A Review of “Adaptive fingerprint image

enhancement with fingerprint image quality analysis”,

by Yun & Cho

Malcolm McMillan

From “Image & Vision Computing”, volume 24.All images taken from article except where quoted.

Talk StructureBackground concerning fingerprint identification.Categorizing fingerprints.Adaptive enhancement method.

Why is fingerprint identification important?Due to uniqueness of a person’s fingerprint, fingerprint analysis plays an important role in identification processes such as:

Crime scene investigations

The Fingerprint

Consists of “ridges” and “valleys”

Ridges = single, curved segments, black lines.

Valleys = area in between ridges, white.

http://perso.wanadoo.fr/fingerchip/biometrics/types/fingerprint.htm

Believed to be unique to each person.

The Identification Process

Quality Issues

Success of fingerprint identification heavily dependent upon quality of fingerprint image.

Fingerprints are often poor quality due to environmental and skin condition factors.

Thus enhancement processes are key to successful identification.

Features used for identification

Identification carried out by extracting location of features, known as “minutiae”.

2 types of feature:

- Ridge Endings

- Ridge Bifurcations (branches).

Quality Issues

a) genuine minutiae being ignored.

b) spurious minutiae being identified

eg a broken ridge will have multiple false ridge endings.

Poor quality images lead to:

The Identification Process

The Identification Process

Enhancement

Aim: To enhance key features of the image in

order to allow minutiae to be more successfully

identified.

EnhancementTraditional techniques have applied a uniform enhancement to all fingerprints,

ie the same method has been used regardless of the state of the original fingerprint.

The aim of this paper is to develop an adaptive enhancement technique,

ie one that takes into account the state of the original image and selects an enhancement technique appropriate to this.

Fingerprint Categories

Fingerprints divided into 3 categories

1. Neutral Image. Image as normal.

Fingerprint Categories

Fingerprints divided into 3 categories

2. Oily Image. Image generally darker due to some parts of valleys being filled

up (thus appearing black rather than white).Ridges either very thick or, in the extreme, merged into one.

Fingerprint Categories

Fingerprints divided into 3 categories

3. Dry Image. Image generally lighter. Ridge lines broken (due to gaps of white along ridge). Ridges lines thin.

Enhancement OverviewSo adaptive enhancement recognises that a single enhancement process is not going to be optimal for all categories. Instead we want to enhance different categories in different ways:

Oily Image:Valley enhancement – dilate/connect thin/disconnected valleys.

Neutral Image:No enhancement required.

Dry Image:Ridge enhancement - dilate/connect thin/disconnected ridges.

Selection Criteria

5 Criteria Used:

1. Mean

2. Variance

3. Block Directional Difference

4. Orientation Change

5. Ridge - valley thickness ratio

Now we need to define the criteria we will use to assign each fingerprint to a particular category.

A Clustering Algorithm using these criteria then assigns fingerprints to the appropriate class.

Adaptive Enhancement

Now that we have assigned fingerprints to their class we are in a position to perform a different enhancement process on each class.

Enhancement of Dry Images

Method:

Extract centre lines of ridges and remove white pixels in ridge (ie connect ridges) using the centre-lined image.

Want to “join up” ridges so false minutiae not detected.

Ridge Enhancement Process

1. Smoothing

Reduces noise.

2. Skeletonizing

Reduces image to basic structure of ridges.

3. DilatingWhite ridge pixels eliminated.

4. Union of dilated and original image taken to give original image with broken ridges “joined up”.

Experimental ResultsNow we have the theory behind the process of Adaptive Enhancement, we must apply it to a set of data to see if it actually improves fingerprint identification.

Analyzed 2000 fingerprints according to 5 criteria outlined previously and used clustering to assign fingerprints to one of dry, oily, or normal.

Experimental Results: Clustering

Oily

Neutral

Dry

Experimental Results: EnhancementNow we have categorized our fingerprints we can perform adaptive enhancement and compare our results with conventional enhancement.

Adaptive filtering yields improved results over conventional methods.

Left-hand side = conventionally enhanced.

Right-hand side = adaptively enhanced.

Feature ExtractionTo the eye, we can see that adaptive enhancement produces a better image.

This is borne out when we extract features from both conventionally enhanced and adaptively enhanced images.

Images (a) & (c) enhanced conventionally.Images (b) & (d) enhanced adaptively.

Conclusions 1

Quality measured quantitatively calculating proportion of correctly identified minutiae.

Measuring Quality Quantitatively

Adaptive enhancement shows an improvement in image quality over conventional enhancement.

Fingerprint identification relies on image quality.

The experimental results indicate an improvement from 92% to 96% in correctly identified fingerprints.

Conclusions 2

Costs

Estimated increase in computational time for adaptive enhancement is approximately 0.5 seconds per fingerprint.

Is the increase in quality worth the wait?

Further Issues• Is 3 the optimum number of classes?• Can we develop other enhancement schemes for classes of fingerprints with different properties?• If so, can we get better results with more classes?