Download as docx, pdf, or txt
Download as docx, pdf, or txt
You are on page 1of 3

Bachelor in Data Science and Engineering – Audio Processing, Video Processing and Computer Vision

Miguel García de Julián 100409352


Jorge Gómez-Bravo Peche 100406641

Introduction
Optical microscopy has become one of the most powerful techniques in biomedical research,
particularly when combined with digital image analysis. Since an optical microscope uses
visible light and a system of lenses to magnify images, as opposed to electrons and
electromagnets present in electron microscopes, the size of the smallest samples an optical
microscope is able to resolve is directly proportional to the wavelength of visible light. Among
all types of optical microscopes, one variety is a fluorescence microscope, which uses
fluorescence instead of reflection and absorption as a source of image contrast.

Optical microscopy exhibits many challenges for digital image analysis. In general, microscopy
volumes are inherently anisotropic, suffer from decreasing contrast with tissue depth, lack
object edge detail, and characteristically have low signal levels.

Table including each proposed technique and the obtained results.

Otsu 0.690
otsu + binary_clossing + labels 0.692
otsu + binary_clossing + binary_fill_holes + labels 0.638
otsu + binary_clossing + labels + median 0.695
Cellpose 0.776
median filter + Cellpose 0.861

A brief description of the proposed techniques.


- Gaussian filter.

Linear filters work by convoluting an image with a smaller template. That small template can
be anything, but commonly it is a well-defined small matrix. For example, it can be a 2D
Gaussian. The Gaussian has default values, and we can simply apply it on an image.
Gaussian filtering does not only blur information, but it also helps removing noise.

1
Bachelor in Data Science and Engineering – Audio Processing, Video Processing and Computer Vision

Miguel García de Julián 100409352


Jorge Gómez-Bravo Peche 100406641

- Median filter.

Another class of filters does not use convolution to filter the image but does some other type
of operation on small neighbourhoods of the image. For example, it can calculate the median
value of each 3x3 sub-region in the image. This is extremely useful if our image has pixels that
are "dead" or have extremely large values.

- Automatic Thresholding.

Several methods have been developed to detect an "ideal" threshold based on the pixel
histogram. One of the most famous ones is the Otsu method: it sets a threshold by dividing the
pixels in two classes and tries to minimize the variance of both classes.

Some other thresholds are niblack, sauvola, local and li. We tried to experiment with these
other methods, but the results were really poor, with the exception of li.

- Erosion and dilation filters.

When working with masks, one can use another class of filters, that just turn pixels ON and
OFF depending on their surroundings. The simplest to understand are the erosion and
dilation filters. As their name indicate, these filters either shrink or enlarge ON objects. The

2
Bachelor in Data Science and Engineering – Audio Processing, Video Processing and Computer Vision

Miguel García de Julián 100409352


Jorge Gómez-Bravo Peche 100406641
size and shape of the filters, specify what region is considered around objects to shrink or
enlarge them. 

- Skeleton filter.

Another meethod to shrink as much as possible all the structures of the image.

- Fill holes.

REFERENCES

https://engineering.purdue.edu/~micros/

You might also like