Given two grayscale images represented by two matrices where each cell indicates the pixel value (shown in the picture). How can I determine whether the two images are the same or not? On looking at the grayscale values, it is clear that the second image is the rotated version of the first one. But is there a mathematical proof or way to show it?
My approach: We can create histograms for both the images and match them. If they don't match, then the images are not the same. But the problem is that histograms don't show the spatial characteristics of the images. There may be two images with the same histograms but are actually totally different from each other.
Two compare 2 images of the same orientation, use this:
diff = np.subtract(img1, img2)
if np.sum(diff != 0) > 0:
print("Images are not same")
else:
print("Images are same")
Now flip one of the images once wrt horizontal axis, and then wrt verticle axis(and also transpose the image once(if that is a case)) and compare these images one by one with the second image. If any of these images are the same, then the 2 images are same else they are not the same.
Instead of flipping, you can use rotate(keep in mind that you rotate by 180 degrees).
See this to know how to flip and rotate
Related
I'm trying to show students how the RGB color model works to create a particular color (or moreover to convince them that it really does). So I want to take a picture and convert each pixel to an RGB representation so that when you zoom in, instead of a single colored pixel, you see the RGB colors.
I've done this but for some very obvious reasons the converted picture is either washed out or darker than the original (which is a minor inconvenience but I think it would be more powerful if I could get it to be more like the original).
Here are two pictures "zoomed out":
Here is a "medium zoom", starting to show the RGB artifacts in the converted picture:
And here is a picture zoomed in to the point that you can clearly see individual pixels and the RGB squares:
You'll notice the constant color surrounding the pixels; that is the average RGB of the picture. I put that there so that you could see individual pixels (otherwise you just see rows/columns of shades of red/green/blue). If I take that space out completely, the image is even darker and if I replace it with white, then the image looks faded (when zoomed out).
I know why displaying this way causes it to be darker: a "pure red" will come with a completely black blue and green. In a sense if I were to take a completely red picture, it would essentially be 1/3 the brightness of the original.
So my question is:
1: Are there any tools available that already do this (or something similar)?
2: Any ideas on how to get the converted image closer to the original?
For the 2nd question, I could of course just increase the brightness for each "RGB pixel" (the three horizontal stripes in each square), but by how much? I certainly can't just multiply the RGB ints by 3 (in apparent compensation for what I said above). I wonder if there is some way to adjust my background color to compensate for me? Or would it just have to be something that needs to be fiddled with for each picture?
You were correct to assume you could retain the brightness by multiplying everything by 3. There's just one small problem: the RGB values in an image use gamma correction, so the intensity is not linear. You need to de-gamma the values, multiply, then gamma correct them again.
You also need to lose the borders around each pixel. Those borders take up 7/16 of the final image which is just too much to compensate for. I tried rotating every other pixel by 90 degrees, and while it gives the result a definite zig-zag pattern it does make clear where the pixel boundaries are.
When you zoom out in an image viewer you might see the gamma problem too. Many viewers don't bother to do gamma correction when they resize. For an in-depth explanation see Gamma error in picture scaling, and use the test image supplied at the end. It might be better to forgo scaling altogether and simply step back from the monitor.
Here's some Python code and a crop from the resulting image.
from PIL import Image
im = Image.open(filename)
im2 = Image.new('RGB', (im.size[0]*3, im.size[1]*3))
ld1 = im.load()
ld2 = im2.load()
for y in range(im.size[1]):
for x in range(im.size[0]):
rgb = ld1[x,y]
rgb = [(c/255)**2.2 for c in rgb]
rgb = [min(1.0,c*3) for c in rgb]
rgb = tuple(int(255*(c**(1/2.2))) for c in rgb)
x2 = x*3
y2 = y*3
if (x+y) & 1:
for x3 in range(x2, x2+3):
ld2[x3,y2] = (rgb[0],0,0)
ld2[x3,y2+1] = (0,rgb[1],0)
ld2[x3,y2+2] = (0,0,rgb[2])
else:
for y3 in range(y2, y2+3):
ld2[x2,y3] = (rgb[0],0,0)
ld2[x2+1,y3] = (0,rgb[1],0)
ld2[x2+2,y3] = (0,0,rgb[2])
Don't waste so much time on this. You cannot make two images look the same if you have less information in one of them. You still have your computer that will subsample your image in weird ways while zooming out.
Just pass a magnifying glass through the class so they can see for themselves on their phones or other screens or show pictures of a screen in different magnification levels.
If you want to stick to software, triple the resolution of your image, don't use empty rows and columns or at least make them black to increase contrast and scale the RGB components to full range.
Why don't you keep the magnified image for the background ? This will let the two images look identical when zoomed out, while the RGB strips will remain clearly visible in the zoom-in.
If not, use the average color over the whole image to keep a similar intensity, but the washing effect will remain.
An intermediate option is to apply a strong lowpass filter on the image to smoothen all details and use that as the background, but I don't see a real advantage over the first approach.
Light Field captures the scene from slightly different points. This means I would have two images of the same scene with a slight shift, as shown in the following figure:
Assuming the red squares in the images above are pixels. I know that the spatial difference between those two pixels is a shift. Nevertheless, what other information do these two pixels give us in terms of scene radiance? I mean is there a way to find (or compute) the difference in image irradiance values between those two points?
Look for color space representations other than RGB. Some of them have explicit channel(s) carrying luminance information of a pixel.
A varaiant of the same idea is to convert to a Black and White image and examine the pixel values.
I have a customized camera, which contains 3 individual lens+filters arranged in a triangle so in every shot I get 3 single band grayscale images (r, g, b). I want to merge them to get an RGB.
The problem is, since the 3 lens are physically separated, the image captured by them are not aligned. As a result, when I use command qdal_merge in the software pack QGIS, the result looks weird. I may also need to adjust the weight of the r,g,b. I put the raw r,g,b images and the output I generated using qgis in this dropbox folder.
Is there existing open-source tool to do the alignment and merge? If not, how can I do it using opencv?
Combining R,G,B images is possible using a simple pixel intensity distance metric like Sum of Squared Distances (SSD). A better metric is the Normalized Cross-Correlation (NCC) (see Wikipedia) which first normalizes an image matrix into a unit vector, and computes the dot product of such unit vectors (from 2 input images). The higher the NCC value, the greater the similarity of the two input images.
However, NCC similarity may be insufficient for computing the best alignment of two high resolution images, such as the TIFF images you provide. One should therefore use a downsampling method as described below
to align two input images at a smaller size and then simply compute the offset as you rescale the images.
So for the input images, red, green and blue, there are two approaches to align them into a single RGB image:
Consider the blue image as the reference image for example, w.r.t. which we align the red and green images. Now consider red and blue images. Within a certain window, compute the best alignment offset of the red and blue images using the NCC similarity metric, and find the shifted_red image. Do the same for the green and blue images. Now combine the shifted_red, shifted_green and blue images to get the final RGB image.
For high-resolution images, decide a scale_count. Recursively, at each step resize the image by half, compute the offset of the red image w.r.t. the blue image, rescale the offset and apply it. The benefit of doing such a recursive multi-scale alignment is decrease in computation time and increase in accuracy of alignment (you don't know the best window size for searching for alignment offsets for solution (1), so this will work better). Repeat this approach for computing the alignment for green and blue channels, and then combine the final results as in (1).
Since this problem is common in assignments of computational photography courses, I am not going to share any code. I have, however implemented the two approaches and experimented with the images you provide. I don't know which of the input images is red, so I have two results (rescaled to decrease file size):
If IMG_0290_1.tif is Red, IMG_0290_2.tif is Green and IMG_0290_3.tif is blue:
RGB result if red:1, green:2, blue:3
If IMG_0290_3.tif is Red, IMG_0290_2.tif is Green and IMG_0290_1.tif is blue (this looks more correct to me):
RGB result if red:3, green:2, blue:1
Here's what I would like to do:
I have an image of rice leaf. I have another image of rice leaf that got brown color spots on the leaf. What I want to do is separate the color pixels that are not common for two images using opencv.(color of the spots can be vary)
I tried to do this using histogram intersection. But only managed to find number of pixels common between two images.
Is there any way to do this using opencv? Please be me kind enough to help me.
if the 2 images match perfectly
if they match use RhinoDevel approach:
so loop through all pixels of first image
and compare each pixel with corresponding pixel from second image
if the difference is higher then treshold
you found not matching Pixel and do what you need to do
like add pixel to some output map or recolor (brown) pixel to color from first image or whatever
if the 2 images does not match
so you just got some reference leaf image and the processed image can have any position/rotation skew
create list of colors per each image
sort them ascending by color
cross compare booth lists
if any color is in list2 but not in list1
then recolor/copy all pixels that contains such color in/from image2
this approach is slower O(xs*ys*n)
xs,ys is image 2 resolution and n is number of non common colors
[Notes]
RGB is usually fine but may be you got better result on HSV color space
in HSV you can compare all 3 parameters or just few of them like ignoring V value...
I'd like to know what would be the best strategy to compare a group of contours, in fact are edges resulting of a canny edges detection, from two pictures, in order to know which pair is more alike.
I have this image:
http://i55.tinypic.com/10fe1y8.jpg
And I would like to know how can I calculate which one of these fits best to it:
http://i56.tinypic.com/zmxd13.jpg
(it should be the one on the right)
Is there anyway to compare the contours as a whole?
I can easily rotate the images but I don't know what functions to use in order to calculate that the reference image on the right is the best fit.
Here it is what I've already tried using opencv:
matchShapes function - I tried this function using 2 gray scales images and I always get the same result in every comparison image and the value seems wrong as it is 0,0002.
So what I realized about matchShapes, but I'm not sure it's the correct assumption, is that the function works with pairs of contours and not full images. Now this is a problem because although I have the contours of the images I want to compare, they are hundreds and I don't know which ones should be "paired up".
So I also tried to compare all the contours of the first image against the other two with a for iteration but I might be comparing,for example, the contour of the 5 against the circle contour of the two reference images and not the 2 contour.
Also tried simple cv::compare function and matchTemplate, none with success.
Well, for this you have a couple of options depending on how robust you need your approach to be.
Simple Solutions (with assumptions):
For these methods, I'm assuming your the images you supplied are what you are working with (i.e., the objects are already segmented and approximately the same scale. Also, you will need to correct the rotation (at least in a coarse manner). You might do something like iteratively rotate the comparison image every 10, 30, 60, or 90 degrees, or whatever coarseness you feel you can get away with.
For example,
for(degrees = 10; degrees < 360; degrees += 10)
coinRot = rotate(compareCoin, degrees)
// you could also try Cosine Similarity, or even matchedTemplate here.
metric = SAD(coinRot, targetCoin)
if(metric > bestMetric)
bestMetric = metric
coinRotation = degrees
Sum of Absolute Differences (SAD): This will allow you to quickly compare the images once you have determined an approximate rotation angle.
Cosine Similarity: This operates a bit differently by treating the image as a 1D vector, and then computes the the high-dimensional angle between the two vectors. The better the match the smaller the angle will be.
Complex Solutions (possibly more robust):
These solutions will be more complex to implement, but will probably yield more robust classifications.
Haussdorf Distance: This answer will give you an introduction on using this method. This solution will probably also need the rotation correction to work properly.
Fourier-Mellin Transform: This method is an extension of Phase Correlation, which can extract the rotation, scale, and translation (RST) transform between two images.
Feature Detection and Extraction: This method involves detecting "robust" (i.e., scale and/or rotation invariant) features in the image and comparing them against a set of target features with RANSAC, LMedS, or simple least squares. OpenCV has a couple of samples using this technique in matcher_simple.cpp and matching_to_many_images.cpp. NOTE: With this method you will probably not want to binarize the image, so there are more detectable features available.