Panoramic_view_creation

This project allows you to create a panoramic view by selecting a corresponding point. Using this correspondences, warp the src img by backward warping and this allows you to make panoramic img using stitch function.

get_coord.py

functions: get_coord

homography.py

functions: get_homography(src, dst),
ransac(src_points, dst_points, iteration, threshold)

stitch.py

functions: warp(image, H, target_w, target_h),
stitch(src_img, dst_img)

main.py

functions: stitch_all()

get_coord(ssrc_img, ddst_img): This function allows you to choose correspondences between two images

Args:
    ssrc_img: np.ndarray with shape (src_h,src_w,c)
    ddst_img: np.ndarray with shape (dst_h,dst_w,c)
Returns:
    src_points: np.ndarray with shape (num_points,2), 2=> xy(or wh) axis
    dst_points: np.ndarray with shape (num_points,2), 2=> xy(or wh) axis
    (i)th correspondence: src_points[i] <-> dst_points[i]

[Usage guideline]
- Two windows (left for src image and right for dst image) are created
1) Click a point at src image and press enter
2) Click the corresponding point at dst image and press enter
- One point correspondence is registered
3) Repeat multiple times as you want
4) After indicating multiple point correspondences,
    press enter twice in a row (started by src image) to exit

get_homography(src_points, dst_points): Input of this function can be generated by get_coord function. The output is homography matrix between src_img and dst_img by your own correspondences

  Args:
    src_points: np.ndarray with shape (num_points,2), 2=> xy(or wh) axis
    dst_points: np.ndarray with shape (num_points,2), 2=> xy(or wh) axis
    (i)th correspondence: src_points[i] <-> dst_points[i]
    
  Returns:
      H: Homography matrix with shape (3,3)
      H[2,2] = 1

Ransac(src_points, dst_points, iteration, threshold) :This function allows you to find the best homography among all possibles. The best homography implies that this H has the biggest number of inliers

 Args:
    src_points: np.ndarray with shape (num_points,2), 2=> xy(or wh) axis
    dst_points: np.ndarray with shape (num_points,2), 2=> xy(or wh) axis
    (i)th correspondence: src_points[i] <-> dst_points[i]
    iteration : the number of iterations to find best homography
    threshold : If the error ( input dst_points and calculated dst_points using homography) is less than threshold, then this point becomes inlier.
    
    Output: 
      H: Best Homography matrix with shape(3,3) , H[2,2] = 1

warp(image, H, target_w, target_h): Warp src_img (using homography matrix H) to the target plane with a window size=(target_h,target_w)

  Args:
    src_img: np.ndarray with shape (src_h,src_w,c)
    H: np.ndarray with shape (3,3)
    target_h: int
    target_w: int
Returns:
    Warped src image
    np.ndarray with shape (target_h,target_w,c)

stitch(src_img, dst_img):

1)Manually indicate corresponding points
2) Find src_img -> dst_img homography matrix
3) Warp src_img and stitch with dst_img 

Args:
    src_img: np.ndarray with shape (src_h,src_w,c)
    dst_img: np.ndarray with shape (dst_h,dst_w,c)
Returns:
    src,dst stitched image. np.ndarray with shape (target_h,target_w,c)
Notice:
    get_homography() or ransac() function should be used for computing homography
    warp() function should be used for warping an image

[Example]

#stitched boudary