How to detect the black line from a captured image?

Hi,

I have both the same image, while one is printed out and another one is edited through softcopy.

Captured Image

I printed it out and take a photo of it using my phone camera.

Picture a

Softcopy

Picture b

My problem is, I can detect the presence of the black line in Picture b but I couldnt when I use Picture a.

Here is my code,

%%  Convert RGB to grayscale image 
if numberOfColorChannels > 1
  grayImage = rgb2gray(originalImage); 
else
  grayImage = originalImage; 
end

%% Histogram Equalization 
subplot(2, 3, 2)
imhist(grayImage);
grid on;
title('Histogram of Gray Scale Image', 'FontSize', fontSize);

% Binarize Image 
% Turn it into a binary image.
binaryImage = grayImage < 10;
% Display the image.
subplot(2, 3, 3);
imshow(binaryImage, []);
title('Binarized Image', 'FontSize', fontSize, 'Interpreter', 'None');
axis('on', 'image');
% Remove small objects.
binaryImage = imclearborder(binaryImage);
binaryImage = bwareaopen(binaryImage, 300);
% Display it.
subplot(2, 3, 4);
imshow(binaryImage, []);
title('Cleaned Binary Image', 'FontSize', fontSize);

%% Invert Binarization
binaryImage = ~ binaryImage;
subplot(2, 3, 5);
imshow(binaryImage);
title('Inverted Binary Image', 'FontSize', fontSize);
%% Scanning of black pixels 
thresh = 10;
[y1,x1] = find(binaryImage == 0, 1 );
[y2,x2] = find(binaryImage == 0, 1, 'last');

In y1 and x1, the value shows empty for Picture a.

Can someone explain to me why and provide me a solution to solve this problem?

Thank you

Answer 1

[Benjamin Großmann]

If you print it out and make a photo, then black is not pure black any more but some kind of grayish or with some kind of color fault. That being said, your limit of 10 inside this line of code

binaryImage = grayImage < 10;

is not picking anythin. You could adapt the limit or look at the color image with the Color Thresholder app. The seperation by color is not that easy anymore with a printed and captured image.

If you give more details about your problem. than i am pretty sure, that you will get more advanced and robust approaches.

[WanYu]

Hi,

Thanks for answering my question.

I am not aware of the black colour is no longer the pure black anymore earlier but I got the idea now.

Answer 2

[Adam Danz]

If you zoom in to the lower side of the image histogram you created, you'll notice a hump just before x = 40.

imhist(grayImage);
xlim([0,60])

Based on those values, it looks like a good threshold to detect the black like would be about 40 which results in the following binarized image

binaryImage = grayImage < 40;
imshow(binaryImage, []);

If the line is always expected to be horizontal (or vertical), you could use the "BoundingBox" property of regionprops() to get the position of the black line.

stats = regionprops(binaryImage, 'BoundingBox');  % acting on the cleaned image
hold on
rectangle('Position', stats.BoundingBox, 'EdgeColor', 'r')

[WanYu]

Hi,

Thanks for answering my question.

I missed out the function of my histogram equalization, now I realized where my problem is.

Is 'regionprops' a good way to detect the presence of the box? Yes, my object is always a straight line.

[Adam Danz]

As you can see in my answer, the red rectangled outlines the black line well. If the line is always horizontal or vertical, this my solution should work for you. If the line is not horizontal or vertical, you'll need to use a different regionprops property since the bounding box will only define the horizontal and vertical extent of the object.

[WanYu]

Hi,

Can the bounding box be able to get the coordinates of the vertices?

I need the coordinates of each vertex to do rotate and scaling.

[Adam Danz]

Please take a few minutes to think about what th code is doing.

The rectangle() function in my answer draws the red rectangle around the black line. That can't be done without knowing the cordinates of the rectangle (ie, the vertices). The stats.BoundingBox value has 4 values [x y w h] where x is the x coordinate of the lower, left corner of the rectangle, y is the y coordinate of the lower left corner of the rectangle, w is the width, and h is the height. You can use those value to compute the coordinates of the 4 corners of the rectangle.

Note that you could also use the ConvexHull property of regionprops but if the line is always horizontal or vertical, using the BoundingBox will be cleaner.

[Image Analyst]

Wan, don't do a global histogram equalization. It is almost NEVER a good thing to do.

An alternative way would be to look for low saturation regions using the Color Thresholder on the Apps tab of the tool ribbon. Segment in HSV color space and put the saturation thresholds to something like 0 and 0.1, and the value thresholds to 0 to 100 or something. That should get you just gray/black regions. You can then clean up with bwareafilt(bw, 1) to extract the largest blob only.

[WanYu]

Hi Adam,

I got the idea now. Thanks for explaining.

[WanYu]

Hi Image Analyst,

How is it different by using the Colour Thresholder app? Also, I don't understand the part for HSV colour space, I am new to image processing yet.