21-08-2014, 12:52 PM
A Survey on Hybrid Image Compression Techniques
for Video Transmission
A Survey on Hybrid Image.pdf (Size: 43.01 KB / Downloads: 56)
Abstract
With the advent of internet, large number of images is transmitted. Memory
space and channel capacity are the major challenges during image
transmission. Hence image compression plays a major role during image
transmission. Considerable research is carried out in the literature in both
spatial and transform domain. Performance metrics used are Peak Signal to
Noise Ratio (PSNR) and Compression Ratio (CR). In this paper an extensive
literature survey is done on various image compression techniques.
Keywords: DCT, DWT, Neural Network, Huffman Coding, SPIHT,
Quantization
I. INTRODUCTION
Image Compression aims at removing coding, inter pixel and psycho visual
redundancies. In order to remove coding redundancy, different symbol encoders
namely Huffman Coding, Run length Coding and Arithmetic Coding are used. LZW
coding is used for removing interpixel redundancy. Improved Gray Scale
Quantization is used for removing psycho visual redundancy. While the former two
techniques result in lossless compression, the later leads to lossy compression. Image
compression can be performed directly in the spatial domain or in the transform
domain. Transform coding techniques involve sub image separation, forward
transformation, Quantization and symbol encoding at the transmitter’s side
II. DWT BASED IMAGE COMPRESSION TECHNIQUES
Prasanthi Jasmine et al (2012) proposed wavelet and ridgelet based compression
methods. Methodology involved is: the RGB image is converted to gray scale and is
de-noised with Gaussian filter; Discrete Wavelet Transform (DWT) is performed on
the de-noised image; Finite Ridgelet Transform (FRT) is employed to obtained
wavelet coefficients; compressed image of reduced size is obtained; decompression is
done by applying Inverse FRT and DWT and the original image is obtained without
loss of data. This hybrid image compression technique results in compression of the
image in an effective manner without losing data. Indrit Enesi (2012) proposed a
combination of wavelet technique with algebraic Generalized Principal Component
Analysis (GPCA) that provides compression of multimedia information without
reducing its quality. The proposed algorithm is as follows: Load the RGB image;
wavelet transform is applied to the image; the approximation co-efficient will be
decomposed into a sub-band tree; Hybrid Linear Modeling is performed on the
approximation coefficients; Entropy encoding is executed to obtain the compressed
image; Reconstruction of the image is done by reverse process. Performance of the
proposed method is better than the classic wavelet method and achieves a higher
performance. PSNR-values were found to be 15% larger.
III. HYBRID IMAGE COMPRESSION TECHNIQUES USING
DCT AND DWT
Aree Ali Mohammed and Jamal Ali Hussein (2011) presented a scheme for medical
image compression based on hybrid compression technique (DWT and DCT) to
achieve higher compression rates. The proposed technique is as follows: Load the
RGB image and convert into YCbCr; to obtain approximate 8x8 coefficient bands
apply Forward Discrete Wavelet Transform (FDWT); Perform Forward Discrete
Cosine Transform (FDCT) on the image and apply DCT and DWT quantization;
Discrete Pulse Code Modulation (DPCM) is implemented to convert the bands into
positive values and Variable Shift Coding algorithm is employed; Reconstruction is
done by the reverse procedure. Experimental results show that these images preserve
its quality where quantization factor is less than 0.5. Parveen Banu and Venkatramani
(2011) proposed a hybrid image compression scheme which comprises of three
techniques for efficient storage and delivery of data. The original color image is
converted into luminance and chrominance components.
IV. SPIHT BASED COMPRESSION TECHNIQUES
Prathyusha Reddi et al (2013) proposed a new image compression scheme by
combining Hyper analytical Wavelet Transform (HWT) and Set Partitioning in
Hierarchical Tree (SPIHT) which resulted in appreciable increase in PSNR and
compression ratio. The proposed algorithm comprises of the following steps: Source
image is converted into a hyper analytical image by Hilbert Transform; each
component of the image is decomposed into wavelet coefficients by 2D DWT
method; Encoding is done by using SPIHT technique to achieve desired compression
ratio; Reconstruction of the image is obtained by the reverse process. As a result the
combination of HWT and SPIHT produces better quality of reconstructed images
when compared with the combination of DWT and SPIHT. Salija et al (2013)
proposed a technique to achieve high compression ratio by using block based seam
carving with hybrid transform and SPIHT algorithm. The stages involved are: The
RGB image is input and converted into YCbCr format; Image analysis is performed to
extract the Region of interest (ROI); manually define the region and sharpen that ROI
region by using filter to give contrast to ROI and high weighing factor is given to ROI
is to get high energy value; DWT is performed on the carved images; DCT is applied
to the wavelet coefficients; SPIHT is used for coding the transformed coefficients;
Recovery of image can be done by applying the reverse process. This method is not
only efficient for obtaining high compression ratio but also to obtain images with high
quality in given bitrate with less complexity. Also it provides good quality and
efficient method to avoid duplication of data with less complexity and storage space.
V. IMAGE COMPRESSION USING NEURAL NETWORKS
Sridhar et al (2013) illustrated a wavelet transform and neural network based model
for image compression. The demonstrated technique comprises of the following steps:
Store a color image of a moderate size; Discrete Wavelet Transform (DWT) is used to
decompose the image to obtain approximation coefficients; the coefficient bands are
compressed using DPCM and Neural Network techniques; Huffman Coding is
performed on the bit stream to obtain the compressed image; Reconstruction is done
by the reverse process. This illustrated technique results in improved quality of
reconstructed images and eliminates blocking effects associated with DCT. Moreover
it can be used in Bar code creation and can also be used in various fields like space,
medical, defense and many more.
Abdul Khader Jilani Saudagar and Abdul Sattar Syed (2013) proposed Image
compression by transforming the image into another domain with ridgelet function
and then quantizing the coefficients. The methodology is as follows: Decompose the
input data into a set of wavelets bands into smooth blocks of side length 1; Non
redundancy transformation is achieved by the FRIT; In Finite RAdon Transform
(FRAT), 2-D wavelet transform is performed; high-pass filter and a low-pass filter are
applied to the approximation bands and smooth partitioning is done on each window;
FRIT is performed on each window; hybrid neural network with Back propagation
algorithm is carried out on the given input; Reconstruction of the image is done by
inverse process. It is observed that proposed algorithm is able to achieve good quality
performance with a simple algorithm. Also it does not require complicated bit
allocation procedures.
V. IMAGE COMPRESSION USING NEURAL NETWORKS
Sridhar et al (2013) illustrated a wavelet transform and neural network based model
for image compression. The demonstrated technique comprises of the following steps:
Store a color image of a moderate size; Discrete Wavelet Transform (DWT) is used to
decompose the image to obtain approximation coefficients; the coefficient bands are
compressed using DPCM and Neural Network techniques; Huffman Coding is
performed on the bit stream to obtain the compressed image; Reconstruction is done
by the reverse process. This illustrated technique results in improved quality of
reconstructed images and eliminates blocking effects associated with DCT. Moreover
it can be used in Bar code creation and can also be used in various fields like space,
medical, defense and many more.
Abdul Khader Jilani Saudagar and Abdul Sattar Syed (2013) proposed Image
compression by transforming the image into another domain with ridgelet function
and then quantizing the coefficients. The methodology is as follows: Decompose the
input data into a set of wavelets bands into smooth blocks of side length 1; Non
redundancy transformation is achieved by the FRIT; In Finite RAdon
Transform
(FRAT), 2-D wavelet transform is performed; high-pass filter and a low-pass filter are
applied to the approximation bands and smooth partitioning is done on each window;
FRIT is performed on each window; hybrid neural network with Back propagation
algorithm is carried out on the given input; Reconstruction of the image is done by
inverse process. It is observed that proposed algorithm is able to achieve good quality
performance with a simple algorithm. Also it does not require complicated bit
allocation procedures.
V. IMAGE COMPRESSION USING NEURAL NETWORKS
Sridhar et al (2013) illustrated a wavelet transform and neural network based model
for image compression. The demonstrated technique comprises of the following steps:
Store a color image of a moderate size; Discrete Wavelet Transform (DWT) is used to
decompose the image to obtain approximation coefficients; the coefficient bands are
compressed using DPCM and Neural Network techniques; Huffman Coding is
performed on the bit stream to obtain the compressed image; Reconstruction is done
by the reverse process. This illustrated technique results in improved quality of
reconstructed images and eliminates blocking effects associated with DCT. Moreover
it can be used in Bar code creation and can also be used in various fields like space,
medical, defense and many more.
Abdul Khader Jilani Saudagar and Abdul Sattar Syed (2013) proposed Image
compression by transforming the image into another domain with ridgelet function
and then quantizing the coefficients. The methodology is as follows: Decompose the
input data into a set of wavelets bands into smooth blocks of side length 1; Non
redundancy transformation is achieved by the FRIT; In Finite RAdon Transform
(FRAT), 2-D wavelet transform is performed; high-pass filter and a low-pass filter are
applied to the approximation bands and smooth partitioning is done on each window;
FRIT is performed on each window; hybrid neural network with Back propagation
algorithm is carried out on the given input; Reconstruction of the image is done by
inverse process. It is observed that proposed algorithm is able to achieve good quality
performance with a simple algorithm. Also it does not require complicated bit
allocation procedures.
VI. CONCLUSION
An extensive literature survey on various lossy image compression techniques is
performed in this paper. From the survey, the major conclusions are as follows: DCTDWT and SPIHT provides higher compression ratio and good quality output images.
However the performance of these techniques is affected in the presence of noise. Soft
computing based compression technique works well in robust environment and
provides higher compression ratio and higher PSNR and good quality. Neural
Network algorithm results in improved quality of reconstructed images as it
eliminates blocking effects associated with DCT. Moreover it can be used in Bar code
creation and can also be used in various fields like space, medical, defense and many