I'm trying to measure the areas of each particle shown in this image:
I managed to get the general shape of each particle using MSER shown here:
but I'm having trouble removing the background. I tried using MATLAB's imfill, but it doesn't fill all the particles because some are cut off at the edges. Any tips on how to get rid of the background or find the areas of the particles some other way?
Cheers.
Edit: This is what imfill looks like:
Edit 2: Here is the code used to get the outline. I used this for the MSER.
%Compute region seeds and elliptial frames.
%MinDiversity = how similar to its parent MSER the region is
%MaxVariation = stability of the region
%BrightOnDark is used as the void is primarily dark. It also prevents dark
%patches in the void being detected.
[r,f] = vl_mser(I,'MinDiversity',0.7,...
'MaxVariation',0.2,...
'Delta',10,...
'BrightOnDark',1,'DarkOnBright',0) ;
%Plot region frames, but not used right now
%f = vl_ertr(f) ;
%vl_plotframe(f) ;
%Plot MSERs
M = zeros(size(I)) ; %M = no of overlapping extremal regions
for x=r'
s = vl_erfill(I,x) ;
M(s) = M(s) + 1;
end
%Display region boundaries
figure(1) ;
clf ; imagesc(I) ; hold on ; axis equal off; colormap gray ;
%Create contour plot using the values
%0:max(M(:))+.5 is the no of contour levels. Only level 0 is needed so
%[0 0] is used.
[c,h]=contour(M,[0 0]) ;;
set(h,'color','r','linewidth',1) ;
%Retrieve the image data from the contour image
f = getframe;
I2 = f.cdata;
%Convert the image into binary; the red outlines are while while the rest
%is black.
I2 = all(bsxfun(#eq,I2,reshape([255 0 0],[1 1 3])),3);
I2 = imcrop(I2,[20 1 395 343]);
imshow(~I2);
Proposed solution / trick and code
It seems you can work with M here. One trick that you can employ here would be to pad zeros all across the boundaries of the image M and then fill its holes. This would take care of filling the blobs that were touching the boundaries before, as now there won't be any blob touching the boundaries because of the zeros padding.
Thus, after you have M, you can add this code -
%// Get a binary version of M
M_bw = im2bw(M);
%// Pad zeros all across the grayscale image
padlen = 2; %// length of zeros padding
M_pad = padarray(M_bw,[padlen padlen],0);
%// Fill the holes
M_pad_filled = imfill(M_pad,'holes');
%// Get the background mask after the holes are gone
background_mask = ~M_pad_filled(padlen+1:end-padlen,padlen+1:end-padlen);
%// Overlay the background mask on the original image to show that you have
%// a working background mask for use
I(background_mask) = 0;
figure,imshow(I)
Results
Input image -
Foreground mask (this would be ~background_mask) -
Output image -
Related
I'm using the watershed algorithm to segment bright spots on a dark background. The code is provided below, along with some images it generates.
In the second image, I've marked with red crosses the areas of enclosed background which are segmented as 'cells' (they're not biological cells, just using the word) - this is incorrect, they're part of the background, just enclosed by 'cells'. I see that this creates a false minimum, any help on how to prevent this?
% Improve contrast, binarize
RFP_adjust = imadjust(RFP_blur, stretchlim(RFP_blur, 0.001));
figure, imshow(RFP_adjust), title('Contrast adjust');
RFP_binarized = imbinarize(RFP_adjust);
RFP_perimeters = bwperim(RFP_binarized);
% figure, imshow(RFP_binarized), title('Otsu thresholding');
%2B - SEGMENTATION BY WATERSHED METHOD
% Discover putative cell centroids and process
RFP_maxs = imextendedmax(RFP_adjust, 3000);
RFP_maxs = imclose(RFP_maxs, strel('disk',5));
RFP_maxs = imfill(RFP_maxs, 'holes');
RFP_maxs = bwareaopen(RFP_maxs, 5);
RFP_max_overlay = imoverlay(RFP_adjust, RFP_perimeters | RFP_maxs, [1 .3 .3]);
figure, imshow(RFP_max_overlay), title('Maxima');
% Obtain complement - maxima become low-points (required for watershed)
RFP_comp = imcomplement(RFP_adjust);
RFP_imposemin = imimposemin(RFP_comp, ~RFP_binarized | RFP_maxs);
figure, imshow(RFP_imposemin), title('Inverted Maxima');
% Apply watershed
RFP_watershed = watershed(RFP_imposemin);
mask = im2bw(RFP_watershed, 1);
overlay3 = imoverlay(RFP_adjust, mask, [1 .3 .3]);
figure, imshow(overlay3), title('Segmented cells');
% Segment
RFP_cc = bwconncomp(RFP_watershed);
RFP_label_matrix = labelmatrix(RFP_cc);
whos labeled;
RFP_label = label2rgb(RFP_label_matrix, #spring, 'c', 'shuffle');
figure, imshow(RFP_label), title('Cells segmented');
Image 0 - result for image titled 'Maxima' (i.e. adjusted original image with maxima and outlines overlaid).
Image 1 - the result for image titled 'inverted maxima'
Image 2 - the result for image titled 'Cells segmented'
I would suggest something like what is done in the example included for the watershed function: use the background mask to set those pixels to Inf, perform the watershed operation, then set the background pixels in the result to 0. I believe you could change the watershed section of your code like so to achieve this:
% Apply watershed
RFP_watershed = RFP_imposemin; % Added
RFP_watershed(~RFP_binarized) = Inf; % Added
RFP_watershed = watershed(RFP_watershed); % Modified
RFP_watershed(~RFP_binarized) = 0; % Added
mask = im2bw(RFP_watershed, 1);
overlay3 = imoverlay(RFP_adjust, mask, [1 .3 .3]);
figure, imshow(overlay3), title('Segmented cells');
There no magic bullet, but a few things you can try.
One is to filter the image with a very large circular disc, creating a blurry image that looks like the background. Then subtract it from the original image. That will tend to force the actual background to zero.
Another is to Otsu threshold to separate foreground from background. That creates a binary image. Then do a morphological open operation using a mask designed to look like the actual cells.
I am making a script in Matlab that takes in an image of the rear of a car. After some image processing I would like to output the original image of the car with a rectangle around the license plate of the car. Here is what I have written so far:
origImg = imread('CAR_IMAGE.jpg');
I = imresize(origImg, [500, NaN]); % easier viewing and edge connecting
G = rgb2gray(I);
M = imgaussfilt(G); % blur to remove some noise
E = edge(M, 'Canny', 0.4);
% I can assume all letters are somewhat upright
RP = regionprops(E, 'PixelIdxList', 'BoundingBox');
W = vertcat(RP.BoundingBox); W = W(:,3); % get the widths of the BBs
H = vertcat(RP.BoundingBox); H = H(:,4); % get the heights of the BBs
FATTIES = W > H; % find the BBs that are more wide than tall
RP = RP(FATTIES);
E(vertcat(RP.PixelIdxList)) = false; % remove more wide than tall regions
D = imdilate(E, strel('disk', 1)); % dilate for easier viewing
figure();
imshowpair(I, D, 'montage'); % display original image and processed image
Here are some examples:
From here I am unsure how to isolate the letters of the license plate, particularly like in the second example above where each letter has a decreased area due to the perspective of the image. My first idea was to get the bounding box of all regions and keep only the regions where the perimeter to area ratio is "similar" but this resulted in removing the letters of the plate that were connected when I dilate the image like the K and V in the fourth example above.
I would appreciate some suggestions on how I should go about isolating these letters. No code is necessary, and any advice is appreciated.
So I continued to work despite not receiving any answers here on SO and managed to get a working version through trial and error. All of the following code comes after the code in my original question and all plots below are from the first example image above. First, I found the variance for every single pixel row of the image and plotted them like so:
V = var(D, 0, 2);
X = 1:length(V);
figure();
hold on;
scatter(X, V);
I then fit a very high order polynomial to this scatter plot and saved the values where the slope of the polynomial was zero and the variance value was very low (i.e. the dark row of pixels immediately before or after a row with some white):
P = polyfit(X', V, 25);
PV = polyval(P, X);
Z = X(find(PV < 0.03 & abs(gradient(PV)) < 0.0001));
plot(X, PV); % red curve on plot
scatter(Z, zeros(1,length(Z))); % orange circles on x-axis
I then calculate the integral of the polynomial between any consecutive Z values (my dark rows), and save the two Z values between which the integral is the largest, which I mark with lines on the plot:
MAX_INTEG = -1;
MIN_ROW = -1;
MAX_ROW = -1;
for i = 1:(length(Z)-1)
TEMP_MIN = Z(i);
TEMP_MAX = Z(i+1);
Q = polyint(P);
TEMP_INTEG = diff(polyval(Q, [TEMP_MIN, TEMP_MAX]));
if (TEMP_INTEG > MAX_INTEG)
MAX_INTEG = TEMP_INTEG;
MIN_ROW = TEMP_MIN;
MAX_ROW = TEMP_MAX;
end
end
line([MIN_ROW, MIN_ROW], [-0.1, max(V)+0.1]);
line([MAX_ROW, MAX_ROW], [-0.1, max(V)+0.1]);
hold off;
Since the X-values of these lines correspond row numbers in the original image, I can crop my image between MIN_ROW and MAX_ROW:
I repeat the above steps now for the columns of pixels, crop, and remove any excess black rows of columns to result in the identified plate:
I then perform 2D cross correlation between this cropped image and the edged image D using Matlab's xcorr2 to locate the plate in the original image. After finding the location I just draw a rectangle around the discovered plate like so:
I'm using the watershed algorithm to try and segment touching nuclei. A typical image may look like:
or this:
I'm trying to apply the watershed algorithm with this code:
show(RGB_img)
%Convert to grayscale image
I = rgb2gray(RGB_img);
%Take structuring element of a disk of size 10, for the morphological transformations
%Attempt to subtract the background from the image: top hat is the
%subtraction of the open image from the original
%Morphological transformation to subtract background noise from the image
%Tophat is the subtraction of an opened image from the original. Remove all
%images smaller than the structuring element of 10
I1 = imtophat(I, strel('disk', 10));
%Increases contrast
I2 = imadjust(I1);
%show(I2,'contrast')
%Assume we have background and foreground and assess thresh as such
level = graythresh(I2);
%Convert to binary image based on graythreshold
BW = im2bw(I2,level);
show(BW,'C');
BW = bwareaopen(BW,8);
show(BW,'C2');
BW = bwdist(BW) <= 1;
show(BW,'joined');
%Complement because we want image to be black and background white
C = ~BW;
%Use distance tranform to find nearest nonzero values from every pixel
D = -bwdist(C);
%Assign Minus infinity values to the values of C inside of the D image
% Modify the image so that the background pixels and the extended maxima
% pixels are forced to be the only local minima in the image (So you could
% hypothetically fill in water on the image
D(C) = -Inf;
%Gets 0 for all watershed lines and integers for each object (basins)
L = watershed(D);
show(L,'L');
%Takes the labels and converts to an RGB (Using hot colormap)
fin = label2rgb(L,'hot','w');
% show(fin,'fin');
im = I;
%Superimpose ridgelines,L has all of them as 0 -> so mark these as 0(black)
im(L==0)=0;
clean_img = L;
show(clean_img)
For whatever reason after C = ~BW; the whole image goes dark. This very same code block has worked on a handful of other images, all of which were more "solid" or not as grainy as these. However, I thought I compensated for this with BW = bwdist(BW) <= 1;. I've experimented a ton and I don't really know what's happening. Any help would be great!
Ps. this is the image after BW = bwareaopen(BW,8);
Before the top-hat, you should perform a closing and an opening in order to reduce the noise.
If you perform an area opening on a noisy image, you may end up with the result on your black and white image.
So it would be:
Closing and opening
Top-Hat
Area opening if still necessary
Thresholding
Erosion and dilation to find the inner and outer markers respectively
Watershed (never use a watershed without markers).
can any one please help me in filling these black holes by values taken from neighboring non-zero pixels.
thanks
One nice way to do this is to is to solve the linear heat equation. What you do is fix the "temperature" (intensity) of the pixels in the good area and let the heat flow into the bad pixels. A passable, but somewhat slow, was to do this is repeatedly average the image then set the good pixels back to their original value with newImage(~badPixels) = myData(~badPixels);.
I do the following steps:
Find the bad pixels where the image is zero, then dilate to be sure we get everything
Apply a big blur to get us started faster
Average the image, then set the good pixels back to their original
Repeat step 3
Display
You could repeat averaging until the image stops changing, and you could use a smaller averaging kernel for higher precision---but this gives good results:
The code is as follows:
numIterations = 30;
avgPrecisionSize = 16; % smaller is better, but takes longer
% Read in the image grayscale:
originalImage = double(rgb2gray(imread('c:\temp\testimage.jpg')));
% get the bad pixels where = 0 and dilate to make sure they get everything:
badPixels = (originalImage == 0);
badPixels = imdilate(badPixels, ones(12));
%# Create a big gaussian and an averaging kernel to use:
G = fspecial('gaussian',[1 1]*100,50);
H = fspecial('average', [1,1]*avgPrecisionSize);
%# User a big filter to get started:
newImage = imfilter(originalImage,G,'same');
newImage(~badPixels) = originalImage(~badPixels);
% Now average to
for count = 1:numIterations
newImage = imfilter(newImage, H, 'same');
newImage(~badPixels) = originalImage(~badPixels);
end
%% Plot the results
figure(123);
clf;
% Display the mask:
subplot(1,2,1);
imagesc(badPixels);
axis image
title('Region Of the Bad Pixels');
% Display the result:
subplot(1,2,2);
imagesc(newImage);
axis image
set(gca,'clim', [0 255])
title('Infilled Image');
colormap gray
But you can get a similar solution using roifill from the image processing toolbox like so:
newImage2 = roifill(originalImage, badPixels);
figure(44);
clf;
imagesc(newImage2);
colormap gray
notice I'm using the same badPixels defined from before.
There is a file on Matlab file exchange, - inpaint_nans that does exactly what you want. The author explains why and in which cases it is better than Delaunay triangulation.
To fill one black area, do the following:
1) Identify a sub-region containing the black area, the smaller the better. The best case is just the boundary points of the black hole.
2) Create a Delaunay triangulation of the non-black points in inside the sub-region by:
tri = DelaunayTri(x,y); %# x, y (column vectors) are coordinates of the non-black points.
3) Determine the black points in which Delaunay triangle by:
[t, bc] = pointLocation(tri, [x_b, y_b]); %# x_b, y_b (column vectors) are coordinates of the black points
tri = tri(t,:);
4) Interpolate:
v_b = sum(v(tri).*bc,2); %# v contains the pixel values at the non-black points, and v_b are the interpolated values at the black points.
Let's say I have 9 MxN black and white images that are in some way related to one another (i.e. time lapse of some event). What is a way that I can display all of these images on one surface plot?
Assume the MxN matrices only contain 0's and 1's. Assume the images simply contain white lines on a black background (i.e. pixel value == 1 if that pixel is part of a line, 0 otherwise). Assume images are ordered in such a way as to suggest movement progression of line(s) in subsequent images. I want to be able to see a "side-view" (or volumetric representation) of these images which will show the surface that a particular line "carves out" in its movement across the images.
Coding is done in MATLAB. I have looked at plot (but it only does 2D plots) and surf, which does 3D plots but doesn't work for my MxNx9 matrix of images. I have also tried to experiment with contourslice, but not sure what parameters to pass it.
Thanks!
Mariya
Are these images black and white with simple features on a "blank" field, or greyscale, with more dense information?
I can see a couple of approaches.
You can use movie() to display a sequence of images as an animation.
For a static view of sparse, simple data, you could plot each image as a separate layer in a single figure, giving each layer a different color for the foreground, and using AlphaData to make the background transparent so all the steps in the sequenc show through. The gradient of colors corresponds to position in the image sequence. Here's an example.
function plotImageSequence
% Made-up test data
nLayers = 9;
x = zeros(100,100,nLayers);
for i = 1:nLayers
x(20+(3*i),:,i) = 1;
end
% Plot each image as a "layer", indicated by color
figure;
hold on;
for i = 1:nLayers
layerData = x(:,:,i);
alphaMask = layerData == 1;
layerData(logical(layerData)) = i; % So each layer gets its own color
image('CData',layerData,...
'AlphaData',alphaMask,...
'CDataMapping','scaled');
end
hold off
Directly showing the path of movement a "line" carves out is hard with raster data, because Matlab won't know which "moved" pixels in two subsequent images are associated with each other. Don't suppose you have underlying vector data for the geometric features in the images? Plot3() might allow you to show their movement, with time as the z axis. Or you could use the regular plot() and some manual fiddling to plot the paths of all the control points or vertexes in the geometric features.
EDIT: Here's a variation that uses patch() to draw each pixel as a little polygon floating in space at the Z level of its index in the image sequence. I think this will look more like the "surface" style plots you are asking for. You could fiddle with the FaceAlpha property to make dense plots more legible.
function plotImageSequencePatch
% Made-up test data
nLayers = 6;
sz = [50 50];
img = zeros(sz(1),sz(2),nLayers);
for i = 1:nLayers
img(20+(3*i),:,i) = 1;
end
% Plot each image as a "layer", indicated by color
% With each "pixel" as a separate patch
figure;
set(gca, 'XLim', [0 sz(1)]);
set(gca, 'YLim', [0 sz(2)]);
hold on;
for i = 1:nLayers
layerData = img(:,:,i);
[x,y] = find(layerData); % X,Y of all pixels
% Reshape in to patch outline
x = x';
y = y';
patch_x = [x; x+1; x+1; x];
patch_y = [y; y; y+1; y+1];
patch_z = repmat(i, size(patch_x));
patch(patch_x, patch_y, patch_z, i);
end
hold off