IMAGE COMPRESSION USING SOFM AND SPIHT WITH
WAVELET
PROJECT GUIDE:
Mr.Sivaprakash M.E.,
Assistant professor.
Submitted by
M.Aarathy
Presentation Details Abstract Image Compression Project Description Wavelet Transform Sub band Decomposition Vector Quantization Self Organizing Feature Map of Kohonen Image Reconstruction Bibliography
i) Abstract The main objective of this project is to implement
the concept of wavelet based compression to gray scale images using two different techniques namely SOFM and SPIHT
Wavelet Transform is a superior approach to other time frequency analysis tools because its time scale width of the window can be stretched to match the original signal especially in image analysis.
It is more advantageous than the Fourier transform
By using SOFM technique,we have made an attempt in employing lossy technique i.e., Vector Quantisation to encode the sub bands formed by the application of wavelet Transform.
ii
In the second Technique,ROI coding functionality is incorporated with the set partitioning in hierarchical trees algorithm for wavelet based image coding.
Both the Compression Techniques use wavelet transform output as the input for SOFM and SPIHT encoding.
Image Compression Image compression operation reduce the data
content of a digital image and represent the image in more compact form,usually before storage or transmission.
Compression Techniques are classified as • Loss less• Lossy
Lossy Compression results in the decompressed image being similar but not the same as the original image.
Much higher compression is achievable,and under normal viewing conditions,no visible loss is perceived
Project Description The Main Objective of this project is to implement
the concept of wavelet based compression to gray scale images.Vector quantisation is used to encode the sub bands formed by the application of wavelet Transform
We have also used a clustering property of self organizing Feature Map of Kohonen,an unsupervised training algorithm formulated by Kohonen.
Sofm serves as a tool for selecting the best vectors as they are being trained and the codebooks are formed using the trained vectors.Instead of storing the grayscale image,we store only the codebook and their corresponding index values.This reduces the space required to store the image,hence the compression of the image is achieved
Block Diagram representation of the compression algorithm
Input Image
Sub band Decomposition using wavelet
Vector Quantisation of the Sub bands
Code book formation using SOFM
a
Storage of the codebook and their indices
a
Mapping of index values with the code vectors
Application of IDWT on the index mapped code vectors
Arranging the sub bands in proper order
Reconstructed image
Wavelet Transform A wavelet is a waveform of effectively limited
duration. Wavelet is a small wave,which has its energy
concentrated in time to give a tool for the analysis of time varying phenomena.wavelets are suited to modeling phenomena whose signals are not continuous.
wavelets are well suited for approximating data with sharp discontinuities.
Wavelets not only have an oscillating characteristic but also have the ability to allow simultaneous time and frequency analysis with a flexible mathematical foundation.
According to wavelet transformation,a function,which can represent an image,a curve,signal etc.,can be described in terms of a coarse level description in addition to others with details
Wavelets are constructed by considering a complex valued window function (t) called the Mother Wavelet or a Basic Wavelet.
The compressed version packs all its oscillation in a small interval while the stretched version spreads them.
Discrete Wavelet Transform: The Discrete wavelet Transform of a finite
length signal x(n) having N components.Each wavelet coefficient represents information in a certain frequency range at a certain spatial location.
Its basis function is a scale varying function,which can be used to extract information from a given function at different scales.
The important application of wavelets is separating the smooth variations and details of the image,which is done by wavelet decomposition of the image using DWT.
Advantages of DWT: It is fast,linear in its operation. Invertible and orthogonal,hence
reconstruction is easier Window size is variable They are capable of providing the time and
frequency information simultaneously.
Wavelet Filter Coefficients: A Particular set of wavelets is specified by the
particular set of numbers called wavelet filter coefficients.
Any input signal f(t) can be expressed in the notation for wavelet transform as:
f(t)= cj(k) (2j t-k) + dj (2j t-k)
where,
cj(k) are the approximation coefficients
dj(k) are the detail coefficients
(t) is the scaling function
(t) is the wavelet function
Haar Scaling Function:
The Haar scaling function is the simple unit-width,unit-length,pulse-function can be used to construct (t) by
(t)= (2t)+ (2t-1) with the scaling coefficient h(n)
1 0 t
Sub band Decomposition Images being a two dimensional matrix,filtering
is applied to both horizontal and vertical elements of the image matrix.
To begin the decomposition, the input image is divided into sub bands and sub sampled.Each coefficient represents a spatial area corresponding to approximately a 2X2 area of the original image.
The total number of components we have after vertical and horizontal decompositions is four.These components are referred as sub bands.
Sub bands arranged after two level
decomposition
LL LH
HL HH
LL LH
HL HH
Parent
Children
Schematic of a vector quantisation block
Group into Vectorblocks
Input image Sub band
Find ClosestCode-Vector
--- Look up Table
--
Reconstruction
Unblock
Decoder
Encodercodebook
Index
codebook Index
Advantage and Disadvantage of VQ
Advantages: For a given rate,use of Vector Quantisation
results in a lower distortion Higher compression is achieved
Disadvantage: Changing either the block size or the codebook
size will allow the compression ratio to vary,but involves the training and storage of many code books.
Large memory is required
Self Organizing Feature Map Of Kohonen Self Organizing Feature Map (SOFM) developed
by Kohonen is an unsupervised training algorithm.
In unsupervised learning,the NET seeks to find patterns of regularity of the input data without the aid of the tutor.
We have used to train the vectors formed from each sub band and to select the best vector to form the code book.
The new weight vectors can be found by
Wj(new)=Wj (old)+ [x-wj(old)] X----input vector
wj----weight vector for unit j ---Learning rate coefficient
Image ReconstructionCodebook Remapping: In the reconstruction segment i.e., at the
decoder or decompressor,the code vectors are arranged according to the indices.
The resulting output will be an approximation of the input image.In this approximate image,the sub bands are only obtained.To get the reconstructed image,the approximate image should be subjected to IDWT
Reconstruction using IDWT: Reconstruction of the original image
coefficients can be obtained from a combination of the scaling function and wavelet coefficient.
The filter pair used here is called as synthesis filter.
Bibliography Books: 1.A primer on wavelets by James.s.Walker. 2. Digital Image Processing by Gonzalez. 3. Neural Networks and Application by Lauren Facet.
Website: www.wavelet.org www.sanbi.ac.za/tdrcourse/materials.html
IEEE Reference:• “ Initialization and Training Methods for Kohonen
Self Organizing Feature Map in Image Quantization” by Xiao Rei, chip-Hong Chang.
• “Image Compression by Vector Quantisation”by Robert.S.H
Wavelet Transform Output
Input image Single Decomposition output
Compression Output
Literature SurveyTill Date: