest_list(coord_b1)
coord_a2 = unnest_list(coord_a2)
coord_b2 = unnest_list(coord_b2)
coord_a3 = unnest_list(coord_a3)
coord_b3 = unnest_list(coord_b3)

def unify_coords(coords1,coords2,coords3):
    unified = []
    unified.extend([coords1, coords2, coords3])
    return unified

# Create a 2 lists containing 3 pairs of coordinates
unified_pair1 = unify_coords(coord_a1, coord_a2, coord_a3)
unified_pair2 = unify_coords(coord_b1, coord_b2, coord_b3)

# Convert elements of lists to numpy arrays with data type float32
unified_pair1 = np.asarray(unified_pair1, dtype=np.float32)
unified_pair2 = np.asarray(unified_pair2, dtype=np.float32)

# Get result of the affine transformation
trans = cv2.getAffineTransform(unified_pair1, unified_pair2)

# Apply the affine transformation to original image
result = cv2.warpAffine(image_a, trans, (image_a.shape[1] + image_b.shape[1], image_a.shape[0]))
result[0:image_b.shape[0], image_b.shape[1]:] = image_b

showimage(result)
cv2.imwrite('result.png', result)

Did you ever manage to solve this problem? I'm having a similar issue.

I would like to create a panoramic image by combining 2 images in which the same feature, a plus sign.

I've used cv2.xfeatures2d.SIFT_create() to find keypoints in the image however it doesn't find the plus symbol very well. Is there some way I can improve this by making it search specifically for a plus-shaped feature?

import cv2

image1 = cv2.imread('example_image.png')
sift = cv2.xfeatures2d.SIFT_create()
kp = sift.detect(grey_image1, None)
kp_image = cv2.drawKeypoints(grey_image1, kp, None)

def showimage(image, name="No name given"):
    cv2.imshow(name, image)
    cv2.waitKey(0)
    cv2.destroyAllWindows()
    return

showimage(kp_image)

The source image is here, second image to align is here. Here is the resulting image from the code above. This is an example of the desired output made using GIMP and manually aligning the two images (the second image will need to transformed to fit properly).

NB I'm open to using other approaches outside of OpenCV/Python to solve this problem.