Introduction to Image Processing§
author: | David Coeurjolly |
---|
author: | David Coeurjolly |
---|
We only focus on values of the functions using stochastic representation
Important object: histograms
The image can be seen as a random variable with probability density function
Exemple:
Histograms sum up colorimetric information of the image without considering spatial relationships.
Few examples:
Translation
Histogram remapping
Inversion
…
More generally
Propagation to image <demo>
In previous examples, was prescribed. Now, we are looking for to map histogram (empiric or observed) to (model).
Example: Equalization
Be careful not always a good idea
Image Segmentation classify pixels into classes such that intra-class pixels share the same visual properties (colorimetric information, geometrical properties …)
Let’s go back to initial image:
Histogram Thresholding
<slides>
Take into account spatial relationships between pixel values
Key tool: convolution
Image as a function
We consider the result of the convolution product of by
or a discrete version for compact support kernel
Algorithmically
Direct discrete computation of the support is small
Or by regular product on Fourier space:
(for )
Linearity
Scaling
Transaltion
Modulation for real
Convolution
DCT component
(magnitude of Fourier transform leads to translation and rotational invariant descriptors)
(http://www.cs.unm.edu/~brayer/vision/fourier.html)
Pure cosine signals | Magnitude vs. Phase | Real image example |
(http://www.cs.unm.edu/~brayer/vision/fourier.html)
Continuous vs. discrete FT | Edge effect reduction through convolution |
Periodic structure:
(http://www.cs.unm.edu/~brayer/vision/fourier.html)
Lowpass filter | Highpass filter |
Smoothing / Low-pass filters
G : Gaussian kernel approximation function
binomial coefficients, ….
Match between filtering and noise model
Gaussian noise | ||
Specular noise salt/pepper | ||
Contour spread |
Complementary filters to low-pass filters
Median filter
15
…but…
complexity for a size domain?
Context
loci if high gradient vector norm of the function
Zero-crossing of the Laplacian …
We will focus later to segmentation process.
Definition
Finite difference approximation
(other masks exist along specific directions )
Amplitude = gradient vector norm
Orientation
a contour can be characterized has pixels with high gradient vector norm
(Prewitt)
Gradient of ta filterer image and Gaussian smoothing filters
let us consider Gaussian kernel filter , we have
g | g’ | g’‘ |
Let be a convolution filter ( Impulse response of a filter). The filter is separable if
Thus:
and for partial derivatives:
direct consequences when implementing filters (only 1D convolutions)
i.e. Prewitt’s filter corresponds to finite difference approximation of the result of a constant smoothing filtering of
Canny’s criteria (1983)
Noise model + signal model + Criteria modeling + initial conditions Optimal filter as solution of a PDE
Shen-Castan : | Deriche : |
The Gaussian filter is a good approximation of such optimal filters
Contour extraction requires gradient vector norm (or zero-crossing laplacian) thresholding
Naive approach: belongs to a contour iff
From hysteresis : belongs to a contour iff
or
and neighbor of such that
[Tomasi & Manduchi, 98]
Non-linear filter with a pair of spatial intensity kernels
(From [Bilateral Filtering: Theory and Applications Sylvain Paris, Pierre Kornprobst, Jack Tumblin, and Frédo Durand Foundations and Trends in Computer Graphics and Vision, 2009])
Question: How to efficiently implement bilateral filtering ?
[Perona, Malik]
Images as functions
Diffusion process defined on time-varying images:
with for example:
Anisotropic diffusion process that removes noise preserving edges
Rationale: for some we want to minimize
and thus
N.B Bilateral filtering is asymptotically close to Perona-Malik’s diffusion
Idea Variational formulation edge aware smoothing
But:
convergence mathemtatical framework
:-)
[Matheron, Serra, …]
Idea
Translation
Dilation by a structuring element B
Erosion by a structuring element B
Properties
with and
Structuring element: Euclidean disc
(blue: set , gray circle: structuring element and cyan: result of the operators)
Opening B
Closing
3x3 structuring element
Properties
Opening is anti-extensive:
Closing is extensive:
Useful tool for granulometric analysis
Sequence of increasing structuring elements k times
is called the granulometry function of and the spectrum
Intuitive explanation
is defined as the union of grains and is the size of the set defined by grains larger than k
Operators on gray-level images
Example
Let suppose a grayscale image and a constant structuring element (whose origin is it mid-point, so-called flat structuring element)
Then
Basic “à-la” finite difference definition
Gradient
Laplacian
Mathemtatical model
Operators acting oncomplete lattices
Principle
Given an specific image
Select the best structuring element(s)
Specify the combination of fundamental operators (e.g. series of opening/closing)
is a filter iff it is increasing and idempotent
are filters
are filters and
[Soille]