non pipeline combined

Desirable Properties

-            Complete

-            Congruent

-            Compact

-            Invariant

Region descriptions – Area

-            Size of region à gait silhouette size

-            Invariant to translation and rotation

Region descriptions – Perimeter

-            Boundary length

-            Invariant to translation and rotation

Region descriptions – Compactness

-            Shape roundness

-            Invariant to scale, translation, rotation

Region descriptions – Dispersion

-            Spread of pixels

-            Invariant to scale, translation, rotation

Region descriptions – Moments

-            Zero-order = binary area, total mass

-            First-order = centroid / centre of mass

-            Second-order = orientation and spread

-            Central moments = translation invariant

-            Normalised central = translation and scale invariant

-            Hu moments = translation, scale, and rotation invariant

Region descriptions – Moments Ads and Disadvantages

Advantages

-            Work on greyscale and binary objects

-            Can utilise pixel brightness

-            Access to detail

 

Disadvantages

-            Assumes only one object present

-            Computationally expensive

-            High order moments (large numbers, sensitive to noise)

Fourier Descriptors – What and when

What

-            Represent shapes boundary by taking Fourier transform of contour

-            Produces compact set of coefficients that describe shape

 

When

-            Need a compact, rotation/scale/translation invariant description of 2D shape

-            Used in gait silhouettes, palm geometry, face outlines

Fourier Descriptors – Why and limitations

Why:

-            Fourier coefficients capture global shape structure efficiently

-            Invariance is easy (ignore DC term for translation, normalise magnitude for scale, use magnitude only for rotation)

-            Excellent for comparing shapes that differ only by pose or size

 

Limitations:

-            Sensitive to local noise or boundary errors

-            Poor at representing sharp corners or fine details

-            Requires closed contour

-            Not idea when shape undergoes non-rigid deformation

Elliptic Fourier descriptors – What and when

What

-            Represent contour using separate Fourier series for x and y coordinates

-            Produces set of elliptical harmonics that describe shape more accurately

 

When

-            Shape is complex, non-circular, not well captured by standard FD

-            Common for palm geometry, silhouette contours, irregular boundaries

Elliptic Fourier Descriptors – Why and limitations

Why:

-            Captures more detailed shape info

-            Handles asymmetric and non-smooth shapes better

-            Still allows easy invariance

 

Limitations:

-            More computationally expensive

-            Still sensitive to boundary noise and segmentation errors

-            Requires consistent contour sampling

-            Not ideal for shapes with topological changes

HAAR

-            Uses simple rectangular intensity differences

-            Used for fast face detection in real-time systems

LBP

-            Encodes local texture by comparing pixel to neighbours

-            Used for illumination-robust face recognition and local texture analysis

Active shape model – What and when

What

-            Model that learns the statistical variation of landmark points across training examples

-            Then fits shape to new images using optimisation

 

When

-            When landmark-based shape fitting under moderate variation needed

-            Face recognition, facial landmark detection, iris boundary refinement, palm geometry

Active shape model – Why and limitations

Why

-            Captures typical shape variation

-            Provides robust landmark detection even with noise

-            Enforces shape plausibility

-            Good for model-based recognition

 

Limitations

-            Requires manual landmark annotation for training

-            Struggles with large pose changes, occlusion, extreme expressions

-            Only models shape, not texture,

-            Sensitive to initialisation

Cumulative Match Curves (CMC)

-            Plots probability that the correct identity appears in the top-k matches

-            For identification performance

Correct Recognition Rate (CRR)

-            Percentage of correctly identified subjects

-            used as a simple identification accuracy measure

Inter- vs Intra-class variation

-            Used to judge how separable identities are

-            Good have low intra, high inter

Confusion Matrix

-            Table showing predicted vs actual classes

-            Used to analysis misclassifications in identification systems

TAR, FAR, FRR, TRR

-            Used to quantify verification performance and trade-offs between security and usability

Verification threshold

-            Decision boundary on similarity score

-            Used to control FAR vs FRR depending on security requirements

ROC curve

-            Plots TAR vs FAR across thresholds

-            AUC summarises performance

-            Used to compare verification systems independent of threshold

EER

-            Point where FAR = FRR

-            Used a single number summary of verification accuracy

KNN process

1)        Choose k

2)        Compute distance from test pattern to all training patterns

3)        Sort distances in ascending order

4)        Select k smallest distances

5)        Vote on these k examples

K-means process

1)        Choose k

2)        Randomly assign examples to one of the k sets

3)        Compute the mean value for each set

4)        Re-assign each point to nearest mean

5)        Repeat from step 3 until mean values remain unchanged

Fingerprint Verification/identification process

Data → processing → features → match → decision

Fingerprint Enhancements

1)        Input

2)        Image normalisation

3)        Orientation estimation

4)        Frequency image estimation

5)        Region mask generation

6)        Filtering

7)        Output

Eigenface process overview

1)        Calculate set of weights based on input image and eigenfaces by projecting input image onto each eigenface using eigen vector transform

2)        Determine if image is a face by checking if image sufficiently close to face space

3)        If face, classify weight pattern as either a known person or unknown

4)        Update eigen faces and/or weight patterns

5)        If same unknown face seen several times, calculate its characteristic weight pattern and incorporate into known faces

Active shape models process

1)        Label face points in training data

2)        Compress data using PCA

3)        Evaluate eigenvectors

4)        Acquire face image

5)        Iterate model to find face features

6)        Determine feature vector

Average silhouette process

1)        Background taken from each frame and pixels binary thresholded

2)        Normalise silhouettes by height to account for distance

3)        Add all silhouettes together and divide by number of frames

4)        Resulting image is the signature

5)        The average silhouette then changed into feature vector

Iris main strategy

1)        Detect outer boundary

2)        Detect inner boundary

3)        Extract region

4)        Normalise

5)        Encode texture

6)        Compare via hamming distance

Palm physical method

1)        Input palm

2)        Ring source

3)        Lens

4)        CCD camera

5)        A/D converter

6)        Computer

Fourier Transform – What and When

What:

-            Converts image from spatial domain to frequency domain

 

When:

-            Removing periodic noise

-            Analysing texture

-            Fast convolution (via FFT)

Fourier Transform – Why and Limitations

Why:

-            Efficient for large kernels

-            Reveals global frequency structure

 

Limitation:

-            Loses spatial localisation

Background Subtraction – What and When

What:

-            Separate moving subject from static background

 

When:

-            Gait silhouette extraction

-            Removing background clutter before segmentation

Background Subtraction – Why and Limitations

Why:

-            Simple and computationally efficient

-            Good for detecting motion

-            Clean silhouettes

 

Limitations:

-            Illumination changes

-            Dynamic backgrounds

Temporal Averaging

-            Good for stable scenes,

-            fails with illumination change

Spatiotemporal Averaging

-            more robust,

-            smooth noise across space and time

Temporal Median

-            best for removing intermittent motion,

-            robust to outliers

Mixture of Gaussians – What and When

What:

-            Models each pixel as multiple Gaussians

 

When:

-            Outdoor gait videos

-            Waving trees

-            Shadows

Mixture of Gaussians – Why and Limitations

Why:

-            more robust than simple averaging

-            handles dynamic backgrounds

-            silhouette extraction in gait

 

Limitations:

-            computationally heavier

-            Sudden illumination changes

-            If person stands still, absorbs into background

Template Convolution – What and When

What:

-            Apply filters

 

When:

-            Edge detection

-            Template matching

Template Convolution – Why and Limitations

Why:

-            Simple

-            Fast

-            Deterministic

-            Robust to small noise

 

Limitations:

-            Requires alignment

-            Computationally expensive

-            Sensitive to illumination changes

-            Not invariant to deformation

Statistical Filters – Mean

-            Smooths noise

-            Blurs edges

Statistical Filters – Median

-            Removes salt-and-pepper noise

-            Preserves edges

Statistical Filters – Gaussian

-            Smooths while preserving structure

-            Good pre-processing for edges

Gabor – What and when

What:

-            Capture local frequency and orientation

 

When:

-            Fingerprint ridge orientation

-            Iris texture encoding

-            Face recognition

-            Image coding

-            Image restoration

Gabor – Why and Limitations

Why:

-            Excellent for texture-based biometrics

-            Good for noise and illumination changes

-            multi-scale and multi-orientation

 

Limitations

-            computationally expensive

-            Sensitive to misalignment and occlusion

-            High-dimensional feature vectors

Intensity and Spatial Processing – Histogram Equalisation

-            Redistributes intensities to make histogram uniform

-            Boosts global contrast

Intensity and Spatial Processing – histogram normalisation

-            Forces image to have specific mean and variance

-            Consistent brightness/contrast across samples

Intensity and Spatial Processing – scaling

Linear – linearly maps pixel values from original range to new range

 

Linear with clipping – same as linear but clamps extreme values to avoid amplifying noise

 

Absolute value scaling – takes absolute pixel intensities before scaling (when filters produce negative values that need to be converted to usable form)

Intensity and Spatial Processing – Histogram stretching

-            Expands narrow intensity range to fill full dynamic range

-            Enhances contrast

Edge detection – First and Second derivative

First = gradient

-            Detects edge magnitude and direction

 

Second = Laplacian

-            Detects zero-crossings à edge localisation

 

When

-            Extracting boundaries (iris, palm lines, silhouette)

-            Preprocessing for shape descriptors

Edge Detection Operators

Roberts – computes diagonal gradients using tiny 2x2 operator, for fast, simple, minimal cost

 

Prewitt – uses 3x3 masks to estimate gradients, basic, noise-tolerant in low-quality

 

Sobel – Like prewitt but stronger centre weighting, more robust gradient estimation, better noise suppression

Feature Extraction – Template Matching

-            Slides template over an image to find similarity peaks

-            Used to locate known patterns when alignment controlled

Feature Extraction – Hough transform

-            Detects parametric shapes via voting

-            Iris boundaries, pupils, palm lines even with noise or partial occlusion

Feature Extraction – YOLO

-            Real-time deep learning object detector

-            Fast, robust face/person detection in unconstrained environments

Feature Extraction – Active contours

-            Deformable curves that lock onto object boundaries

-            Used for precise segmentation when edges are smooth but noisy

Feature Extraction – Unet

-            Deep learning segmentation network

-            Used for high-quality pixel-level segmentation when background subtraction unreliable

Feature Extraction – Symmetry

-            Measures bilateral symmetry

-            Used because humans exhibit stable symmetry patterns that help recognition and covariate analysis

Basic Thresholding

-            Converts image to binary using fixed threshold

-            Used for simple segmentation when lighting is stable

Otsu Thresholding

-            Automatically finds threshold that maximises inter-class variance

-            Used for robust binarization when foreground/background intensities differ

Fingerprint Patterns

• Ridges

• Singularities (Whorl, loop, arch)

• Minutiae (terminations, bifurcations)

Fingerprint challenges

• Pressure / skin deformation

• Image quality and forensic use

• Skin condition

• Legal issues

• spoofing

Fingerprint pressure correction

Finger pressure against sensor not uniform, but decreases moving from centre towards the borders

• Close-contact region = high pressure, surface friction doesn’t allow skin slippage

• Transitional region = elastic distortion produced to combine the other regions

• External region = low pressure allows skin to be dragged by finger movement

Face recognition challenges

-            Lighting / illumination

-            Viewpoint

-            Occlusion

-            Resolution

-            Facial expression

-            Aging

-            Make-up / cosmetics

Gait Applications, advantages, challenges

Applications

-            Security / surveillance

-            Immigration

-            Forensics

-            Medicine

 

Advantages

-            Perceivable at distance

-            Hard to disguise

 

Challenges

-            Clothing

-            Carrying objects

-            Viewpoint variation

-            Walking speed

Gait - Existing silhouette descriptions

-            Temporal symmetry

-            Velocity moments

-            Unwrapped silhouette

-            Average silhouette (most popular)

Gait – Modelling movement techniques

-            Pendular thigh motion model

-            Coupled and forced oscillator

-            Anatomically guided skeleton

Iris advantages and problems

Advantages

-            Uniqueness

-            Stable over lifetime

-            Non-contact

 

Problems

-            Lighting

-            Eyebrows

-            Eyelashes

-            Eyelids

-            Peculiarities

Palm features

-            Principal lines (NU)

-            Wrinkles (NU)

-            Texture

-            Palm shape

-            Minutiae-points

Palm main techniques

-            Gabor

-            Hamming distance for matching

-            Eigenpalms

-            Multi-modal extraction

Palm challenges

-            Incorrect placement

-            Illumination

-            Time-based changes