I use the 2D-FFT from NumPy to calculate the differential phase of a patterned image. So I have an image with 20x20 spots which shift within several images and I want to get the shift / differential phase.
Therefore I use the following:
picfft = np.fft.fft2(data* hanning_window)
picfft_shifted = np.fft.fftshift(picfft)
Now I want to crop the different parts of the higher 1st order harmonics. From the documentation of NumPy I read that, before shifting, the zero frequency is in the low-order corner and the positive frequencies are in the first half of the dimensions. After the fftshift this leads to the fact that the zero frequency is in the center and the positive frequencies in the left bottom quarter.
If I now crop the 1st order areas as in the image and transform it back I get a quite reliable differential phase. However the sign is changed. I know that my spots shift to the middle of the image, but in the (1,0) and (0,1) orders the signs indicate a shift to the outer area of the image.
Have I interchanged the directions of crop, so is my (1,0) order in reality the (-1,0) order and so on? Because with this the results would fit to reality. But then my understanding of the documentation is different.
Hopefully anyone here is familiar with 2D-FFT and phase information.
You have a description of what you do, rather than a demonstration (it’s always better to show code), but I think I know what might be amiss.
From your description it seems that you are cropping each of the modes of the frequency domain, and inverse transforming them. The crop has the origin (zero frequency) in the middle. It is imperative that you apply ifftshift to these crops before the inverse transform, such that the origin be moved to the top-left corner. Otherwise the phase of the inverse transform will be wrong.
Related
I have a webcam looking down on a surface which rotates about a single-axis. I'd like to be able to measure the rotation angle of the surface.
The camera position and the rotation axis of the surface are both fixed. The surface is a distinct solid color right now, but I do have the option to draw features on the surface if it would help.
Here's an animation of the surface moving through its full range, showing the different apparent shapes:
My approach thus far:
Record a series of "calibration" images, where the surface is at a known angle in each image
Threshold each image to isolate the surface.
Find the four corners with cv2.approxPolyDP(). I iterate through various epsilon values until I find one that yields exactly 4 points.
Order the points consistently (top-left, top-right, bottom-right, bottom-left)
Compute the angles between each points with atan2.
Use the angles to fit a sklearn linear_model.linearRegression()
This approach is getting me predictions within about 10% of actual with only 3 training images (covering full positive, full negative, and middle position). I'm pretty new to both opencv and sklearn; is there anything I should consider doing differently to improve the accuracy of my predictions? (Probably increasing the number of training images is a big one??)
I did experiment with cv2.moments directly as my model features, and then some values derived from the moments, but these did not perform as well as the angles. I also tried using a RidgeCV model, but it seemed to perform about the same as the linear model.
If I'm clear, you want to estimate the Rotation of the polygon with respect to the camera. If you know the length of the object in 3D, you can use solvePnP to estimate the pose of the object, from which you can get the Rotation of the object.
Steps:
Calibrate your webcam and get the intrinsic matrix and distortion matrix.
Get the 3D measurements of the object corners and find the corresponding points in 2d. Let me assume a rectangular planar object and the corners in 3d will be (0,0,0), (0, 100, 0), (100, 100, 0), (100, 0, 0).
Use solvePnP to get the rotation and translation of the object
The rotation will be the rotation of your object along the axis. Here you can find an example to estimate the pose of the head, you can modify it to suit your application
Your first step is good -- everything after that becomes way way way more complicated than necessary (if I understand correctly).
Don't think of it as 'learning,' just think of it as a reference. Every time you're in a particular position where you DON'T know the angle, take a picture, and find the reference picture that looks most like it. Guess it's THAT angle. You're done! (They may well be indeterminacies, maybe the relationship isn't bijective, but that's where I'd start.)
You can consider this a 'nearest-neighbor classifier,' if you want, but that's just to make it sound better. Measure a simple distance (Euclidean! Why not!) between the uncertain picture, and all the reference pictures -- meaning, between the raw image vectors, nothing fancy -- and choose the angle that corresponds to the minimum distance between observed, and known.
If this isn't working -- and maybe, do this anyway -- stop throwing away so much information! You're stripping things down, then trying to re-estimate them, propagating error all over the place for no obvious (to me) benefit. So when you do a nearest neighbor, reference pictures and all that, why not just use the full picture? (Maybe other elements will change in it? That's a more complicated question, but basically, throw away as little as possible -- it should all be useful in, later, accurately choosing your 'nearest neighbor.')
Another option that is rather easy to implement, especially since you've done a part of the job is the following (I've used it to compute the orientation of a cylindrical part from 3 images acquired when the tube was rotating) :
Threshold each image to isolate the surface.
Find the four corners with cv2.approxPolyDP(), alternatively you could find the four sides of your part with LineSegmentDetector (available from OpenCV 3).
Compute the angle alpha, as depicted on the image hereunder
When your part is rotating, this angle alpha will follow a sine curve. That is, you will measure alpha(theta) = A sin(theta + B) + C. Given alpha you want to know theta, but first you need to determine A, B and C.
You've acquired many "calibration" or reference images, you can use all of these to fit a sine curve and determine A, B and C.
Once this is done, you can determine theta from alpha.
Notice that you have to deal with sin(a+Pi/2) = sin(a). It is not a problem if you acquire more than one image sequentially, if you have a single static image, you have to use an extra mechanism.
Hope I'm clear enough, the implementation really shouldn't be a problem given what you have done already.
I have an Image (or several hundreds of them) that need to be analyzed. The goal is to find all black spots close to each other.
For example all black spots with a Horizontal distance of 160 pixel and vertical 40 pixel.
For now I just look at each Pixel and if there is a black pixel I call a recursive Method to find its neighbours (i can post the code too if you want to)
It works, but its very slow. At the moment the script runs about 3-4 minutes depending on image size.
Is there some easy/fast way to accomplish this (best would be a scikit-image method to help out here) I'm using Python.
edit: I tried to use scikit.measure.find_contours, now i have an array with arrays containing the contours of the black spots. Now I only need to find the contours in the neighbourhood of these contours.
When you get the coordinates of the different black spots, rather than computing all distances between all pairs of black pixels, you can use a cKDTree (in scipy.spatial, http://docs.scipy.org/doc/scipy/reference/generated/scipy.spatial.cKDTree.html#scipy.spatial.cKDTree). The exact method of cKDTree to use depends on your exact criterion (you can for example use cKDTree.query_ball_tree to know whether there exists a pair of points belonging to two different labels, with a maximal distance that you give).
KDTrees are a great method to reduce the complexity of problems based on neighboring points. If you want to use KDTrees, you'll need to rescale the coordinates so that you can use one of the classical norms to compute the distance between points.
Disclaimer: I'm not proficient with the scikit image library at all, but I've tackled similar problems using MATLAB so I've searched for the equivalent methods in scikit, and I hope my findings below help you.
First you can use skimage.measure.label which returns label_image, i.e. an image where all connected regions are labelled with the same number. I believe you should call this function with background=255 because from your description it seems that the background in your images is the while region (hence the value 255).
This is essentially an image where the background pixels are assigned the value 0 and the pixels that make up each (connected) spot are assigned the value of an integer label, so all the pixels of one spot will be labelled with the value 1, the pixels of another spot will be labelled with the value 2, and so on. Below I'll refer to "spots" and "labelled regions" interchangeably.
You can then call skimage.measure.regionprops, that takes as input the label_image obtained in the previous step. This function returns a list of RegionProperties (one for each labelled region), which is a summary of properties of a labelled region.
Depending on your definition of
The goal is to find all black spots close to each other.
there are different fields of the RegionProperties that you can use to help solve your problem:
bbox gives you the set of coordinates of the bounding box that contains that labelled region,
centroid gives you the coordinates of the centroid pixel of that labelled region,
local_centroid gives you the centroid relative to the bounding box bbox
(Note there are also area and bbox_area properties which you can use to decide whether to throw away very small spots that you might not be interested in, thus reducing computation time when it comes to comparing proximity of each pair of spots)
If you're looking for a coarse comparison, then comparing the centroid or local_centroid between each pair of labelled regions might be enough.
Otherwise you can use the bbox coordinates to measure the exact distance between the outer bounds of any two regions.
If you want to base the decision on the precise distance between the pixel(s) of each pair of regions that are closest to each other, then you'll likely have to use the coords property.
If your input image is binary, you could separate your regions of interest as follows:
"grow" all the regions by the expected distance (actually half of it, as you grow from "both sides of the gap") with binary_dilation, where the structure is a kernel (e.g. rectangular: http://scikit-image.org/docs/dev/api/skimage.morphology.html#skimage.morphology.rectangle) of, let's say, 20x80pixels;
use the resulting mask as an input to skimage.measure.label to assign different values for different regions' pixels;
multiply your input image by the mask created above to zero dilated pixels.
Here are the results of proposed method on your image and kernel = rectange(5,5):
Dilated binary image (output of step 1):
Labeled version of the above (output of step 2):
Multiplication results (output of step 3):
I want to segment defective areas in images using MATLAB/Python-OpenCV.
Original image:
With Defects:
http://imgur.com/fyDkpcZ
Defect can be seen at 3rd rectangle.
What I tried so far:
Extract borders of rectangles with LoG filter / threshold graylevel (but not helps much because of shadows)
Trace their boundaries
Get centroid
Find distance between boundary points and centroid with respect to angle ( increment angle by 0.5 degrees for better resolution)
Find a good template rectangle and save it
Find the difference between template rectangle and candidate rectangle
Based on that result I can find the faulty regions but the false-alarm rate increases when I try to increase the sensitivity of algorithm.
I need to get boundaries much more precise and non-noisy. Because of the shadows, the edges of rectangle may vary vastly.
How can I get edges of rectangles more robust to shadows?
What can be done instead of what I tried so far?
Thanks for your help!
A Laplace of Gaussian filter is a zero mean operation. If you feed it an 8-bit image with intensities centered on 127, it will return you image data centered on zero. You must use a filter bias of arbitrary value, usually half the container's max value (so in this 8-bit example, the bias would be 127). You can also adjust the filter strength by multiplying the result pixels by a constant, this makes the log filter's effect more apparent.
The log filter will make one white and one black edge for very strong transitions. In the horizontal or vertical direction, finding the actual edge is very easy, as you need only take the average of position of both. This gives you sub-pixel resolution if integrated over a small distance.
If the illumination of these images is very similar, you can use registration and subtraction:
Normalize both the image suspected to contain defects and a reference image to some intensity.
Register (align) them; you could do this by detecting three points on a rectangle and then moving and rotating one of the images.
Subtract the suspect image from the reference image. This gives you an error map. You can apply a small blur and then a tight LoG filter to it to remove noise and make detection more accurate.
The image below shows two circles of same radius, rendered with antialiasing, only that the left circle is shifted half pixel horizontally (notice that the circle horizontal center is at the middle of a pixel at the left, and at the pixel border at the right).
If I perform a cross-correlation, I can take the position of the maximum on the correlation array, and then calculate the shift. But since pixel positions are always integers, my question is:
"How can I obtain a sub-pixel (floating point) offset between two images using cross-correlation in Numpy/Scipy?"
In my scripts, am using either of scipy.signal.correlate2d or scipy.ndimage.filters.correlate, and they seem to produce identical results.
The circles here are just examples, but my domain-specific features tend to have sub-pixel shifts, and currently getting only integer shifts is giving results that are not so good...
Any help will be much appreciated!
The discrete cross-correlation (implemented by those) can only have a single pixel precision. The only solution I can see is to interpolate your 2D arrays to a finer grid (up-sampling).
Here's some discussion on DSP about upsampling before cross-correlation.
I had a very similar issue, also with shifted circles, and stumbled upon a great Python package called 'image registration' by Adam Ginsburg. It gives you sub-pixel 2D images shifts and is fairly fast. I believe it's a Python implementation of a popular MATLAB module, which only upsamples images around the peak of the x-correlation.
Check it out: https://github.com/keflavich/image_registration
I've been using 'chi2_shifts.py' with good results.
My software should judge spectrum bands, and given the location of the bands, find the peak point and width of the bands.
I learned to take the projection of the image and to find width of each peak.
But I need a better way to find the projection.
The method I used reduces a 1600-pixel wide image (eg 1600X40) to a 1600-long sequence. Ideally I would want to reduce the image to a 10000-long sequence using the same image.
I want a longer sequence as 1600 points provide too low resolution. A single point causes a large difference (there is a 4% difference if a band is judged from 18 to 19) to the measure.
How do I get a longer projection from the same image?
Code I used: https://stackoverflow.com/a/9771560/604511
import Image
from scipy import *
from scipy.optimize import leastsq
# Load the picture with PIL, process if needed
pic = asarray(Image.open("band2.png"))
# Average the pixel values along vertical axis
pic_avg = pic.mean(axis=2)
projection = pic_avg.sum(axis=0)
# Set the min value to zero for a nice fit
projection /= projection.mean()
projection -= projection.min()
What you want to do is called interpolation. Scipy has an interpolate module, with a whole bunch of different functions for differing situations, take a look here, or specifically for images here.
Here is a recently asked question that has some example code, and a graph that shows what happens.
But it is really important to realise that interpolating will not make your data more accurate, so it will not help you in this situation.
If you want more accurate results, you need more accurate data. There is no other way. You need to start with a higher resolution image. (If you resample, or interpolate, you results will acually be less accurate!)
Update - as the question has changed
#Hooked has made a nice point. Another way to think about it is that instead of immediately averaging (which does throw away the variance in the data), you can produce 40 graphs (like your lower one in your posted image) from each horizontal row in your spectrum image, all these graphs are going to be pretty similar but with some variations in peak position, height and width. You should calculate the position, height, and width of each of these peaks in each of these 40 images, then combine this data (matching peaks across the 40 graphs), and use the appropriate variance as an error estimate (for peak position, height, and width), by using the central limit theorem. That way you can get the most out of your data. However, I believe this is assuming some independence between each of the rows in the spectrogram, which may or may not be the case?
I'd like to offer some more detail to #fraxel's answer (to long for a comment). He's right that you can't get any more information than what you put in, but I think it needs some elaboration...
You are projecting your data from 1600x40 -> 1600 which seems like you are throwing some data away. While technically correct, the whole point of a projection is to bring higher dimensional data to a lower dimension. This only makes sense if...
Your data can be adequately represented in the lower dimension. Correct me if I'm wrong, but it looks like your data is indeed one-dimensional, the vertical axis is a measure of the variability of that particular point on the x-axis (wavelength?).
Given that the projection makes sense, how can we best summarize the data for each particular wavelength point? In my previous answer, you can see I took the average for each point. In the absence of other information about the particular properties of the system, this is a reasonable first-order approximation.
You can keep more of the information if you like. Below I've plotted the variance along the y-axis. This tells me that your measurements have more variability when the signal is higher, and low variability when the signal is lower (which seems useful!):
What you need to do then, is decide what you are going to do with those extra 40 pixels of data before the projection. They mean something physically, and your job as a researcher is to interpret and project that data in a meaningful way!
The code to produce the image is below, the spec. data was taken from the screencap of your original post:
import Image
from scipy import *
from scipy.optimize import leastsq
# Load the picture with PIL, process if needed
pic = asarray(Image.open("spec2.png"))
# Average the pixel values along vertical axis
pic_avg = pic.mean(axis=2)
projection = pic_avg.sum(axis=0)
# Compute the variance
variance = pic_avg.var(axis=0)
from pylab import *
scale = 1/40.
X_val = range(projection.shape[0])
errorbar(X_val,projection*scale,yerr=variance*scale)
imshow(pic,origin='lower',alpha=.8)
axis('tight')
show()