Image Processing - Basics

• nature soft -> no straight lines or harsh corners -> difficult to use SIFT

• but images produced for specific purpose -> settings can be easily adjusted

• image representation in frequency space

• theory: every curve is representable as the integral of several sinus-curves

• most common operation for filter application

• pixel-wise calculate sum of pointwise multiplication of filter (kernel) with i mage matrix of surrounding neighboorhoud

• suppress unwanted structure -> low pass filter

• image appears smoother/blurry

• 2D sx = sy; 3D sx = sy != sz

• for each pixel, pick median value in neighborhood

• conserves shaprness

reversible - given filtered image and operation applied, original image can be restored

• Converting 8-bit image to 16-bit

• Changing color space

• Gaussian filter

irreversible - Image can not be restored even with exact details of operation

• Converting 16-bit image to 8-bit

• Decreasing resolution

• Median filter

• (Almost) Anything you do on the microscope

• image given by microscope is convolution between original object and point spread function

• deconvolution = restoration technique

• FFT for numerical approximation of PSF

• saturate percentage of pixels by stretching histogram -> contrast enhancement

• flatten histogram -> enhance hidden structures

• CLAHE: clip peaks and distribute it evenly

• find boundaries based on local intensity change

• magnitude: sqrt(dI/dx²+dI/dy²)

• gradient direction: arg(dI/dx,dI/dy)

• sample might move or deform between images -> map onto each other

• image-based -> minimize error between images

• feature-based -> position of given features (ladnmarks) matched