def testComputeTransformedBBox():
im1=io.imread('stata/stata-1.png')
im2=io.imread('stata/stata-2.png')
pointList1=[np.array([209, 218, 1]), np.array([425, 300, 1]), np.array([209, 337, 1]), np.array([396, 336, 1])]
pointList2=[np.array([232, 4, 1]), np.array([465, 62, 1]), np.array([247, 125, 1]), np.array([433, 102, 1])]
listOfPairsS=zip(pointList1, pointList2)
HS=a6.computeHomography(listOfPairsS)
shape = np.shape(im2)
print a6.computeTransformedBBox(shape, HS)
def compositeNImages(listOfImages, listOfH, weights, abrupt=False):
# Computes the composite image. listOfH is of the form returned by computeNHomographies.
# Hint: You will need to deal with bounding boxes and translations again in this function.
bbox = a6.computeTransformedBBox(listOfImages[0].shape, listOfH[0])
for img, H in zip(listOfImages, listOfH):
currentbbox = a6.computeTransformedBBox(img.shape, H)
bbox = a6.bboxUnion(currentbbox, bbox)
trans = a6.translate(bbox)
out = np.zeros((bbox[1][0] - bbox[0][0], bbox[1][1] - bbox[0][1], 3))
weightIm = np.zeros((out.shape[0], out.shape[1], 3))
for img, H in zip(listOfImages, listOfH):
applyHomographyFast(img, out, trans.dot(H), weights, weightIm, abrupt)
weightIm[weightIm == 0] = np.min(weightIm[weightIm.nonzero()]) # no zero in denominators
return out / weightIm
def compositeNImages(listOfImages, listOfH, weights, abrupt=False):
# Computes the composite image. listOfH is of the form returned by computeNHomographies.
# Hint: You will need to deal with bounding boxes and translations again in this function.
bbox = a6.computeTransformedBBox(listOfImages[0].shape, listOfH[0])
for img, H in zip(listOfImages, listOfH):
currentbbox = a6.computeTransformedBBox(img.shape, H)
bbox = a6.bboxUnion(currentbbox, bbox)
trans = a6.translate(bbox)
out = np.zeros((bbox[1][0] - bbox[0][0], bbox[1][1] - bbox[0][1], 3))
weightIm = np.zeros((out.shape[0], out.shape[1], 3))
for img, H in zip(listOfImages, listOfH):
applyHomographyFast(img, out, trans.dot(H), weights, weightIm, abrupt)
weightIm[weightIm == 0] = np.min(
weightIm[weightIm.nonzero()]) # no zero in denominators
return out / weightIm
def compositeNImages_blended(listOfImages, listOfHomographies, weights, onlyHighestWeight=False):
bbox = a6.computeTransformedBBox(np.shape(listOfImages[0]), listOfHomographies[0])
for (i, H) in enumerate(listOfHomographies):
bboxI = a6.computeTransformedBBox(np.shape(listOfImages[i]), H)
bbox = a6.bboxUnion(bboxI, bbox)
height = bbox[1][0] - bbox[0][0]
width = bbox[1][1] - bbox[0][1]
out = np.zeros([height, width, 3])
out_weight = np.zeros([height, width, 3])
translation = a6.translate(bbox)
for i in xrange(len(listOfImages)):
applyHomographyFast_blended(
listOfImages[i], out, np.dot(translation, listOfHomographies[i]), out_weight, weights, onlyHighestWeight
)
out_weight[out_weight == 0] = 1e-12
return out / out_weight
def applyHomographyFast(source, out, H, weights, weightIm, abrupt=False):
# takes the image source, warps it by the homography H, and adds it to the composite out.
# This version should only iterate over the pixels inside the bounding box of source's image in out.
bbox = a6.computeTransformedBBox(source.shape, H)
Hinv = linalg.inv(H)
for y in xrange(bbox[0][0], bbox[1][0]):
for x in xrange(bbox[0][1], bbox[1][1]):
yp, xp, w = Hinv.dot(np.array([y, x, 1.0]))
yp, xp = (yp / w, xp / w)
if yp >= 0 and yp < source.shape[0] and xp >= 0 and xp < source.shape[1]:
if abrupt and weights[yp, xp, 0] > weightIm[y, x, 0]:
weightIm[y, x] = weights[yp, xp]
out[y, x] = weights[yp, xp] * a6.interpolateLin(source, yp, xp)
elif not abrupt:
weightIm[y, x] += weights[yp, xp]
out[y, x] += weights[yp, xp] * a6.interpolateLin(source, yp, xp)
def applyHomographyFast_blended(source, out, H, out_weight, weight_map, onlyHighestWeight=False):
"""Takes the image source, warps it by the homography H, and adds it to the composite out.
This version should only iterate over the pixels inside the bounding box of source's image in out. """
bbox = a6.computeTransformedBBox(np.shape(source), H)
for y in xrange(bbox[0][0], bbox[1][0]):
for x in xrange(bbox[0][1], bbox[1][1]):
(yp, xp, wp) = np.dot(linalg.inv(H), np.array([y, x, 1]))
(ypp, xpp) = (yp / wp, xp / wp)
if a6.within_bounds(source, ypp, xpp):
if not onlyHighestWeight:
out[y, x] += a6.interpolateLin(source, ypp, xpp) * weight_map[ypp, xpp]
out_weight[y, x] += weight_map[ypp, xpp]
elif np.sum(weight_map[ypp, xpp]) > np.sum(out_weight[y, x]):
out[y, x] = a6.interpolateLin(source, ypp, xpp) * weight_map[ypp, xpp]
out_weight[y, x] = weight_map[ypp, xpp]
else:
pass
def applyHomographyFast(source, out, H, weights, weightIm, abrupt=False):
# takes the image source, warps it by the homography H, and adds it to the composite out.
# This version should only iterate over the pixels inside the bounding box of source's image in out.
bbox = a6.computeTransformedBBox(source.shape, H)
Hinv = linalg.inv(H)
for y in xrange(bbox[0][0], bbox[1][0]):
for x in xrange(bbox[0][1], bbox[1][1]):
yp, xp, w = Hinv.dot(np.array([y, x, 1.0]))
yp, xp = (yp / w, xp / w)
if yp >= 0 and yp < source.shape[
0] and xp >= 0 and xp < source.shape[1]:
if abrupt and weights[yp, xp, 0] > weightIm[y, x, 0]:
weightIm[y, x] = weights[yp, xp]
out[y, x] = weights[yp, xp] * a6.interpolateLin(
source, yp, xp)
elif not abrupt:
weightIm[y, x] += weights[yp, xp]
out[y, x] += weights[yp, xp] * a6.interpolateLin(
source, yp, xp)
