I am analyzing multiple images and need to be able to tell if they are shifted compared to a reference image. The purpose is to tell if the camera moved at all in between capturing images. I would ideally like to be able to correct the shift in order to still do the analysis, but at a minimum I need to be able to determine if an image is shifted and discard it if it's beyond a certain threshold.
Here are some examples of the shifts in an image I would like to detect:
I will use the first image as a reference and then compare all of the following images to it to figure out if they are shifted. The images are gray-scale (they are just displayed in color using a heat-map) and are stored in a 2-D numpy array. Any ideas how I can do this? I would prefer to use the packages I already have installed (scipy, numpy, PIL, matplotlib).
As Lukas Graf hints, you are looking for cross-correlation. It works well, if:
The scale of your images does not change considerably.
There is no rotation change in the images.
There is no significant illumination change in the images.
For plain translations cross-correlation is very good.
The simplest cross-correlation tool is scipy.signal.correlate. However, it uses the trivial method for cross-correlation, which is O(n^4) for a two-dimensional image with side length n. In practice, with your images it'll take very long.
The better too is scipy.signal.fftconvolve as convolution and correlation are closely related.
Something like this:
import numpy as np
import scipy.signal
def cross_image(im1, im2):
# get rid of the color channels by performing a grayscale transform
# the type cast into 'float' is to avoid overflows
im1_gray = np.sum(im1.astype('float'), axis=2)
im2_gray = np.sum(im2.astype('float'), axis=2)
# get rid of the averages, otherwise the results are not good
im1_gray -= np.mean(im1_gray)
im2_gray -= np.mean(im2_gray)
# calculate the correlation image; note the flipping of onw of the images
return scipy.signal.fftconvolve(im1_gray, im2_gray[::-1,::-1], mode='same')
The funny-looking indexing of im2_gray[::-1,::-1] rotates it by 180° (mirrors both horizontally and vertically). This is the difference between convolution and correlation, correlation is a convolution with the second signal mirrored.
Now if we just correlate the first (topmost) image with itself, we get:
This gives a measure of self-similarity of the image. The brightest spot is at (201, 200), which is in the center for the (402, 400) image.
The brightest spot coordinates can be found:
np.unravel_index(np.argmax(corr_img), corr_img.shape)
The linear position of the brightest pixel is returned by argmax, but it has to be converted back into the 2D coordinates with unravel_index.
Next, we try the same by correlating the first image with the second image:
The correlation image looks similar, but the best correlation has moved to (149,200), i.e. 52 pixels upwards in the image. This is the offset between the two images.
This seems to work with these simple images. However, there may be false correlation peaks, as well, and any of the problems outlined in the beginning of this answer may ruin the results.
In any case you should consider using a windowing function. The choice of the function is not that important, as long as something is used. Also, if you have problems with small rotation or scale changes, try correlating several small areas agains the surrounding image. That will give you different displacements at different positions of the image.
Another way to solve it is to compute sift points in both images, use RANSAC to get rid of outliers and then solve for translation using a least squares estimator.
as Bharat said as well another is using sift features and Ransac:
import numpy as np
import cv2
from matplotlib import pyplot as plt
def crop_region(path, c_p):
"""
This function crop the match region in the input image
c_p: corner points
"""
# 3 or 4 channel as the original
img = cv2.imread(path, -1)
# mask
mask = np.zeros(img.shape, dtype=np.uint8)
# fill the the match region
channel_count = img.shape[2]
ignore_mask_color = (255,)*channel_count
cv2.fillPoly(mask, c_p, ignore_mask_color)
# apply the mask
matched_region = cv2.bitwise_and(img, mask)
return matched_region
def features_matching(path_temp,path_train):
"""
Function for Feature Matching + Perspective Transformation
"""
img1 = cv2.imread(path_temp, 0) # template
img2 = cv2.imread(path_train, 0) # input image
min_match=10
# SIFT detector
sift = cv2.xfeatures2d.SIFT_create()
# extract the keypoints and descriptors with SIFT
kps1, des1 = sift.detectAndCompute(img1,None)
kps2, des2 = sift.detectAndCompute(img2,None)
FLANN_INDEX_KDTREE = 0
index_params = dict(algorithm = FLANN_INDEX_KDTREE, trees = 5)
search_params = dict(checks = 50)
flann = cv2.FlannBasedMatcher(index_params, search_params)
matches = flann.knnMatch(des1, des2, k=2)
# store all the good matches (g_matches) as per Lowe's ratio
g_match = []
for m,n in matches:
if m.distance < 0.7 * n.distance:
g_match.append(m)
if len(g_match)>min_match:
src_pts = np.float32([ kps1[m.queryIdx].pt for m in g_match ]).reshape(-1,1,2)
dst_pts = np.float32([ kps2[m.trainIdx].pt for m in g_match ]).reshape(-1,1,2)
M, mask = cv2.findHomography(src_pts, dst_pts, cv2.RANSAC,5.0)
matchesMask = mask.ravel().tolist()
h,w = img1.shape
pts = np.float32([ [0,0],[0,h-1],[w-1,h-1],[w-1,0] ]).reshape(-1,1,2)
dst = cv2.perspectiveTransform(pts,M)
img2 = cv2.polylines(img2, [np.int32(dst)], True, (0,255,255) , 3, cv2.LINE_AA)
else:
print "Not enough matches have been found! - %d/%d" % (len(g_match), min_match)
matchesMask = None
draw_params = dict(matchColor = (0,255,255),
singlePointColor = (0,255,0),
matchesMask = matchesMask, # only inliers
flags = 2)
# region corners
cpoints=np.int32(dst)
a, b,c = cpoints.shape
# reshape to standard format
c_p=cpoints.reshape((b,a,c))
# crop matching region
matching_region = crop_region(path_train, c_p)
img3 = cv2.drawMatches(img1, kps1, img2, kps2, g_match, None, **draw_params)
return (img3,matching_region)
Related
So I have a template and an image. I want to find the location and orientation of the template inside the image. I am using SIFT to find features and description.
Problem is only one feature is consistently correct at recognizing the image. Homography requires at least 4 features to work. error: (-28:Unknown error code -28) The input arrays should have at least 4 corresponding point sets to calculate Homography in function 'cv::findHomography'
Since I am working with 2D image (with same scale), position and rotation of even one correct feature should be enough to provide the location and rotation of the template in the image.
From OpenCV Docs https://docs.opencv.org/3.4/da/df5/tutorial_py_sift_intro.html
OpenCV also provides cv.drawKeyPoints() function which draws the small
circles on the locations of keypoints. If you pass a flag,
cv.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS to it, it will draw a circle
with size of keypoint and it will even show its orientation.
However the image I am working is too low resolution to actually see the circles and I need numbers which can compared.
All the other examples of finding orientation I find on the internet use edge detection. However There is no straight edge whose slope can be easily calculated exist in my template.
This solution can help, however my images could potentially have other unwanted objects which will mess with "minAreaRect". If there is any other solution, please let me know.
I have looked for tutorial, books, documentation on how to crunch the numbers in 'keypoints' and 'description', but I could not find any.
Perhaps I should use SURF -which is faster with 2d, same color images- but it is not available in latest opencv version.
Template to be searched
Image to be searched in
Matched
sift = cv.SIFT_create()
# find the keypoints and descriptors with SIFT
kp1, des1 = sift.detectAndCompute(img1, None)
kp2, des2 = sift.detectAndCompute(img2, None)
print (des1)
# BFMatcher with default params
bf = cv.BFMatcher()
matches = bf.knnMatch(des1,des2,k=2)
# Apply ratio test
good = []
good_match = []
for m,n in matches:
if m.distance < .5*n.distance:
good.append([m])
good_match.append(m)
print('good matches are')
print(good)
print(good_match)
# cv.drawMatchesKnn expects list of lists as matches.
img3 = cv.drawMatchesKnn(img1,kp1,img2,kp2,good,None,flags=cv.DRAW_MATCHES_FLAGS_DRAW_RICH_KEYPOINTS)
plt.imshow(img3),plt.show()
src_pts = np.float32([ kp1[m.queryIdx].pt for m in good_match ]).reshape(-1,1,2)
dst_pts = np.float32([ kp2[m.trainIdx].pt for m in good_match ]).reshape(-1,1,2)
M, mask = cv.findHomography(src_pts, dst_pts, cv.RANSAC,5.0)
matchesMask = mask.ravel().tolist()
h,w = img1.shape
pts = np.float32([ [0,0],[0,h-1],[w-1,h-1],[w-1,0] ]).reshape(-1,1,2)
dst = cv.perspectiveTransform(pts,M)
img2 = cv.polylines(img2,[np.int32(dst)],True,255,3, cv.LINE_AA)
draw_params = dict(matchColor = (0,255,0), # draw matches in green color
singlePointColor = None,
matchesMask = matchesMask, # draw only inliers
flags = 2)
img3 = cv.drawMatches(img1,kp1,img2,kp2,good,None,**draw_params)
plt.imshow(img3, 'gray'),plt.show()
I am working on a project which requires me to stitch images together. I decided to test this with buildings due to a large number of possible key points that can be calculated. I have been following several guides, but the one with the best results for 2-3 images has been this guide: https://towardsdatascience.com/image-stitching-using-opencv-817779c86a83. The way I decided to stitch multiple images is to stitch the first two, then take the output and then stitch that with the third image, so on and so forth. I am confident in the matching of descriptors for the images. But as I stitch more and more images, the previous stitched part gets pushed further and further into -z axis. Meaning they get distorted and smaller. The code I use to accomplish this is as follows:
import cv2
import numpy as np
import os
os.chdir('images')
img_ = cv2.imread('Output.jpg', cv2.COLOR_BGR2GRAY)
img = cv2.imread('DJI_0019.jpg', cv2.COLOR_BGR2GRAY)
#Setting up orb key point detector
orb = cv2.ORB_create()
#using orb to compute keypoints and descriptors
kp, des = orb.detectAndCompute(img_, None)
kp2, des2 = orb.detectAndCompute(img, None)
print(len(kp))
#Setting up BFmatcher
bf = cv2.BFMatcher(cv2.NORM_HAMMING, crossCheck=False)
matches = bf.knnMatch(des, des2, k=2) #Find 2 best matches for each descriptors (This is required for ratio test?)
#Using lowes ratio test as suggested in paper at .7-.8
good = []
for m in matches:
if m[0].distance < .8 * m[1].distance:
good.append(m)
matches = np.asarray(good) #matches is essentially a list of matching descriptors
#Aligning the images
if(len(matches)) >= 4:
src = np.float32([kp[m.queryIdx].pt for m in matches[:, 0]]).reshape(-1, 1, 2)
dst = np.float32([kp2[m.trainIdx].pt for m in matches[:, 0]]).reshape(-1, 1, 2)
#Creating the homography and mask
H, masked = cv2.findHomography(src, dst, cv2.RANSAC, 5.0)
print(H)
else:
print("Could not find 4 good matches to find homography")
dst = cv2.warpPerspective(img_, H, (img.shape[1] + 900, img.shape[0]))
dst[0:img.shape[0], 0:img.shape[1]] = img
cv2.imwrite("Output.jpg", dst)
With the output of the 4th+ stitch looking like such:
As you can see the images are getting further and further transformed in a weird way. My theory for such an event happening is due to the camera position and angle at which the images were taken, but I am not sure. If this might be the case, are there optimal parameters that will produce the best images to stitching?
Is there a way to fix this issue where the content can be pushed "flush" against the x axis?
Edit: Adding source images: https://imgur.com/zycPQuV
I need to detect the vein junctions of wings bee (the image is just one example). I use opencv - python.
ps: maybe the image lost a little bit of quality, but the image is all connected with one pixel wide.
This is an interesting question. The result I got is not perfect, but it might be a good start. I filtered the image with a kernel that only looks at the edges of the kernel. The idea being, that a junction has at least 3 lines that cross the kernel-edge, where regular lines only have 2. This means that when the kernel is over a junction, the resulting value will be higher, so a threshold will reveal them.
Due to the nature of the lines there are some value positives and some false negatives. A single joint will most likely be found several times, so you'll have to account for that. You can make them unique by drawing small dots and detecting those dots.
Result:
Code:
import cv2
import numpy as np
# load the image as grayscale
img = cv2.imread('xqXid.png',0)
# make a copy to display result
im_or = img.copy()
# convert image to larger datatyoe
img.astype(np.int32)
# create kernel
kernel = np.ones((7,7))
kernel[2:5,2:5] = 0
print(kernel)
#apply kernel
res = cv2.filter2D(img,3,kernel)
# filter results
loc = np.where(res > 2800)
print(len(loc[0]))
#draw circles on found locations
for x in range(len(loc[0])):
cv2.circle(im_or,(loc[1][x],loc[0][x]),10,(127),5)
#display result
cv2.imshow('Result',im_or)
cv2.waitKey(0)
cv2.destroyAllWindows()
Note: you can try to tweak the kernel and the threshold. For example, with the code above I got 126 matches. But when I use
kernel = np.ones((5,5))
kernel[1:4,1:4] = 0
with threshold
loc = np.where(res > 1550)
I got 33 matches in these locations:
You can use Harris corner detector algorithm to detect vein junction in above image. Compared to the previous techniques, Harris corner detector takes the differential of the corner score into account with reference to direction directly, instead of using shifting patches for every 45 degree angles, and has been proved to be more accurate in distinguishing between edges and corners (Source: wikipedia).
code:
img = cv2.imread('wings-bee.png')
# convert image to grayscale
gray = cv2.cvtColor(img, cv2.COLOR_BGR2GRAY)
gray = np.float32(gray)
'''
args:
img - Input image, it should be grayscale and float32 type.
blockSize - It is the size of neighbourhood considered for corner detection
ksize - Aperture parameter of Sobel derivative used.
k - Harris detector free parameter in the equation.
'''
dst = cv2.cornerHarris(gray, 9, 5, 0.04)
# result is dilated for marking the corners
dst = cv2.dilate(dst,None)
# Threshold for an optimal value, it may vary depending on the image.
img_thresh = cv2.threshold(dst, 0.32*dst.max(), 255, 0)[1]
img_thresh = np.uint8(img_thresh)
# get the matrix with the x and y locations of each centroid
centroids = cv2.connectedComponentsWithStats(img_thresh)[3]
stop_criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 30, 0.001)
# refine corner coordinates to subpixel accuracy
corners = cv2.cornerSubPix(gray, np.float32(centroids), (5,5), (-1,-1), stop_criteria)
for i in range(1, len(corners)):
#print(corners[i])
cv2.circle(img, (int(corners[i,0]), int(corners[i,1])), 5, (0,255,0), 2)
cv2.imshow('img', img)
cv2.waitKey(0)
cv2.destroyAllWindows()
output:
You can check the theory behind Harris Corner detector algorithm from here.
I want to get the image Difference for the print which is captured using camera.
I tried many solution using python libraries: opencv, image-magic, etc.
The solution I found for image comparison is for better accuracy is:
move image : left to right and look for minimum difference.
move image : right to left and look for minimum difference.
move image : top to bottom and look for minimum difference.
move image : bottom to top and look for minimum difference.
Condition to capture Image :
1. camera will never move (mounted over a fix stand).
2. Object is placed manually over a white sheet, thus the object will never be properly aligned. (slight variation in angle every time, as it is manual )
Image Sample captured using camera for the bellow code :
Image sample 1: white Dots :
Image sample 2: as original image
Image sample 3: black dots
Accepted Output for print with white dots is not available, but it should only mark the difference(defect) :
Currently I am using following Image-magic command for image difference:
compare -highlight-color black -fuzz 5% -metric AE Image_1.png Image_2.png -compose src diff.png
Code :
import subprocess
# -fuzz 5% # ignore minor difference between two images
cmd = 'compare -highlight-color black -fuzz 5% -metric AE Input.png output.png -compose src diff.png '
subprocess.call(cmd, shell=True)
Output after difference is incorrect as the comparison works pixel to pixel, it is not smart enough to mark only the real difference:
The above solution which I mention will work to get required difference as output, but there is no library or image-magic command available for such image comparison.
Any python code OR Image-magic command for doing this?
It seems you are doing some defect detection task. The first solution comes in my mind is the image registration technique.
First try to take the images in the same conditions (lighting, camera angle and ...) (one of your provided images is bigger 2 pixels).
Then you should register two images and match one to the other one, like this
Then wrap them with the help of homography matrix, and generate an aligned image, in this case, the result is like this:
Then take the difference of aligned image with the query image and threshold it, the result:
As I said if you try to take your frames with more precise, the registration result will be better and cause more accurate performance.
The codes for each part: (mostly taken from here).
import cv2
import numpy as np
MAX_FEATURES = 1000
GOOD_MATCH_PERCENT = 0.5
def alignImages(im1, im2):
# Convert images to grayscale
im1Gray = cv2.cvtColor(im1, cv2.COLOR_BGR2GRAY)
im2Gray = cv2.cvtColor(im2, cv2.COLOR_BGR2GRAY)
# Detect ORB features and compute descriptors.
orb = cv2.ORB_create(MAX_FEATURES)
keypoints1, descriptors1 = orb.detectAndCompute(im1Gray, None)
keypoints2, descriptors2 = orb.detectAndCompute(im2Gray, None)
# Match features.
matcher = cv2.DescriptorMatcher_create(cv2.DESCRIPTOR_MATCHER_BRUTEFORCE_HAMMING)
matches = matcher.match(descriptors1, descriptors2, None)
# Sort matches by score
matches.sort(key=lambda x: x.distance, reverse=False)
# Remove not so good matches
numGoodMatches = int(len(matches) * GOOD_MATCH_PERCENT)
matches = matches[:numGoodMatches]
# Draw top matches
imMatches = cv2.drawMatches(im1, keypoints1, im2, keypoints2, matches, None)
cv2.imwrite("matches.jpg", imMatches)
# Extract location of good matches
points1 = np.zeros((len(matches), 2), dtype=np.float32)
points2 = np.zeros((len(matches), 2), dtype=np.float32)
for i, match in enumerate(matches):
points1[i, :] = keypoints1[match.queryIdx].pt
points2[i, :] = keypoints2[match.trainIdx].pt
# Find homography
h, mask = cv2.findHomography(points1, points2, cv2.RANSAC)
# Use homography
height, width, channels = im2.shape
im1Reg = cv2.warpPerspective(im1, h, (width, height))
return im1Reg
if __name__ == '__main__':
# Read reference image
refFilename = "vv9gFl.jpg"
imFilename = "uP3CYl.jpg"
imReference = cv2.imread(refFilename, cv2.IMREAD_COLOR)
im = cv2.imread(imFilename, cv2.IMREAD_COLOR)
# Registered image will be resotred in imReg.
# The estimated homography will be stored in h.
imReg = alignImages(im, imReference)
# Write aligned image to disk.
outFilename = "aligned.jpg"
cv2.imwrite(outFilename, imReg)
for image difference and thresholding:
alined = cv2.imread("aligned.jpg" , 0)
alined = alined[:, :280]
b = cv2.imread("vv9gFl.jpg", 0 )
b = b[:, :280]
print (alined.shape)
print (b.shape)
diff = cv2.absdiff(alined, b)
cv2.imwrite("diff.png", diff)
threshold = 25
alined[np.where(diff > threshold)] = 255
alined[np.where(diff <= threshold)] = 0
cv2.imwrite("threshold.png", diff)
If you have lots of images and want to do defect detecting task I suggest using Denoising Autoencoder to train a deep artificial neural network. Read more here.
Although you do not want point-by-point processing, here is a subimage-search compare using Imagemagick. It pads one image after cropping off the black and then shifts the smaller to find the best match locations with the larger.
crop image1:
convert image1.jpg -gravity north -chop 0x25 image1c.png
crop and pad image2:
convert image2.jpg -gravity north -chop 0x25 -gravity center -bordercolor "rgb(114,151,157)" -border 20x20 image2c.png
do subimage search
compare -metric rmse -subimage-search image2c.png image1c.png null:
1243.41 (0.0189732) # 22,20
now shift and get difference between the two images
convert image2c.png image1c.png -geometry +22+20 -compose difference -composite -shave 22x20 -colorspace gray -auto-level +level-colors white,red diff.png
ADDITION:
If you want to just use compare, then you need to add -fuzz 15% to the compare command:
compare -metric rmse -fuzz 15% -subimage-search image2c.png image1c.png diff.png
Two images are produced. The difference image is the first, so look at diff-0.png
i've been trying to match a scanned formular with its empty template. The goal is to rotate and scale it to match the template.
Source (left), template (right)
Match (left), Homography warp (right)
The template does not contain any very specific logo, fixation cross or rectangular frame that would conveniently help me with feature or pattern matching. Even worse, the scanned formular can be skewed, altered and contains handwritten signatures and stamps.
My approach, after unsuccessfully testing ORB feature matching, was to concentrate on the shape of the formular (lines and column).
The pictures I provide here are obtained by reconstituting lines after a segment detection (LSD) with a certain minimum size. Most of what remains for source and template is the document layout itself.
In the following script (that should work out of the box along with pictures), I attempt to do ORB feature matching, but fail to make it work because it is concentrating on edges and not on the document layout.
import cv2 # using opencv-python v3.4
import numpy as np
from imutils import resize
# alining image using ORB descriptors, then homography warp
def align_images(im1, im2,MAX_MATCHES=5000,GOOD_MATCH_PERCENT = 0.15):
# Detect ORB features and compute descriptors.
orb = cv2.ORB_create(MAX_MATCHES)
keypoints1, descriptors1 = orb.detectAndCompute(im1, None)
keypoints2, descriptors2 = orb.detectAndCompute(im2, None)
# Match features.
matcher = cv2.DescriptorMatcher_create(cv2.DESCRIPTOR_MATCHER_BRUTEFORCE_HAMMING)
matches = matcher.match(descriptors1, descriptors2, None)
# Sort matches by score
matches.sort(key=lambda x: x.distance, reverse=False)
# Remove not so good matches
numGoodMatches = int(len(matches) * GOOD_MATCH_PERCENT)
matches = matches[:numGoodMatches]
# Draw top matches
imMatches = cv2.drawMatches(im1, keypoints1, im2, keypoints2, matches, None)
# Extract location of good matches
points1 = np.zeros((len(matches), 2), dtype=np.float32)
points2 = np.zeros((len(matches), 2), dtype=np.float32)
for i, match in enumerate(matches):
points1[i, :] = keypoints1[match.queryIdx].pt
points2[i, :] = keypoints2[match.trainIdx].pt
# Find homography
h, mask = cv2.findHomography(points1, points2, cv2.RANSAC)
# Use homography
if len(im2.shape) == 2:
height, width = im2.shape
else:
height, width, channels = im2.shape
im1Reg = cv2.warpPerspective(im1, h, (width, height))
return im1Reg, h, imMatches
template_fn = './stack/template.jpg'
image_fn = './stack/image.jpg'
im = cv2.imread(image_fn, cv2.IMREAD_GRAYSCALE)
template = cv2.imread(template_fn, cv2.IMREAD_GRAYSCALE)
# aligh images
imReg, h, matches = align_images(template,im)
# display output
cv2.imshow('im',im)
cv2.imshow('template',template)
cv2.imshow('matches',matches)
cv2.imshow('result',imReg)
cv2.waitKey(0)
cv2.destroyAllWindows()
Is there any way to make the pattern matching algorithm work on the image on the left (source)? (another idea was to leave only lines intersections)
Alternatively, I have been trying to do scale and rotation invariant pattern matching for loops and while keeping max correlation, but it is way too resource consuming and not very reliable.
I'm therefore looking for hints in the right direction using opencv.
SOLUTION
The issue was about reducing the image to what really matters: the layout.
Also, ORB was not appropriate since it is not as robust (rotation and size invariant) as SIFT and AKAZE are.
I proceeded as follows:
convert the images to black and white
use line segment detection and filter lines shorter than 1/60th of the width
reconstruct the image from segments (line width does not have a big impact)
(optional: resize the pictures to speed up the rest)
apply a Gaussian transformation on the line reconstruction, 1/25th of the width
detect and match features using SIFT (patented) or AKAZE (free) algorithm
find a homography and warp the source picture to match the template
Matches for AKAZE
Matches for SIFT
I noted:
the layout of the template has to match, otherwise it will only stick to what it recognizes
line detection is better with higher resolution, then downsizing is possible since Gaussian are applied
SIFT produces more features and seems more reliable than AKAZE