Scribd
Upload
1 views

Lecture16 digital image processing notes

The document discusses object recognition in image processing, highlighting two main approaches: decision-theoretic methods, which utilize quantitative descriptors, and structural methods, which focus on qualitative descriptors. It outlines the process of pattern recognition involving feature selection and classification, as well as the application of wavelet-based face recognition using minimum-distance classification. Additionally, it contrasts decision-theoretic and structural methods based on their use of numeric versus symbolic information.

Uploaded by

gg3385
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
1 views

Lecture16 digital image processing notes

The document discusses object recognition in image processing, highlighting two main approaches: decision-theoretic methods, which utilize quantitative descriptors, and structural methods, which focus on qualitative descriptors. It outlines the process of pattern recognition involving feature selection and classification, as well as the application of wavelet-based face recognition using minimum-distance classification. Additionally, it contrasts decision-theoretic and structural methods based on their use of numeric versus symbolic information.

Uploaded by

gg3385
Copyright
© © All Rights Reserved
Available Formats
Download as PDF, TXT or read online on Scribd
You are on page 1/ 4

Image Processing Lecture 16

Object Recognition
The automatic recognition of objects or patterns is one of the important
image analysis tasks. The approaches to pattern recognition are divided
into two principal areas:
• Decision-theoretic methods: deal with patterns described using
quantitative descriptors, such as length, area, and texture.
• Structural methods: deal with patterns best described by qualitative
descriptors (symbolic information), such as the relational
descriptors.

Patterns and Pattern Classes


• A pattern is an arrangement of descriptors (or features).
• A pattern class is a family of patterns that share some common
properties. Pattern classes are denoted w1, w2, . . . , wN where N is
the number of classes.
• Pattern recognition by machine involves techniques for assigning
patterns to their respective classes—automatically and with as little
human intervention as possible.
The object or pattern recognition task consists of two steps:
Ø feature selection (extraction)
Ø matching (classification)
There are three common pattern arrangements used in practice:
• Numeric vectors (for quantitative descriptions)

= ⋮

©Asst. Lec. Wasseem Nahy Ibrahem Page 1


Image Processing Lecture 16

• Strings and trees (for structural descriptions)


x = abababa….

Recognition Based on Decision-Theoretic Methods


These methods are based on the use of decision functions. Let x = (x1, x2,
... , xn)T represent an n-dimensional pattern vector. For N known pattern
classes w1, w2, ... , wN the idea here is to find N decision functions d1(x),
d2(x),..., dN(x) with the property that, if an unknown pattern x is said to
belong to the ith pattern class if, upon substitution of x into all decision
functions, di(x) yields the largest numerical value.

Matching
Recognition techniques based on matching represent each class by a
prototype pattern vector. Set of patterns of known classes is called the
training set. Set of patterns of unknown classes is called the testing set.
An unknown pattern is assigned to the class to which it is closest in terms
of a predefined metric. The simplest approach is the minimum-distance
classifier, which, as its name implies, computes the (Euclidean) distance
between the unknown and each of the prototype vectors. Then, it chooses
the smallest distance to make a decision.

Wavelet-based Face Recognition Application


In the enrolment stage, each face image in the training set is transformed
to the wavelet domain to extract its pattern vector (i.e. subband). The
choice of an appropriate subband varies depending on the operational
circumstances of the face recognition application. The decomposition

©Asst. Lec. Wasseem Nahy Ibrahem Page 2


Image Processing Lecture 16

level is predetermined based on the efficiency and accuracy requirements


and the size of the face image. In the recognition stage, a minimum-
distance classification method was used to classify the unknown face
images. Figure 15.1 illustrates the stages of this approach.

Figure 16.1 Wavelet-based face recognition

Let the set F = { fi,1, fi,2, fi,3, . . . , fi,m} be a training set of face
images of n subjects, where each subject i has m images. In the enrolment
stage, wavelet transform is applied on each training image so that a set
Wk(F) of multi-resolution decomposed images result. A new set LLk(F) of
all k-level LL-subbands will be obtained from the transformed face
images in the set Wk(F). The new set LLk(F) forms the set of features for
the training images. Thus, the training face image 1 of subject i (fi,1) is

©Asst. Lec. Wasseem Nahy Ibrahem Page 3


Image Processing Lecture 16

expressed by its feature vector LLk,i,1. The collection of feature vectors


LLk,i,1, LLk,i,2, . . ., LLk,i,m represents the stored template of subject i. In a
similar manner, 3 new sets HLk(F), LHk(F), HHk(F) can be created at the
same decomposition level k from the kth-level HL-, LH-, HH-subbands
respectively.

In the recognition phase, a minimum-distance classifier is used to


classify the input face image. When a probe face image is introduced to
the system, it is decomposed by wavelet transform, and a certain subband
(e.g. LLk) is chosen to represent the feature vector of the probe image. A
match score Si,j can now be computed between the probe feature vector
and each of the feature vectors j of the subject i in the feature set LLk(F).
Then, the identity (i.e. class) of the training image which gives the
minimum score is assigned to the probe image:

Si = min (Si,j) (j = 1, . . . , m)

Many similarity measures can be used for the minimum-distance


classifier, for example CityBlock or Euclidean distance functions.

Structural methods
Structural recognition techniques are based on representing objects as
strings, trees or graphs and then defining descriptors and recognition rules
based on those representations.
The key difference between decision-theoretic and structural
methods is that the former uses quantitative descriptors expressed in the
form of numeric vectors, while the structural techniques deal with
symbolic information.

©Asst. Lec. Wasseem Nahy Ibrahem Page 4

You might also like