Adaptive block size selection in a hybrid image compression algorithm employing the DCT and SVD

Garg, Garima; Kumar, Raman

doi:10.2478/ijssis-2024-0005

Figures & Tables

Adopted methodology for image compression. DCT, discrete cosine transform; SVD, singular value decomposition.

Working flow of the proposed hybrid algorithm. DCT, discrete cosine transform; SVD, singular value decomposition.

Comparative analysis of compression ratios attained for several image datasets. DCT, discrete cosine transform; SVD, singular value decomposition

Comparative analysis of PSNR values attained. BSDS, Berkeley segmentation dataset; DCT, discrete cosine transform; PSNR, peak signal-to-noise ratio; SVD, singular value decomposition.

Visual comparison of original and compressed images for different techniques. DCT, discrete cosine transform; SVD, singular value decomposition.

Comparative analysis of CIDs attained for several image datasets. DCT, discrete cosine transform; SVD, singular value decomposition.

SSIM values obtained for several image datasets using the DCT-SVD hybrid technique

Dataset	SSIM
Kodak Lossless True Color Image Suite	0.93
Lena image	0.94
BSDS	0.89
ImageNet	0.92

Compression ratios attained for several image datasets using the DCT-SVD hybrid technique

Dataset	Compression ratio
Kodak Lossless True Color Image Suite	58.34
Lena image	63.12
BSDS	55.76
ImageNet	57.89

CIDs obtained for several image datasets using the DCT-SVD hybrid technique

Dataset	CID
Kodak lossless true color image suite	0.86
Lena image	0.82
BSDS	0.89
ImageNet	0.87

Summary of recent work on image compression

Ref. No.	Dataset used	Adopted methodology	Techniques used	Advantages	Disadvantages	Solutions
[18]	Kodak dataset	Adaptive block size selection and DCT-SVD hybrid	DCT, SVD, and adaptive processing	High compression and good quality	Complexity in hybridization	Adaptive hybridization
[19]	UCID dataset	Wavelet transform	Wavelet transform	Multiresolution representation	Limited to certain images	Improved wavelet selection
[20]	CALTECH dataset	Huffman coding	Huffman coding	No quality loss	Limited compression ratio	Enhanced entropy coding
[21]	ImageNet dataset	DCT-based compression	Discrete cosine transform	Established standard	Lossy compression	Improved quantization
[22]	Custom dataset	Iterated function system	Fractal encoding	Good compression	Iteration limits	Adaptive fractal generation
[23]	MNIST dataset	DCT-DWT hybrid	DCT and DWT	Multifrequency representation	High computational cost	Improved parallel processing
[24]	COCO dataset	Singular value decomposition	Singular value decomposition	Noise robustness	Singular value truncation	Adaptive truncation threshold
[25]	CIFAR-10 dataset	Neural network-based approach	Neural networks	Adaptive learning	Training complexity	Improved model architecture
[26]	ImageNet dataset	Contextual analysis	Contextual processing	Improved quality	Complexity	Efficient context modeling
[27]	Medical images	Adaptive block size selection and transform coding	DCT and Huffman coding	Lossless compression	Limited to medical images	Improved coding strategies
[28]	Custom dataset	Vector quantization	Vector quantization	High compression ratios	Information loss	Enhanced vector codebooks
[29]	COCO dataset	Adaptive processing based on content	DCT and adaptive strategies	Improved quality and efficient compression	Complexity in content analysis	Enhanced adaptive strategies
[30]	ImageNet dataset	Pyramid-based compression	Pyramid transform	Multiresolution representation	Complexity	Optimized pyramid levels
[31]	Kodak dataset	Progressive compression approach	DCT and SVD	Stepwise quality enhancement	Progressive transmission complexity	Improved transmission order
[32]	CALTECH dataset	Block-based processing and Huffman coding	Block processing and Huffman coding	Balanced quality compression	Block artifacts	Enhanced block processing
[33]	ImageNet dataset	Simultaneous compression and encryption	DCT and encryption techniques	Secure compression	Increased complexity	Improved encryption algorithms
[34]	Custom dataset	Arithmetic coding	Arithmetic coding	High compression and lossless compression	Complexity	Enhanced probability modeling
[35]	CIFAR-10 dataset	DCT–neural network hybrid	DCT and neural networks	Adaptive compression and improved quality	Training complexity	Enhanced training strategies
[36]	COCO dataset	Wavelet transform	Wavelet transform	Multifrequency representation	Complexity	Enhanced transform selection
[37]	Custom dataset	Contextual Huffman coding	Contextual analysis and Huffman coding	Improved compression	Complexity	Enhanced context modeling
[38]	ImageNet dataset	Multiresolution encoding	Discrete wavelet transform	Progressive quality and multiresolution	Complexity	Adaptive wavelet selection

PSNR values attained for several image datasets using the DCT-SVD hybrid technique

Dataset	PSNR (dB)
Kodak Lossless True Color Image Suite	38.21
Lena image	39.08
BSDS	36.75
ImageNet	37.52

Dataset used for experimentation

Dataset name	Number of images	Image types	Resolution	Content complexity
Kodak Lossless True Color Image Suite	24	Natural sceneries	Varied	Moderate
Lena image	1	Portrait	512 × 512	Moderate
BSDS	200	Natural sceneries	Varied	High
ImageNet	1000	Various	Varied	High

Adaptive block size selection in a hybrid image compression algorithm employing the DCT and SVD

Figures & Tables

Figure 1:

Figure 2:

Figure 3:

Figure 4:

Figure 5:

Figure 6:

Figure 7:

SSIM values obtained for several image datasets using the DCT-SVD hybrid technique

Compression ratios attained for several image datasets using the DCT-SVD hybrid technique

CIDs obtained for several image datasets using the DCT-SVD hybrid technique

Summary of recent work on image compression

PSNR values attained for several image datasets using the DCT-SVD hybrid technique

Dataset used for experimentation

Paradigm

My account