Image Compression: Efficient Techniques for Visual Data Optimization

By Fouad Sabry

What is Image Compression

When applied to digital photographs, image compression is a form of data compression that helps to reduce the amount of money that is required for their storage or transmission. It is possible for algorithms to make use of visual perception and the statistical aspects of picture data in order to provide higher outcomes when compared to generic data compression approaches that are utilized for other types of digital data.

How you will benefit

(I) Insights, and validations about the following topics:

Chapter 1: Image compression

Chapter 2: Data compression

Chapter 3: JPEG

Chapter 4: Lossy compression

Chapter 5: Lossless compression

Chapter 6: PNG

Chapter 7: Transform coding

Chapter 8: Discrete cosine transform

Chapter 9: JPEG 2000

Chapter 10: Compression artifact

(II) Answering the public top questions about image compression.

(III) Real world examples for the usage of image compression in many fields.

Who this book is for

Professionals, undergraduate and graduate students, enthusiasts, hobbyists, and those who want to go beyond basic knowledge or information for any kind of Image Compression.

• Language: English
• Publisher: One Billion Knowledgeable
• Release date: Apr 29, 2024

Book preview

Chapter 1: Image compression

Digital photographs can be compressed using image compression, a form of data compression, to lower their size and thus their storage and transfer costs. In order to get better outcomes than generic data compression methods used for other digital data, algorithms may exploit visual perception and the statistical features of image data.

Lossy and lossless image compression are both possible. The majority of medical images, technical drawings, clip art, and cartoons benefit from lossless compression for long-term storage. Compression artifacts are introduced by lossy compression techniques, especially when working with low bit rates. When a significant reduction in bit rate is desired but a small (often undetectable) loss of quality is acceptable, lossy approaches are ideal. This is the case with natural pictures like photographs. Visually lossless compression uses a form of lossy compression that yields imperceptible quality losses.

Compression techniques with loss:

The most used technique is transform coding.

The most popular type of lossy compression is the Discrete Cosine Transform (DCT). Invented by Nasir Ahmed, T. Natarajan, and K. R. Rao in 1974, it belongs to a family of transforms that includes the Fourier transform. In the context of a group of discrete cosine transforms, the DCT is frequently referred to as DCT-II (see discrete cosine transform). It is the most effective method of image compression in most cases.

The most widely used lossy format, JPEG, uses DCT, as does the more modern HEIF.

Quantization and entropy coding come next, then the more contemporary wavelet transform.

Color quantization is the process of reducing an image's color palette to a small set of representative hues. The compressed image's header includes a color palette that specifies the colors that were used to create the image. The color index is simply referred to in each pixel. In order to prevent posterization, this technique can be used in tandem with dithering.

Full color palette, commonly used in GIF and PNG files, with a maximum of 256 colors.

In BTC, CCC, S2TC, and S3TC, a block palette of 2 or 4 colors each 4x4 pixel block is employed.

Color space reduction by chroma subsampling. By averaging or eliminating some of the chrominance information in the image, this takes use of the fact that the human eye registers spatial changes of brightness more acutely than those of color.

Fractal compression.

Multilayer perceptrons, Convolutional neural networks, and Generative adversarial networks, all from the realm of Machine Learning, have been used in more contemporary applications.

Lossless compression techniques:

PCX uses run-length encoding by default, whereas BMP, TGA, and TIFF all support it as an alternative mode of encoding.

Area image compression

DPCM makes use of predictive coding.

Mathematical encoding and Huffman encoding are the two most popular forms of entropy encoding.

LZW, a popular adaptive dictionary method used in GIF and TIFF, Use DEFLATE in PNG, MNG, and TIFF files.

Chain codes

Diffusion models

The primary objective of picture compression is to get the highest possible quality at a given compression rate (or bit rate):

In most cases, reducing scalability means sacrificing quality by manipulating the underlying bitstream or file (without decompression and re-compression). Progressive encoding and embedded bitstreams are other names for scalability. Lossless codecs also have scalability, however it takes a different shape (often a scan from coarse to fine pixels) and are used for other purposes. Previewing images while download (in a web browser, for example) or offering varying quality access to resources (such as databases) both benefit greatly from scalability. Different kinds of scalability exist:

Progressive escalation of quality or complexity: Each iteration of the bitstream improves the quality of the rebuilt image.

To encode a higher quality version of an image, the lower resolution version must be encoded first.

Gradual encoding of color from a previously encoded black and white version.

Coding by area of interest. Different regions of the image are encoded at different quality levels. The capacity to scale could be added to this (encode these parts first, others later).

Data about data. Images can be browsed, searched, and categorized with the help of metadata that may be present in compressed data. Color and texture data, preview images, and author/copyright information are all examples of this type of data.

Capacity to compute. Compression methods have varying encoding and decoding computational demands. It takes a lot of cpu resources to run some high compression methods.

The peak signal-to-noise ratio is a common metric used to evaluate the efficacy of a compression technique. It quantifies the amount of noise created by the lossy compression of an image, but the viewer's subjective evaluation is sometimes considered just as, if not more, essential.

Shannon-Fano coding, the precursor of modern entropy coding, was developed in the late 1940s, Data Compression for Images".

In January 1974, IEEE Trans. Computers published Discrete Cosine Transform by Nasir Ahmed, T. Natarajan, and K. R. Rao on pages 90–93.

Maayan, Gilad David (Nov 24, 2021). State of the Art in Artificial Intelligence-Based Image Compression. Towards the Science of Data. Extracted on April 6th, 2023.

High Quality Generative Image Compression. Extracted on April 6th, 2023.

^ Bühlmann, Matthias (2022-09-28).

Compressing images using a stable diffusion method.

Medium.

Retrieved 2022-11-02.

Burt, P.; Adelson, E. (1 April 1983). The Laplacian pyramid is a condensed image code. Communications, IEEE, 31(4), pp.532-540. Doi:10.1109/TCOM.1983.1095851; CiteSeerX:10.1.1.54.299; S2CID:8018433.

^ Shao, Dan; Kropatsch, Walter G.

(February 3–5, 2010).

Špaček, Libor; Franc, Vojtěch (eds.).

Pyramid of Irregular Laplacian Graphs (PDF).

Computer Vision Winter Workshop 2010.

Nové Hrady, Czech Republic: Czech Pattern Recognition Society.

Date of initial publication: May 27, 2013 (PDF).

Thanks to: Claude Elwood Shannon (1948). Alcatel-Lucent (ed.). The Mathematical Theory of Interpersonal Interaction (PDF). 27 (3-4) Bell System Technical Journal, pages 379–423, 623–656. References: hdl:11858/00-001M-0000-002C-4314-2; doi:10.1002/j.1538-7305.1948.tb01338.x. The original (PDF) can be accessed from May 24, 2011. Date Accessed: 21 April 2019.

The original (PDF) version of A method for the creation of minimum-redundancy codes by David Albert Huffman was archived on 2005-10-08 from Proceedings of the IRE, vol. 40, no. 9, pp. 1098-1101 (doi:10.1109/JRPROC.1952.273898).

Reference: Hadamard transform image coding, William K. Pratt, Julius Kane, and Harry C. Andrews, Proceedings of the IEEE 57.1 (1969), pages 58-68.

Nasir Ahmed (January 1991). Discrete Cosine Transform: The Story Behind Its Creation. To cite this article: Digital Signal Processing 1 (January): 4-5.

T.81: REQUIREMENTS AND GUIDELINES FOR THE DIGITAL COMPRESSION AND CODING OF CONTINUOUS-TONE STILL IMAGES (PDF). Published in CCITT September 1992. Retrieved 2000-08-18 from the original (PDF) on that date. Get this on July 12th, 2019!.

There is an explanation of the JPEG image format on BT.com. May 31, 2018, BT Group. As of the 5th of August, 2019.

^ JPEG: What Is It? The Everyday Item That Goes Unnoticed . Published in The Atlantic on September 24, 2013. Extracted on September 13th, 2019.

Chris Baraniuk (15 October 2015). According to BBC News, Copy restrictions could come to JPEGs. Retrieved from the BBC on September 13, 2019.

The GIF Debate: A Programmer's Viewpoint," 27 January 1995. Obtainable as of May 26, 2015.

Author: L. Peter Deutsch (May 1996). Abstract. DOI:10.17487/RFC1951. DEFLATE Compressed Data Format Specification version 1.3. IETF. page 1. section Abstract. Get this: 2014-04-23.

See: Taubman, David; Marcellin, Michael (2012). Basics of Image Compression with JPEG2000: Recommended Practices and Technical Guidelines. ISBN: 9781461507994, published by Springer Science & Business Media.

^ a b Unser, M; Blu, T.

(2003).

JPEG2000 wavelet filter mathematic characteristics (PDF).

IEEE Transactions on Image Processing.

12 (9): 1080–1090.

Bibcode:2003ITIP...12.1080U.

doi:10.1109/TIP.2003.812329.

PMID 18237979.

S2CID 2765169.

Archived from the original (PDF) on 2019-10-13.

From Gary Sullivan (8-12 December 2003). Overall features of temporal subband video coding and design concerns. ITU-Video T's Coding Experts Group. Extracted on September 13th, 2019.

Alan C. Bovik (2009). Video Processing: An Essential Guide, Academic Press, p. 355, ISBN 9780080922508.

As cited in Le Gall, Didier, and Ali J. Tabatabai (1988). Sub-band coding of digital images with arithmetic coding and symmetric short kernel filters. For further information, please visit: http://dx.doi.org/10.1109/ICASSP.1988.196696. S2CID=109186495 ICASSP-88., International Conference on Acoustics, Speech, and Signal Processing: 761-764 volume 2.

To cite: Swartz, Charles S. (2005). The Professional's Guide to Understanding Digital Cinema. 147 Taylor & Francis. ISBN 9780240806174.

{End Chapter 1}

Chapter 2: Data compression

In information theory, data compression, source coding, and other related fields: In common parlance, a device that engages in the process of data compression is known as an encoder, whereas a device that engages in the process's inverse—that is, decompression—is known as a decoder.

Data compression is the process of lowering the size of a data file, and is a term that is used rather often. Source coding is an encoding process that takes place at the original data source, prior to the data being stored or transferred. This process is referred to in the context of data transmission. It is important not to mistake source coding with other types of coding, such as channel coding, which is used for error detection and correction, or line coding, which is a method for mapping data onto a signal.

Data compression is beneficial since it cuts down on the amount of space and bandwidth needed to store and transfer information. The procedures of compression and decompression both need a significant amount of computational resources. The space-time complexity trade-off is something that must be considered while compressing data. For example, a video compression method might call for expensive hardware in order for the video to be decompressed quickly enough to be watched as it is being decompressed. Additionally, the option to fully decompress the video before watching it might be inconvenient or call for additional storage space. When designing data compression schemes, designers must make trade-offs between a number of different factors. These factors include the level of compression achieved, the amount of distortion that is introduced (when using lossy data compression), and the amount of computational resources that are needed to compress and decompress the data.

In order to represent data without losing any information in the process, lossless data compression methods often make use of statistical redundancy. This ensures that the process may be reversed. Because the vast majority of data in the actual world has statistical redundancy, lossless compression is feasible. For instance, a picture may include patches of color that do not change over the course of multiple pixels; in this case, the data may be recorded as 279 red pixels rather of the traditional notation of red pixel, red pixel,... This is a fundamental illustration of run-length encoding; there are many more methods to decrease the size of a file by removing redundant information.

Compression techniques such as Lempel–Ziv (LZ) are now among the most widely used algorithms for lossless data storage. Table entries are replaced for repeating strings of data in the LZ technique