2. Unlike the standard wavelet decomposition, which gives a logarithmic
frequency resolution, the M-band decomposition gives a mixture of a
logarithmic and linear frequency resolution.
Most texture image retrieval systems are still incapable of providing retrieval
result with high retrieval accuracy and less computational complexity.
3. Visual information (images/ video) is one of the most promising sources
of multimedia information, as it plays a key role in the communication
framework.
The term [CBIR] describes the process of retrieving desired images from a
large collection on the basis of features (such as colour, texture and shape)
that can be automatically extracted from the images themselves.
The main advantage of CBIR is its ability to support visual queries . The
challenge in CBIR is to develop the methods that will increase retrieval
accuracy and reduce the retrieval time (computational complexity).
5. A CBIR system consists of two databases namely an Image Database and
Image Feature Database.
The image database contains the original images present in the database.
Similarity between query image and each database image is calculated by
finding the distance between the feature vectors.
The Feature Extraction module processes each of the database images to
extract a description of the content of the image, represented in the form of a
vector called feature vector.
6. There are two important tasks in content-based image retrieval. First one
is feature extraction, and second one is similarity measurement. Our
research is focused on these two important tasks.
This motivates us to explore different similarity measures and different
wavelet based features , which will improve retrieval effectiveness both in
terms of retrieval accuracy
and retrieval time.
A successful CBIR system must be able to deal with textured images in real
world. The majority of existing texture feature extraction methods for CBIR
assumes that all images are acquired from the same viewpoint. This
assumption is not realistic in practical applications.
7. Texture can be defined as, “A region in an image has a constant texture if a set
of local statistics or other local properties of the picture are constant, slowly
varying, or approximately periodic”.
Texture features currently used in CBIR are mainly derived from Gabor
wavelets [62], the conventional discrete wavelet transform (DWT) [53], tree
structured wavelet transform [7], and wavelet frame [91].
Texture feature extraction with DWT gives the edge information in the
horizontal, vertical and diagonal direction.
Texture representation with the real DWT has two main disadvantages of
shift sensitivity and poor directionality (only three directions information).
8. Art Collections
e.g. Fine Arts Museum of San Francisco
Medical Image Databases :CT, MRI, Ultrasound, The Visible Human
Scientific Databases
e.g. Earth Sciences
General Image Collections for Licensing :Corbis, Getty Images
The World Wide Web
Automatic face recognition
9. Color (histograms, gridded layout, wavelets)
Texture (Laws, Gabor filters, local binary partition)
Shape (first segment the image, then use statistical or structural shape
similarity measures)
Objects and their Relationships
10. To avoid the problem of pixel-by-pixel comparison next abstraction level that is
used for representing images is the feature level.
Feature extraction plays an important role in content-based image retrieval to
support efficient and fast retrieval of similar images from image databases.
Significant features should be extracted from image data. Every image is
characterized by a set of features such as texture, color, shape, spatial location,
image semantic features etc.
These features are extracted at the time of injecting new image in image database
and stored in image feature database.
11. Average Retrieval Time:
A new optimization criterion for locating emergency medical care facilities,
level-load retrieval time, is described and applied to Los Angeles County.
The new criterion combines retrieval times from demand points served by each
facility and patient load on each facility
12. The main drawbacks of standard wavelets is that they are not suitable for the
analysis of high- frequency signals with relatively narrow bandwidth. Also the
standard wavelet decomposition gives a logarithmic frequency resolution.
M-Band wavelet on the other hand has two main advantages over the
standard wavelet.
M-band wavelet gives better spectral decomposition for texture images than
standard wavelet, because M-band wavelet decomposition gives a mixture of a
logarithmic and linear frequency resolution.
M-band wavelet decomposition yields a large number of sub bands, which
improves the retrieval accuracy.
The limitation of M–band wavelet is that the computational complexity
increases and hence retrieval time increases with number of bands.
14. The filters hi (n) are analysis filters constituting the analysis filter bank and
the filters gi (n) are the synthesis filters constituting the synthesis filter bank.
Perfect reconstruction of the signal is an important requirement of M-
Channel filter bank. Filter bank is said to be perfect reconstruction if
y(n) = x(n).
15. Disadvantage of using standard wavelets is that they are not suitable for the
analysis of high-frequency signals with relatively narrow bandwidth.
The M-band orthonormal wavelets give a better energy compaction than two
band wavelets by zooming into narrow band high frequency components of a
signal
In M-band wavelets there are M-1 wavelets
17. In the filtering stage we make use of biorthonormal M-band wavelet
transform [94] to decompose the texture image into M ×M -channels,
corresponding to different direction and resolutions.
At each level with M=3, the image is decomposed in to
M ×M (=9) channels. Table 3.1 shows the 3-band wavelet filter coefficients
[120] used in the experiments.
18. The cosine –modulated FIR filter banks are the special class of unitary filter banks,
where the analysis filters hi (n) are all cosine-modulates of a low pass linear-phase
prototype filter g(n).
The fundamental idea behind cosine-modulated filter banks is the following: In
an M-channel filter bank, the analysis and synthesis filters are meant to approximate
ideal M th band filters.
In the filtering stage we make use of filter coefficients for M =2 to decompose the
texture image in to four channels, corresponding to different direction and
resolutions.
After decomposing image with wavelet transform we get horizontal, vertical and
diagonal information. Hsin has reported that diagonal filter gives strong response
to textures with orientations at or close to ± 45° .
19. Subsequently the decomposition was performed column wise.
Thus at the first level of decomposition the original image was decomposed
into M 2 = 9 sub-images.
This would correspond to the decomposition of upper left-hand corner sub-
band of the frequency plane called a complete decomposition.
In general we obtain M 2n sub-bands at the nth level of decomposition.
Rotate these sub bands by +45 deg, we will get the information in directions of
0,45,90,135 degrees.
Calculate the energy for all sub bands and from the feature vector.
21. The objective of these experiments is to illustrate that the proposed texture
features for CBIR using M-band wavelet and cosine modulated wavelet provides
equally better retrieval accuracy to that of the Gabor wavelet based method
along with much reduced retrieval time.
Average retrieval performance with M-Band wavelet (73.65 %) is better than
standard wavelet (71.71 %).
Average retrieval performance of cosine-modulated wavelet is 74.78% and it is
better than standard wavelet.
M-Band wavelet and is marginally better than that in case of Gabor wavelet
method (74.32%) proposed by Ma and Manjunath.
23. In terms of feature extraction time for query image, the Gabor wavelet
is most expensive.
Computational complexity of M-Band wavelet is more as compared to standard wavelet and
cosine-modulated wavelet but five times less as compared to Gabor wavelet.
24. The retrieval performance of M-Band wavelet is consistently superior to
standard wavelet.
If the top 116 (6% of the database) retrievals are considered the performance
increases up to 91.65%, 94.07 %, 94.77%, and 92.375% using, standard wavelet,
M-Band wavelet, cosine-modulated wavelet, and Gabor wavelet respectively.
26. The analysis was performed up to second level (9×2=18 sub bands) of the
wavelet decomposition.
The approach is partly supported by physiological studies of the visual cortex
as reported by Hubel and Wiesel and Daugman .
The energy and standard deviation of decomposed sub bands are computed
as follows:
M N
1
Energy Ek W ij
M N i 1 j 1
1/ 2
M N
1 2
Standard D eviation k
(W ij ij
)
M N i 1 j 1
where W ij is the wavelet-decomposed sub band, M×N is the size of wavelet-
decomposed sub band, k is the number of sub bands (k=18 for two levels), and ij
is the sub band mean value.
28. The performance of the proposed method is tested by conducting
experimentation on Brodatz database.
The results after being investigated show a significant improvement in terms
of average retrieval rate and average retrieval precision as compared to
M_band_DWT, M_band_RWT and other existing transform domain
techniques.