I am trying to extract just the currency image from the an image of a currency taken on white paper. I tried median filter to get the background and use a simple subtraction like:
image=image-background.
But I got two problems: 1) The median filter is too slow. 2) The resulting image is not as my expectation.
Here is the image:
Any better way to do it, please?
You can try detecting the boundary of the note using edge detection and the hough transform. Alternatively, you can find the boundary using the activecontour function.
Another possibility is to detect the note using hue, using the fact that it is green and the background is not. Transform the image to HSV color space using rgb2hsv, and then use greythresh on the hue component.
By the way, this is more of an image segmentation problem. Background subtraction usually refers to separating moving objects from a static background in video.
Related
Could you please advice what image processing transform can I use in order to correct character blurring after text scanning? Afterwards, i am planning to remove uneven background illumination using top-hat transforms.
You need spatially dependent deconvolution. I think, the point scattering function (PSF) here is ellipse (in left part of image).
I need to implement an image segmentation function in MATLAB based on the principles of the connected components algorithm, but with a few modifications. This is intended for very simple, 2D images, with a background color and some objects in different colors.
The idea is that, taking the image as a matrix, I provide a tool to select the background color (it will vary for every image). Then, when the value of the color of the background of the image is selected, I have to segment all the objects in the image, and the result should be a labeled matrix, of the same size of the image, with 0's for the background, and a different number for each object.
This is a graphic example of what I mean:
I understand the idea of how to do it, but I do not know how to implement it on MATLAB. For each pixel (matrix position) I should mark it as visited and then if the value corresponds to the one of the background, assign 0, if not, assign another value. The objects can be formed by different colors, so in the end, I need to segment groups of adjacent pixels, whatever their color is. Also I have to use 8-connectivity, in order to count the green object of the example image as only one object and not 4 different ones. And also, the objects should be counted from top to bottom, and from left to right.
Is there a simple way of doing this in MATLAB? I know the bwlabel function, but it works for binary images only, so I'd like to adapt it to my case.
once you know the background color, you can easily convert your image into a binary mask of the same size:
bw=img!=bg_color;
Once you have a binary mask you can call bwlavel with 8-connectivity argument as you suggested yourself.
Note: you might want to convert your color image from RGB representation to an indexed image using rgb2ind before processing.
I have an ellipse in the image.After segmentation i got a broken ellipse as shown .which morphological operation is used to get the perfect ellipse
Actual input file is
output obtained is
i tried imopen ,but i will lose lower ellipse like structure .how to close the upper ellipse like structure without losing lower ones
Mask i created is
i want to segment the ellipse like structure.but some of these structures are connected with rectangular like bodies.how to separate it. erode will eliminate small ellipses
If you want to reconnect something with a mathematical morphology operator, do not use an opening (it increases the gap), but a closing (imclose)! The names are explicits.
In you case, you want to reconnect something vertically cut, so use a horizontal structuring element (type segment).
And yes, you have to invert your image, black pixels representing the absence of information.
Usually, for closing gaps, you would need the close operator.
However, since most software assume active pixels are white, you would either need to invert the image, or use the open operator.
On this image, in matlab, the following works well:
imopen(I,ones(32))
This uses a square structuring element. You may want to experiment with other shapes.
Your example also looks like you moved half of the ellipse, as opposed to some process which deleted pixels in the middle. No simple morphological operation can create a perfect ellipse out of the sample image, unless you use the knowledge that multiple components can be moved to re-form the ellipse. If that is the actual case, you can scan connected components and try to match them together.
I'd like to resize the components contained in a 3D binary image sequence without changing any of the dimensions of the sequence itself.
I'm not sure if I need to do it on a component-by-component basis, if yes, then how do I create a transform such that the resized components are re-positioned 'correctly' in the image sequence? By 'correctly', I mean with the same centre of mass as the original unprocessed components.
(If that last paragraph doesn't make sense then please ignore)
A 2D example: suppose I wanted to enlarge by 10% the white blobs in the following [295x445] image
How would you do this without making the image itself larger?
you could use the imdilate function to dilate the regions of interest. The examples in the webpage show how to use this function.
I've converted a colored photo to black and white, and bolded the edges. Now i need to convert it back to its original color with the bolded edges. Is there any function in matlab which allows me to do so?
Once you remove the colour from an image, there is no possible way to automatically put it back. You're basically reducing a set of 16,777,216 colours to a set of 256 - on average each shade of grey has 65,536 equivalent colours, and without the original image there's no way to guess which it could be.
Now, if you were to take the bolded lines from your black-and-white image and paint them on top of the original coloured image, that might end up producing what you're looking for.
If what you are trying to do is to use some filter over the B/W image and then use that with the original color. I suggest you convert your image to a color space with Lightness channel that suits your needs (for example L*a*b* if you need the ligtness to be uniformly distributed regarding human recognition of differences) and apply your filter only over the Lightness channel.