... | @@ -13,7 +13,7 @@ For particles of disc shape, a picture with white light transmitted through part |
... | @@ -13,7 +13,7 @@ For particles of disc shape, a picture with white light transmitted through part |
|
|
|
|
|
* Then a low-pass filter is applied to the Fourier transform of the image to remove noises or default impurities by setup from the background and increase contrast.
|
|
* Then a low-pass filter is applied to the Fourier transform of the image to remove noises or default impurities by setup from the background and increase contrast.
|
|
|
|
|
|
* Finally, there are two ways to detect particles, depending on the image contrast quality. (1) If it is possible to binarize the image (1 for particles and 0 for background), then a convolution method is recommended. There are several methods to binarize an image: [imbinarize function in Matlab with a fixed threshold](https://www.mathworks.com/help/images/ref/imbinarize.html), [adaptive thresholding](http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.420.7883&rep=rep1&type=pdf), and [otsu method](https://ieeexplore.ieee.org/document/4310076). After image binarization, a convolution is applied between the image and a pre-set disk mold whose diameter is approximately 3/4 of that of the smallest particle. The resulted convoluted image will be a collection of isolated white bulbs, whose centers reveal the centers of corresponding particles and area is related to particle diameters. (2) If it is too difficult to binarize the image, a circular Hough transform is recommended to be applied to the gray scale image to find centers and diameters of the discs. The circular Hough transform is an algorithm computing the curvature at each pixel point based on the image gray scale gradient. The curvature will automatically give the center and radius. Then a voting process rules out fake circles and combines circles with close enough centers and similar radii.
|
|
* Finally, there are two ways to detect particles, depending on the image contrast quality. (1) If it is possible to binarize the image (1 for particles and 0 for background), then a convolution method is recommended. There are several methods to binarize an image: [imbinarize function in Matlab with a fixed threshold](https://www.mathworks.com/help/images/ref/imbinarize.html), [adaptive thresholding](http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.420.7883&rep=rep1&type=pdf), and [otsu method](https://ieeexplore.ieee.org/document/4310076). After image binarization, a convolution is applied between the image and a pre-set disk mold whose diameter is approximately 3/4 of that of the smallest particle. The resulted convoluted image will be a collection of isolated white bulbs. Centers of these bulbs correspond to the centers of particles, and their areas are related to particle diameters, which can be pre-assigned. (2) If it is too difficult to binarize the image, a circular Hough transform is recommended to be applied to the gray scale image to find centers and diameters of the discs. The circular Hough transform is an algorithm computing the curvature at each pixel point based on the image gray scale gradient. The curvature will automatically give the center and radius. Then a voting process rules out fake circles and combines circles with close enough centers and similar radii.
|
|
|
|
|
|
The procedure described above with results at each step is illustrated in the figure below.
|
|
The procedure described above with results at each step is illustrated in the figure below.
|
|
![Fig_DisksPos](uploads/fa70a6add46ab4038df159167cede62e/Fig_DisksPos.png)
|
|
![Fig_DisksPos](uploads/fa70a6add46ab4038df159167cede62e/Fig_DisksPos.png)
|
... | | ... | |