Unit-4

Shape Representation & Segmentation

Shape representation generally looks for effective and perceptually important shape features based on
either shape boundary information or boundary plus interior content.

Superquadrics are a family of parametric shapes that are used as primitives for shape representation in
computer graphics and computer vision.

Shape representation and description techniques can be generally classified into two broad categories:
Contour-based methods and Region-based methods.

Contour Based Representation

Contour Representation

We represent contours with chain codes and shape numbers. These parameters help in clear
representation and a better understanding of object contour. However, computing these parameters is
purely optional in the process of finding contours.
There are 3 kinds of contour lines you'll see on a map: intermediate, index, and supplementary. Index
lines are the thickest contour lines and are usually labeled with a number at one point along the line.
This tells you the elevation above sea level.
Types of Contour Lines
Blind Contour Line. A blind contour line is created by drawing an object while looking only at the object,
not at the paper.
Continuous Contour Line. Continuous contour lines are created by drawing a subject without lifting the
pen from the paper.

Cross Contour Line.

Implied Contour Line.

Shapes of Contour lines

V-shaped contour lines indicate valleys, ravines, or gullies. Closely-spaced contour lines depict steep
slopes. More widely- spaced contour lines indicate gradual slopes. Irregular, zigzag contour lines indicate
uneven terrain.

Region Based Representation

A region-based approach to shape representation and similarity measure is presented. The shape
representation is invariant to translation, scale and rotation. The similarity measure conforms to human
similarity perception, i.e., perceptually similar shapes have high similarity measure. Graph
representation based on region skeleton
Objects are represented by a planar graph with nodes representing subregions resulting from region
decomposition, and region shape is then described by the graph properties.

Region-based approaches try to find partitions of image pixels into sets corresponding to coherent
image properties such as brightness, color and texture. Contour- based approach usually start with a
first stage of edge detection followed by a linking process that seeks to exploit curvilinear continuity.

Region based active contour method

The active contours technique, also called snakes, is an iterative region-growing image segmentation
algorithm. Using the active contour algorithm, you specify initial curves on an image and then use the
active contour function to evolve the curves towards object boundaries.
Deformable Curves & surfaces

A deformable curve or active contour is a curved defined within the domain of an image. The properties
of the active contour and its behaviour are specified by the energy functional. A partial differential
equation controlling the snake causes it to evolve so as to reduce the energy.

Deformable models are curves or surfaces defined within an image domai n that can move under the
influence of internal forces coming within the model itself and external forces computed from the image
data.

Parametric Deformable Model

Parametric deformable models, which synthesize parametric curves or surfaces within an image domain
and allow them to move toward desired features, usually edges.

Geometric Deformable Model

A deformable curve or active contour is a curved defined within the domain of an image. The properties
of the active contour and its behaviour are specified by the energy functional. A partial differential
equation controlling the snake causes it to evolve so as to reduce the energy.

Snakes & Active Contours

The Greedy Snake algorithm is a feature extraction technique, also known as active contours. It works
like stretched Elastic Band being released. The Initial Points are defined around feature to be extracted.
The points then move through an iterative process to move to a point with the lowest energy function.
Active contours, or snakes, are computer-generated curves that move within images to find object
boundaries. The active contour model is a method to minimize the energy function from an external and
internal force.

Snake active contour algorithm

The active contours technique, also called snakes, is an iterative region-growing image segmentation
algorithm. Using the active contour algorithm, you specify initial curves on an image and then use the
active contour function to evolve the curves towards object boundaries. The different types of active
contour models defined are traditional snake, balloon, gradient vector flow and geodesic active contour
model.

Level Set Representations

Level set methods in computer vision: Image segmentation, tracking, statistical shape modeling,
multiview reconstruction, ... There are numerous methods to speed up level set methods. In
combination these lead to segmentation speeds close to real time for volumetric data of 512 x 512 x 64
voxels. Level set is one of active contour models, which is good at handling complex topologies and
capturing boundary.

The level set methods are specially used in image with intensity inhomogeneity, such as medical image,
SAR image, etc. Level Sets are an important category of modern image segmentation techniques based
on partial differential equations (PDE), i.e. progressive evaluation of the differences among neighboring
pixels to find object boundaries. It can be used in numerical simulation of flow, for example, if we are
working with discretization of the interface water-air, compacts at sixth order, ensures the accurate and
fast calculation of the interface equations.
The LSM uses a distance function to locate different fluids. The level set and phase field methods are
both field-based methods in which a free fluid surface is represented as an isosurface of the level set or
phase field functions. The free liquid surface corresponds to the phase boundary between the liquid and
the gas and is represented on a fixed mesh. A level set of a function of two variables f(x,y) is a curve in
the two-dimensional xy-plane, called a level curve. A level set of a function of three variables f(x,y,z) is a
surface in three-dimensional space, called a level surface. Level sets, the gradient, and gradient flow are
methods of extracting specific features of a surface. You've heard of level sets and the gradient in vector
calculus class – level sets show slices of a surface and the gradient shows a sort of 2D “slope” of a
surface.

Zero Level Set Method

The basic idea is to represent the curves or surfaces as the zero level set of a higher dimens ional hyper-
surface. This technique not only provides more accurate numerical implementations but also handle
topological change very easily. The blue-green surface on the right below is called the level set function,
because it accepts as input any point in the plane and hands back its height as output. The red front is
called the zero level set, because it is the collection of all points that are at height zero.
The zero set Z(f) of an analytic function f as a multiset or a sequence, where each zero w is listed with its
multiplicity, which is m if z ↦→ f(z)/(z − w)m is analytic and does not vanish at w. The level -set method
is an approach to topology optimization that uses a flexible implicit description of the material domain.
This structural domain is represented by a level-set function whose zero-level contour defines the
structural boundary.
Particle Level Set Method

The method, a hybrid “Particle Level Set Method”, uses Lagrangian marker particles to rebuild a level set
representation of the interface in regions which are under-resolved. This is often the case for flows
undergoing stretching and tearing. Level set function φ is used to represent interface based on concept
of implicit surfaces; with a constant value at the interface. Although the physically relevant thickness is
negligible, a numerical rel- evant interface of thickness 2ǫ is defined in LSM.

The Level Set Equation

φt + F|∇φ| = 0.

Fourier & wavelet Descriptors

Fourier Descriptors

Fourier descriptors and moment invariants are the most widely used shape representation schemes. The
main idea of a Fourier descriptor is to use the Fourier transformed boundary as the shape feature.
Moment invariant technique uses region-based moments, which are invariant to transformations, as the
shape features.

The advantage is that it is possible to capture coarse shape properties with only a few numeric values,
and the level of detail can be increased (or decreased) by adding (or removing) descriptor elements. The
Fourier transform is a representation of an image as a sum of complex exponentials of varying
magnitudes, frequencies, and phases. A generic Fourier descriptor (GFD) is proposed to overcome the
drawbacks of existing shape representation techniques. The proposed shape descriptor is derived by
applying a 2D Fourier transform on a polar raster sampled shape image. The acquired shape descriptor is
application independent and robust.
The image feature descriptors are divided into three categories according to the characteristics of
different, namely local gradient-based descriptors, image intensity-based descriptors, and learning-
based descriptors.

Fourier analysis is used most frequently as a univariate method for either simplifying data or for
modeling. It can also be used as a multivariate technique for data analysis. In a sense, Fourier analysis is
similar to trend analysis. It evaluates the relationship of sets of data from a different perspective.

The Fourier Transform possesses the following properties: 1) Linearity. 2) Time shifting. 3) Conjugation
and Conjugation symmetry. 4) Differentiation. The main advantage of Fourier analysis is that very little
information is lost from the signal during the transformation. The Fourier transform maintains
information on amplitude, harmonics, and phase and uses all parts of the waveform to translate the
signal into the frequency domain. There are two types of descriptors: data descriptors and non-data
ones. If a descriptor implements both 1 get () and set () , it's called a data descriptor; otherwise
is a non-data descriptor.

Wavelet Descriptors

The wavelet descriptors are formed on the basis of wavelet representation of the original sequence
(xi,yi), describing the boundary of the shape.

Wavelet transforms
Continuous wavelet transform (CWT)

Discrete wavelet transform (DWT)

Fast wavelet transform (FWT)
Lifting scheme and generalized lifting scheme.

Wavelet packet decomposition (WPD)

Stationary wavelet transform (SWT)

Fractional Fourier transform (FRFT)

Fractional wavelet transform (FRWT)

There are two types of wavelet transforms: the continuous wavelet transform (CWT) and the discrete
wavelet transform (DWT). Specifically, the DWT provides an efficient tool for signal coding.

A wavelet is a mathematical function applied in digital image processing and compression. Its main aim
is to improve the image quality. Also, wavelets can divide signals into time and frequency components.
The method of wavelet transforms, which provides a decomposition of functions in terms of a fixed
orthogonal family of functions of constant shape but varying scales and locations, recently acquired
broad significance in the analysis of signals and of experimental data from various physical phenomena.

A zero phase wavelet is symmetrical with a maximum at time zero. Zero phase wavelets have energy
before time zero, which makes them noncausal, and therefore they are not physically realizable.

Fourier transforms approximate a function by decomposing it into sums of sinusoidal functions, while
wavelet analysis makes use of mother wavelets. Both methods are capable of detecting dominant
frequencies in the signals; however, wavelets are more efficient in dealing with time-frequency analysis.
While the Fourier transform creates a representation of the signal in the frequency domain, the wavelet
transform creates a representation of the signal in both the time and frequency domain, thereby
allowing efficient access of localized information about the signal.

In the case of db 20 wavelet decomposition of the signal having 60 Hz, 300 Hz and 660 Hz frequency
components, the energy can be seen to be concentrated in level 2 (40 Hz to 80 Hz), level 4 (160 Hz to
320 Hz), level 6 (640 Hz to 1280 Hz) and level 5 (320 Hz to 640 Hz) of the wavelet decomposition. The
wavelet transform analysis relies on the following three parameters: mother wavelet , sampling
frequency f, and level of decomposition l. The outputs A and D are the reconstruction wavelet
coefficients: A: The approximation output, which is the low frequency content of the input signal
component. D: The multidimensional output, which gives the details, or the high frequency components,
of the input signal at various levels (up to level 6).

Two types of wavelet transform are

Continuous wavelet transform (CWT)

Discrete wavelet transform (DWT)

The CWT and the discrete wavelet transforms differ in how they discretize the scale parameter. The
CWT typically uses exponential scales with a base smaller than 2, for example 21/12 . The discrete
wavelet transform always uses exponential scales with the base equal to 2.

Medical Representations & Multi Resolution Analysis

The wavelet representation then consists of the low-pass band at the lowest resolution and the highpass
bands obtained at each step. This transform is invertible and nonredundant. In order to develop a
multiresolution analysis, a scaling function is needed, together with the dilated and translated version of
it. The wavelet representation is a wavelet series representation of a square -integrable function with
respect to either a complete, orthonormal set of basis functions, or an overcomplete set or frame of a
vector space, for the Hilbert space of square-integrable functions.
Wavelet transforms are now being adopted for a vast number of applications, often replacing the
conventional Fourier Transform. Wavelet Transforms (WT) can also be used in the field of Image
compression, Feature extraction, image denosing and other medical image technology.

MultiResolution Analysis(MRA)

A multiresolution analysis (MRA) or multiscale approximation (MSA) is the design method of most of the
practically relevant discrete wavelet transforms (DWT) and the justification for the algorithm of the fast
wavelet transform (FWT).

Multiresolution analysis provides information on both the spatial and frequency domains. Here we
describe multiresolution analysis from a wavelet perspective and provide a simple example. The wavelet
transform is the foundation of techniques for analysis, compression and transmission of images.
Multiresolution offers an efficient framework for extracting information from images at various levels of
resolution. Several applications in diverse fields use this idea. Pyramid algorithms use a multiresolution
architecture to reduce the computational complexity of image processing operations.
A multiresolution image is a type of multi-scale representation, in which an image is subject to repeated
scaling. With a multiresolution format, large raster image files such as aerial photographs or satellite
imagery are compressed and can be quickly viewed without having to decompress the entire file.
The multiresolution analysis satisfies the following conditions:

𝖴 j = − ∞ ∞ V j is dense in L2(R).
∩j = −∞ ∞ Vj = { 0} .

f(x) ∈ Vj if and only if f(2x) ∈ Vj + 1, ∀j ∈ Z.

Central to the multiresolution approach is to somehow express short-range, mid-range, and long-range
relationships explicitly. The main reasons for a multiresolution approach is one of: improving
performance, by capturing long-range phenomena that would otherwise not be utilized. Multiresolution
Motion Estimation (MRME) and Compensation has been reported in literature. for the exploitation of
temporal redundancy in video compression. This technique estimates the. motion vectors hierarchically
from lower resolution to higher resolution subimages and thus. reduces the complexity significantly.
A multiresolution analysis of L2(R) consists of a nested sequence of closed subspaces ⋯ V j − 1 ⊂ V j ⊂ V
j + 1 ⋯ for approximating L2(R) functions. The multiresolution analysis satisfies the following conditions:
1. 𝖴 j = − ∞ ∞ V j is dense in L2(R).