Multimedia Database Systems
description
Transcript of Multimedia Database Systems
Multimedia Database Systems
Relevance Feedback
Department of InformaticsAristotle University of Thessaloniki
Fall 2008
Outline
Motivation for Relevance Feedback (RF)Introduction to RF RF techniques in Image Databases (5 techniques
are studied)Other RF techniquesConclusionsBibliography
Motivation
Initial work on content-based retrieval focused on using low-level features like color and texture for image representation.
After each image is associated with a feature vector, similarity between images is measured by computing distances between feature vectors in the feature space.
It is generally assumed that the features are able to locate visually similar images close to each other in the feature space so that non-parametric approaches, like the k-nearest neighbor search, can be used for retrieval.
Motivation
There are cases where the user is not satisfied by the answers returned.
Several relevant objects may not be retrieved or in addition to the relevant objects there are a lot of non-relevant ones.
Possible solutions:– Request more answers (e.g., next 10)– Rephrase and reexecute the query– Relevance feedback
A Possible Solution: RF
Take advantage of user relevance judgments in the retrieval process:– User issues a query and gets back an initial hit list– User marks hits as relevant or non-relevant– The system computes a better representation of the
information need based on this feedback– This process can be repeated more than once.
Idea: you may not know what you’re looking for, but you’ll know when you see it.
Forms of RF
Explicit feedback: users explicitly mark relevant and irrelevant documents
Implicit feedback: system attempts to infer user intentions based on observable behavior
Blind feedback (also known as pseudofeedback): feedback in absence of any evidence, explicit or otherwise
The Goal of RF
x
x
xx
oo
o
Revised queryx non-relevant objectso relevant objects
o
o
ox
x
x x
xx
x
x
xx
x
x
x
x
Initial query
x
RF in Text Retrieval
RF was originally proposed for text-based information retrieval.
The goal is to improve the quality of the returned documents.
Fundamental work: Rocchio
Rocchio Method
Used in practice:
New query– Moves toward relevant objects– Away from irrelevant objects
nrjrj Dd
jnrDd
jr
m dD
dD
110
qm = modified query vector; q0 = original query vector;α,β,γ: weights (hand-chosen or set empirically); Dr = set of known relevant doc vectors; Dnr = set of known irrelevant doc vectors
Rocchio Example
vector feedback negative
vector feedback positive
vectorquery originalvectorquery
0 4 0 8 0 0
1 2 4 0 0 1
2 0 1 1 0 4
-1 6 3 7 0 -3
0 4 0 8 0 0
2 4 8 0 0 2
8 0 4 4 0 16
Original query
Positive Feedback
Negative feedback
0.1
5.0
25.0
(+)
(-)
New query
Typically, <
RF Example
RF Example: initial results
RF Example: user selection
RF Example: revised results
RF Example: alternative interface
Some RF Techniques
1. Yong Rui, Thomas S. Huang and Sharad Mehrotra. “Content-Based Image Retrieval with Relevance Feedback in MARS”, International Conference on Image Processing (ICIP), 1997.
2. Selim Aksoy, Robert M. Haralick, Faouzi A. Cheikh, Moncef Gabbouj. “A Weighted Distance Approach to Relevance Feedback”, International Conference on Pattern Recognition (ICPR), 2000.
3. Zhong Su, Hongjiang Zhang, Stan Li, and Shaoping Ma. “Relevance Feedback in Content-Based Image Retrieval: Bayesian Framework, Feature Subspaces, and Progressive Learning”, IEEE Transactions on Image Processing, 2003.
4. DeokHwan Kim, ChinWan Chung. “Qcluster: Relevance Feedback Using Adaptive Clustering for ContentBased Image Retrieval”, SIGMOD, 2003.
5. Junqi Zhang Xiangdong Zhou Wei Wang Baile Shi1 Jian Pei. “Using High Dimensional Indexes to Support Relevance Feedback Based Interactive Images Retrieval”, VLDB, 2006.
CBIR with RF in MARS
There is an urgent need to develop integration mechanisms to link the image retrieval model to text retrieval model, such that the well established text retrieval techniques can be utilized.
This paper studies approaches of converting image feature vectors (Image Processing domain) to weighted-term vectors (IR domain).
Furthermore, the relevance feedback technique from the IR domain is used in content-based image retrieval to demonstrate the effectiveness of this conversion.
Experimental results show that the image retrieval precision increases considerably by using the proposed integration approach.
The method has been implemented in the MARS prototype system developed at the University of Illinois @ Urbana Campaign.
Weighted Distance Approach
Selim Aksoy, Robert M. Haralick, Faouzi A. Cheikh, Moncef Gabbouj. A Weighted Distance Approach to Relevance Feedback, Proceedings of International Conference on Pattern Recognition (ICPR), 2000.
Weighted Distance Approach
kjrelF ,
kR
krelR
kjF
Q
K number of iterations
number of features in feature vector
retrieval set after the k-th iteration
set of objects in kR marked as relevant
values of the j-th feature component of images in kR
values of the j-th feature component of images in krelR
Weighted Distance Approach
The similarity between images is measured by computing distances between feature vectors in the feature space.
Given two feature vectors x and y and the weight vector w, we use the weighted distances L1 or L2:
Q
jjjj yxwwyxL
11 |)(|);,(
2/1
1
22 |)(|);,(
Q
jjjj yxwwyxL
Weighted Distance Approach
From the pattern recognition point of view, for a feature to be good, its variance among all the images in the database should be large but its variance among the relevant images should be small.
Any one of these is not enough alone but characterizes a good feature when combined with the other.
Weighted Distance Approach
Let denote the weight of the j-th feature component in the k+1 iteration.
This weight is given by the following equation:
kjw
kjrel
jkjw
,
0
where:
)( 00jj Fstd )( ,,
kjrel
kjrel Fstd
Weighted Distance Approach
According to the values of and there are four different cases:
kjrel ,0
j
best case
worst case
Weighted Distance Approach
Case 1 When is large and is small, becomes
large. This means that the feature has a diverse set of values
in the database but its values for relevant images are similar.
This is a desired situation and shows that this feature is very effective in distinguishing this specific relevant image set, so a large weight assigns more importance to this feature.
0j k
jrel , kjw
Weighted Distance Approach
Case 2 When both and are large, is close to 1. This means that the feature may have good
discrimination characteristics in the database but is not effective for this specific relevant image group.
The resulting weight does not give any particular importance to this feature.
0j k
jrel , kjw
Weighted Distance Approach
Case 3 When both and are small, is again close to
1. This is a similar but slightly worse situation than the
previous one. The feature is not generally effective in the database
and is not effective for this relevant set either. No importance is given to this feature.
0j k
jrel , kjw
Weighted Distance Approach
Case 4 When is small and is large, becomes small. This is the worst case among all the possibilities. The feature is not generally effective and even causes
the distance between relevant images to increase. A small weight forces the distance measure to ignore
the effect of this feature.
0j k
jrel , kjw
Weighted Distance Approach
Retrieval Algorithm[1] initialize all weights uniformly.[2] compute j = 1, 2, …, Q.[3] for k = 1, k <= K, k++
- search the DB using and retrieve - get feedback from user and populate
- compute j = 1, 2, …, Q
- compute j = 1, 2, …, Q
- normalize
Qw j /10 j = 1, 2, ..., Q0j
1kjw kR
krelR
kjrel ,
kjrel
jkjw
,
0
Q
j
kj
kjk
jw
ww
1
Weighted Distance Approach
Precision results
Weighted Distance Approach
Precision results
Bayesian Classification: a short tutorial
The problem of classification
Given a number of classes and an unclassified object x determine the class that x belongs to.
Examples: given the age and the income of a person determine if she will
buy a laptop or not given the color, type and origin of car determine if it will be
stolen given the age, income and job of a person determine if the bank
will give a loan or not.
In Bayesian Classification we use probabilities to determine the “best” class to assign a new item.
Bayesian Classification: a short tutorial
Bayes Theorem
)(
)()|()|(
xP
hPhxPxhP
P(h) prior probability of hypothesis h P(x) evidence of training data xP(h | x) probability of h given x (posterior probability)P(x | h) probability of x given h (likelihood)
x data itemh hypothesis
evidence
priorlikelihoodposterior
Bayesian Classification: a short tutorial
Each data item x is composed of several attributes. In our example we are interested in determining if a car with specific characteristics will be stolen or not.
Car attributes: color, type and origin.Given a color, type, origin triplet we are interested
in determining if the car will be stolen or not.
(Thanks to Eric Meisner for this example)
Bayesian Classification: a short tutorial
Training data
Bayesian Classification: a short tutorial
Naive Bayesian Classification
It is evident that each data item has several attributes (color, type, origin in our example). To calculate P(x | h) we use the independence assumption among different attributes.
Therefore:
j
j haPhxP )|()|(
Bayesian Classification: a short tutorial
The number of available classes are known. For each class ωi a discriminant function gi(x) is defined.
Item x is assigned to the k-th class when gk(x) > gj(x) for all j <> k.
In Bayesian Classification gi(x) is set to P(ωi | x).
Bayesian Classification: a short tutorial
Since for an item x the value of P(x) does not depend on the class, it can be eliminated without affecting the class ranking.
)()|()( iii PxPxg
))(ln())|(ln())()|(ln()( iiiii PxPPxPxg
Logarithms may also be used as follows:
(e.g., with Gaussian classifiers)
Bayesian Classification: a short tutorial
h1: the car will be stolen (1st hypothesis)
h2: the car will NOT be stolen (2nd hypothesis)
x: a red domestic sports car (color=“red”, type=“sports”, origin=“domestic”)
Determine if the car will be stolen or not by using a Naive Bayesian Classifier.
Bayesian Classification: a short tutorial
We need to calculate the following quantities: P(h1) : probability that a car will be stolen regardless of color, type and origin
(prior probability)P(h2) : probability that a car will not be stolen regardless of color, type and
origin (prior probability)P(x) : probability that a car from our set of cars is a red domestic sports car
(evidence)P(x | h1) : probability that the car has a red color, it is domestic and it is a sports
car, given that it is stolen (likelihood)P(x | h2) : probability that the car has a red color, it is domestic and it is a sports,
given that it is not stolen (likelihood)P(h1 | x): probability that car x will be stolen given that we know its color, type
and origin (posterior probability).P(h2 | x): probability that car x will not be stolen given that we know its color,
type and origin (posterior probability).
Bayesian Classification: a short tutorial
P(h1) = 5/10 = 0.5P(h2) = 5/10 = 0.5P(x) = 2/10 = 0.2
P(x | h1) = P(color=“red” | h1) x P(type=“sports” | h1) x P(origin=“domestic” | h1) = = 3/5 x 4/5 x 2/5 = 0.192
P(x | h2) = P(color=“red” | h2) x P(type=“sports” | h2) x P(origin=“domestic” | h2) = = 2/5 x 2/5 x 3/5 = 0.096
We need to calculate P(h1 | x) and P(h2 | x)
Bayesian Classification: a short tutorial
By substitution and calculations we get:
096.05.0192.0)()|()( 111 hPhxPxg
048.05.0096.0)()|()( 222 hPhxPxg
Since 0.096 > 0.048 we conclude that probably the car will be stolen.
RF with Bayesian Estimation
Zhong Su, Hongjiang Zhang, Stan Li, and Shaoping Ma. “Relevance Feedback in Content-Based Image Retrieval: Bayesian Framework, Feature Subspaces, and Progressive Learning”, IEEE Transactions on Image Processing, 2003.
RF with Bayesian Estimation
In the proposed relevance feedback approach, positive and negative feedback examples are incorporated in the query refinement process with different strategies.
To incorporate positive feedback in refining image retrieval, we assume that all of the positive examples in a feedback iteration belong to the same semantic class whose features follow a Gaussian distribution. Features of all positive examples are used to calculate and update the parameters of its corresponding semantic Gaussian class and we use a Bayesian classifier to re-rank the images in the database.
To incorporate negative feedback examples, we apply a penalty function in calculating the final ranking of an image to the query image. That is, if an image is similar to a negative example, its rank will be decreased depending on the degree of the similarity to the negative example.
RF with Bayesian Estimation
Low-level features used
Classifier Details
The Gaussian density is often used for characterizing probability because of its computational tractability and the fact that it adequately models a large number of cases.
The probability density function of the Gaussian distribution is:
)()(
2
1
2/12/
1
||2
1
xx
d
t
ep(x)
2
12
2
1)(
x
exp
univariate (x assumes single values)
multivariate (x is a vector)
x: variableμ: meanσ: standard deviation
x: variabled: dimensionalityμ: meanΣ: d x d cov matrix|Σ|: determinant of Σ
Classifier Details
Recall the classifier functions:
Assuming the measurements are normally distributed, we have:
By substitution we get:
)(ln)|(ln)( iii Pxpxg
)()(
2
1
2/12/
1
||2
1)|(
iit
i xx
idi exp
)(ln||ln2
1)()(
2
1)( 1
iiiit
ii Pxxxg
Positive Feedback
ix
in
D collection of images in the database
n number of positive examples
Q query image
m number of feature types used
feature vector for the i-th feature type
dimensionality of the i-th feature type
An image is represented as a vector
],...,,...,[ 1 mi xxxx
Positive Feedback
Each feature type is Gaussian distributed.
),(~ iii RxNx
iRis the ni x ni covariance matrix
ix is the ni dimensional mean vector
Positive Feedback
It is reasonable to assume that all the positive examples belong to the class of images containing the desired object or semantic meaning and the features of images belonging to the semantic classes obey the Gaussian distribution.
The parameters for a semantic Gaussian class can be estimated using the feature vectors of all the positive examples.
Hence, the image retrieval becomes a process of estimating the probability of belonging to a semantic class and the query refinement by relevance feedback becomes a process of updating the Gaussian distribution parameters.
Positive Feedback
U: the set of positive examples in the current iteration
|U|: number of positive examples in the current iteration
u: an item in U
The update of the Gaussian parameters is performed as follows:
n, n’: total number ofpositive examples accumulatedbefore and after the iteration
Positive Feedback Algorithm
[1] Feature normalization.
[2] Initialize: n = 1, = I, =
[3] User provides feedback and selects positive examples (set U).
[4] Update of retrieval parameters n, ,
[5] For image X in the database calculate the value of the discriminant function gi(x).
[6] A new ranking is returned to the user.
ixixi
iix
Negative Feedback
Negative examples are treated differently than positive ones.
Negative examples are often isolated and treated independently.
Only images near the negative examples are punished. The rest are not affected.
The penalty involves the increase of the distance by a constant factor (details in the paper).
Some Experimental Results
Corel image gallery has been used 10,000 images of 79 semantic categories have been selected The proposed techniques has been compared to two previously
proposed ones, termed Nuno’s approach and Rui’s approach (see bibliography)
The measurement of interest is the accuracy which is defined as follows:
accuracy = relevant images in top-k
k
Some Experimental Results
Some Experimental Results
Some Experimental Results
Some Experimental Results
PCA Feature Subspaces
The other major contribution in the proposed relevance feedback approach to content-based image retrieval is to apply the principal component analysis (PCA) technique to select and updated a proper feature subspace during the feedback process.
This algorithm extracts more effective, lower-dimensional features from the originally given ones, by constructing proper feature subspaces from the original spaces, to improve the retrieval performance in terms of speed, storage requirement and accuracy.
Some More Results
Some More Results
Some More Results