Pipelines

Minutiae-based fingerprint recognition

-            pressure deformation,

-            image quality,

-            skin condition,

-            forensic partials,

-            spoofing

 

1)        Preprocessing

a.        Normalisation

b.       Histogram equalisation

c.        Median filtering

2)        Ridge structure estimation and enhancement

a.        Orientation image estimation

b.       Frequency image estimation

c.        Region mask generation

d.       Gabor filtering

3)        Binarisation and skeletonization

a.        Otsu Thresholding

b.       Morphological Thinning to 1-pixel ridge

4)        Feature Extraction

a.        Minutiae Detection (Terminations and Bifurcations

b.       Remove spurious minutiae (orientation consistency)

5)        Matching

a.        Elastic matching for pressure/skin deformation

6)        Verification / Identification

a.        Verification = threshold on similarity

b.       Identification = rank list

Ridge / Texture-based Fingerprint Recognition

-            Low quality images

-            Dry/wet skin

-            spoofing

1. Pre‑processing

• Normalisation

• Gaussian averaging

2. Texture Feature Extraction

• Gabor wavelets (multi‑scale, multi‑orientation)

• Statistical filters (mean, median, Gaussian average)

3. Feature Vector Construction

• Block‑wise feature aggregation (pool Gabor/statistical responses)

• Normalisation of feature vector (reduce global contrast change sensitivity)

4. Matching

• Euclidean/cosine distance between texture vectors

5. Verification / Identification

• Thresholding for verification

• k‑NN for identification

Correlation-based fingerprint recognition

-            Small distortions

-            Partial prints

1. Pre‑processing

• Histogram equalisation

• Median filtering

• Ridge enhancement (Gabor optional)

2. Template Construction

• Extract fixed‑size patches around core/delta

• Or use whole image if well aligned

3. Matching

• Template convolution / correlation

• Peak correlation as similarity score

4. Verification / Identification

• Thresholding

• Rank‑based identification

Eigenfaces

-            Mild illumination changes

-            Mild expression changes

1. Face Detection

• HAAR wavelets

2. Pre‑processing

• Cropping and scaling

• Histogram normalisation (reduce illumination differences)

• Gamma correction (mitigate lighting variation)

3. Feature Extraction

• Vectorise face

• PCA → eigenfaces

• Project into PCA subspace

4. Matching

• Euclidean/cosine distance in PCA space

5. Verification / Identification

• Thresholding / k‑NN

LBP-based local texture face recognition

-            Illumination

-            Make-up

-            Aging

-            Expression

1. Face Detection

• HAAR wavelets

2. Pre‑processing

• Cropping and scaling

• Histogram stretching / normalisation

3. Feature Extraction

• Divide face into blocks (preserves spatial layout

• Compute LBP histograms in each block

• Concatenate histograms → feature vector

4. Matching

• Chi‑square or histogram distance

5. Verification / Identification

• Threshold / k‑NN

Gabor-based face recognition

-            Illumination

-            Expression

-            Small pose changes

1. Face Detection

• HAAR wavelets

2. Pre‑processing

• Cropping and scaling

• Histogram stretching

• Gamma correction

3. Feature Extraction

• Apply Multi‑scale, multi‑orientation Gabor filters (texture and edges)

• Pool energy responses over regions (reduce dimensionality but keep spatial)

4. Matching

• Euclidean/cosine distance between feature vectors

5. Verification / Identification

• Threshold / k‑NN

Model-based face recognition

-            Viewpoint variation

-            Occlusion

-            Aging

1.        Face detection

·        HAAR wavelets

2.        Landmark detection and shape fitting

·        Active shape models / active contours

3.        Shape description

·        Region descriptors (describe facial components)

·        Moments (achieve invariance to translation/scale/rotation)

·        Fourier descriptors (compact shape representation)

4.        Matching

·        Distance between shape descriptors

5.        Verification / Identification

·        Threshold on shape distance/ kNN on shape features

Silhouette-based gait

-            Clothing

-            Carrying objects

-            Speed variation

1. Foreground Extraction

• Background subtraction or Mixture of Gaussians

2. Silhouette Extraction

• Binary / Otsu thresholding

• Morphological cleaning (remove noise / fill holes)

3. Feature Extraction

• Average silhouette over gait cycle (capture motion pattern)

• Region descriptors (summarise body shape)

• Hu invariants for shape invariance

• Fourier descriptors of silhouette boundary

4. Matching

• Euclidean distance between silhouette descriptors

5. Verification / Identification

• Threshold / k‑NN

Velocity Moments / Temporal Silhouette features

-            Viewpoint variation

-            Speed variation

1. Silhouette Extraction

• Background subtraction / mixture of Gaussians

• Thresholding and cleaning

2. Temporal Feature Extraction

• Velocity moments (centroid motion, limb movement)

• Temporal shape moments

3. Feature vector construction

• Aggregate temporal descriptors over gait cycle

4. Matching

• Dynamic time warping or distance metrics

5. Verification / Identification

• Threshold on sequence similarity/ k‑NN on temporal features

Model-based gait

-            Clothing

-            Carrying objects

-            viewpoint

1. Pose Estimation

• Skeleton extraction from video

2. Model Fitting

• Extended pendular thigh model

• Extract parameters (frequency, amplitude, phase)

• Bilateral symmetry and Phase coupling (coordination between legs)

3. Feature Vector construction

• Combine oscillator parameters and symmetry measures

4. Matching

• Euclidean/cosine distance

5. Verification / Identification

• Threshold / k‑NN

Main Iris pipeline (Daugman-style + is canonical)

-            Lighting

-            Eyebrows

-            Eyelashes

-            Eyelids

-            Peculiarities

1. Eye Region Detection

• HAAR face → eye localisation

• Histogram equalisation (reduce lighting variation)

2. Boundary Detection

• Edge detection (Sobel/Prewitt) (highlight circular boundaries)

• Hough transform for pupil + iris circles

• Active contours for boundary refinement

• Mask eyelids/eyelashes/eyebrows (exclude occluded regions)

3. Normalisation – Rubber sheet model

• Map circular iris to rectangular strip (invariance to pupil dilation and scale)

4. Feature Extraction

• Gabor wavelets

• Phase quantisation

• Produce binary iris code

5. Matching

• Hamming distance between iris codes (bitwise dissimilarity, occlusion masking)

6. Verification / identification

• Verification threshold on Hamming distance / Identification via nearest iris code/ranking

Palmprint texture

-            Illumination

-            Incorrect placement

1. Pre‑processing

• Histogram equalisation

• Binary segmentation

• Boundary tracking

2. Coordinate system and Region Extraction

• Build coordinate system (for incorrect placement)

• Extract central palm region

3. Feature Extraction

• Gabor filters

• Form feature vector / binary code

4. Matching

• Normalised Hamming distance (if binarised)

• Or Euclidean distance

5. Verification / Identification

• Threshold / k‑NN

Eigenpalms

-            Illumination

-            Mild placement variation

1. Pre‑processing

• Histogram normalisation

• Coordinate system alignment

• Scaling

2. Feature Extraction

• Vectorise palm image

• PCA → eigenpalms

• Project palm image into PCA space

3. Matching

• Euclidean/cosine distance

4. Verification / Identification

• Threshold / k‑NN

Dual-modality palm (geometry + print)

-            Incorrect placement

-            illumination

1. Pre‑processing

• Binary segmentation

• Boundary tracking

• Coordinate system construction

2. Modality 1 — Palm Geometry

• Extract region boundary

• Compute region descriptors:

  • Area, perimeter, compactness, dispersion

  • Moments (central, normalised, Hu invariants)

  • Fourier descriptors (polar/elliptical)

3. Modality 2 — Palmprint Texture

• Extract central region using coordinate system

• Gabor filters

4. Fusion

• Concatenate geometry + texture features or fuse scores

5. Matching

• Distance metrics

6. Verification / Identification

• Verification threshold / k‑NN