Development of Novel Image Compression Algorithms for Portable Multimedia Applications

Senapati, Ranjan Kumar (2012) Development of Novel Image Compression Algorithms for Portable Multimedia Applications. PhD thesis.



Portable multimedia devices such as digital camera, mobile d
evices, personal digtal assistants (PDAs), etc. have limited memory, battery life and processing power.
Real time processing and transmission using these devices requires image compression algorithms that can compress efficiently with reduced complexity. Due to limited
resources, it is not always possible to implement the best algorithms inside these devices. In uncompressed form, both raw and image data occupy an unreasonably large
space. However, both raw and image data have a significant amount of statistical and
visual redundancy. Consequently, the used storage space can be efficiently reduced by compression. In this thesis, some novel low complexity and embedded image compression algorithms are developed especially suitable for low bit rate image compression using these devices.
Despite the rapid progress in the Internet and multimedia technology, demand for data storage and data transmission bandwidth continues to outstrip the capabil-
ities of available technology. The browsing of images over In ternet from the image data sets using these devices requires fast encoding and decodin
g speed with better rate-distortion performance. With progressive picture build up of the wavelet based
coded images, the recent multimedia applications demand goo
d quality images at the earlier stages of transmission. This is particularly important if the image is browsed
over wireless lines where limited channel capacity, storage and computation are the deciding parameters. Unfortunately, the performance of JPEG codec degrades at low bit rates because of underlying block based DCT transforms. Altho
ugh wavelet based codecs provide substantial improvements in progressive picture quality at lower bit
rates, these coders do not fully exploit the coding performance at lower bit rates. It is evident from the statistics of transformed images that the number of significant coefficients having magnitude higher than earlier thresholds are very few. These wavelet
based codecs code zero to each insignificant subband as it moves from coarsest to finest subbands. It is also demonstrated that there could be six to sev
en bit plane passes where wavelet coders encode many zeros as many subbands are likely to be insignificant with respect to early thresholds. Bits indicating
insignificance of a coefficient or subband are required, but they don’t code information that reduces distortion
of the reconstructed image. This leads to reduction of zero distortion for an increase in non zero bit-rate.
Another problem associated with wavelet based coders such as Set partitioning in hierarchical trees (SPIHT), Set partitioning embedded block (SPECK), Wavelet
block-tree coding (WBTC) is because of the use of auxiliary lists. The size of list data structures increase exponentially as more and more eleme
nts are added, removed or moved in each bitplane pass. This increases the dynamic memory requirement
of the codec, which is a less efficient feature for hardware implementations. Later,
many listless variants of SPIHT and SPECK, e.g. No list SPIHT (NLS) and Listless SPECK (LSK) respectively are developed. However, these algorithms have similar
rate distortion performances, like the list based coders. An improved LSK (ILSK)algorithm proposed in this dissertation that improves the low b
it rate performance of LSK by encoding much lesser number of symbols (i.e. zeros) to several insignificant
subbands. Further, the ILSK is combined with a block based transform known as discrete Tchebichef transform (DTT). The proposed new coder isnamed as Hierar-chical listless DTT (HLDTT). DTT is chosen over DCT because of it’s
similar energy compaction property like discrete cosine transform (DCT). It is demonstrated that
the decoded image quality using HLDTT has better visual performance (i.e., Mean Structural Similarity) than the images decoded using DCT based embedded coders
in most of the bit rates.
The ILSK algorithm is also combined with Lift based wavelet tra nsform to show the superiority over JPEG2000 at lower rates in terms of peak signal-to-noise ratio
(PSNR). A full-scalable and random access decodable listless algorithm is also developed which is based on lift based ILSK. The proposed algorithm named as scalable
listless embedded block partitioning (S-LEBP) generates bit
stream that offer increasing signal-to-noise ratio and spatial resolution. These are very useful features for
transmission of images in a heterogeneous network that optimally service each user according to available bandwidth and computing needs. Random access decoding is a
very useful feature for extracting/manipulating certain ar
ea of an image with minimal decoding work. The idea used in ILSK is also extended to encode and decode color
images. The proposed algorithm for coding color images is named as Color listless embedded block partitioning (CLEBP) algorithm. The coding efficiency of CLEBP is compared with Color SPIHT (CSPIHT) and color variant of WBTC
algorithm. From the simulation results, it is shown that CLEBP exhibits a significant PSNR performance improvement over the later two algorithms on various types of images.
Although many modifications to NLS and LSK have been made, the listless modification to WBTC algorithm has not been reported in the literature. Therefore,a listless variant of WBTC (named as LBTC) algorithm is proposed. LBTC not
only reduces the memory requirement by 88-89% but also increases the encoding and decoding speed, while preserving the rate-distortion perform ance at the same time.
Further, the combination of DCT with LBTC (named as DCT
LBT) and DTT with LBTC (named as Hierarchical listless DTT, HLBTDTT) are compared with some state-of-the-art DCT based embedded coders. It is also shown that
the proposed DCT-LBT and HLBT-DTT show significant PSNR improvements over almost all the embedded coders in most of the bit rates.
In some multimedia applications e.g., digital camera, camco
rders etc., the images always need to have a fixed pre-determined high quality. The extra effort required for
quality scalability is wasted. Therefore, non-embedded algo
rithms are best suited for these applications. The proposed algorithms can be made non-embedded by encoding
a fixed set of bit planes at a time. Instead, a sparse orthogonal transform matrix is proposed, which can be integrated in a JEPG baseline coder. The proposed matrix
promises a substantial reduction in hardware complexity with amarginal loss of image quality on a considerable range of bit rates than block based DCT or Integer DCT.

Item Type:Thesis (PhD)
Uncontrolled Keywords:Novel Image Compression, Image representation,Digital Image,
Subjects:Engineering and Technology > Electronics and Communication Engineering > Image Processing
Divisions: Engineering and Technology > Department of Electronics and Communication Engineering
ID Code:5485
Deposited By:Hemanta Biswal
Deposited On:21 Mar 2014 10:57
Last Modified:21 Mar 2014 10:57
Supervisor(s):Pati, U C and Mahapatra, K K

Repository Staff Only: item control page