How Photography Works | JPEG Compression Artefacts — Blocking, Ringing and Texture Loss

Two JPEG files can look almost identical at normal size and still contain different pixel values throughout the image.

One may preserve subtle texture. Another may have thrown that texture away. A third may contain faint ringing around edges. A fourth may have been saved repeatedly until block boundaries begin to appear.

JPEG compression works by keeping visual information that matters more and representing other information more coarsely.

This article continues the canonical How Photography Works | Every Photograph Leaves Something Out knowledge map. Versions owns the identity and provenance problem. This article owns the compression mechanism that can make two versions visually similar while their underlying pixel information diverges.

Quick Read

JPEG is a lossy image-compression system designed to reduce file size by representing image structure efficiently and quantising some information more coarsely than others. Typical JPEG workflows convert colour into luminance-like and chroma-like components, may reduce chroma resolution, divide the image into blocks, transform each block into spatial-frequency coefficients, quantise those coefficients and entropy-code the result. High-frequency and subtle colour information is often reduced most aggressively because human vision is generally less sensitive to some of it. Strong compression can produce blocking, ringing, mosquito noise, colour bleeding and texture loss. Re-saving an already compressed JPEG can introduce additional loss because the new encoder works from the previously altered pixels rather than the original scene data.

pixels → colour transform → blocks → frequency coefficients → quantisation → coding → smaller file

Compression Is Not the Same as Resizing

Resizing changes the number of pixels. JPEG compression can keep the same pixel dimensions while changing how accurately those pixels are represented.

A 3000 × 2000 image can remain 3000 × 2000 after aggressive JPEG compression and still lose substantial fine information. Article 33, Resizing, owns the new-grid problem. JPEG owns the efficient-coding problem.

Lossless and Lossy Compression Solve Different Problems

Lossless compression reduces file size while allowing the exact encoded values to be reconstructed. Lossy compression allows some differences between source and decoded image in exchange for greater size reduction.

JPEG is usually lossy. It is designed around the idea that not every numerical difference has equal visual importance.

JPEG Usually Separates Brightness From Colour Detail

Human vision is often more sensitive to fine brightness structure than to equally fine colour variation. JPEG commonly transforms RGB-like values into a representation with one luminance-like component and two chroma-like components.

This creates an opportunity: preserve brightness detail more strongly while representing colour detail at lower spatial resolution.

Chroma Subsampling Can Remove Colour Resolution Before the Main Compression Begins

In common JPEG encodings, colour-difference channels may be sampled less densely than the luminance channel. A 4:2:0-style representation, for example, carries less chroma detail than full-resolution colour.

Photographs often tolerate this surprisingly well. Small coloured text, saturated graphics and fine colour boundaries reveal the loss more readily than natural photographic texture.

The Image Is Divided Into Small Blocks

Traditional baseline JPEG works on small blocks, commonly 8 × 8 samples. Each block is treated as a local patch whose spatial pattern can be described through a combination of frequency-like components.

This block structure is efficient, but when compression becomes severe the independence among blocks can become visible as square boundaries.

The Transform Rewrites Pixels as Frequencies

Instead of storing every pixel value independently, JPEG uses a discrete cosine transform to represent each block as a weighted mixture of patterns ranging from broad smooth variation to fine alternating detail.

A nearly uniform sky can be described with strong low-frequency coefficients and very little high-frequency information. Fine hair, grass and text require more high-frequency structure.

Quantisation Is Where Most JPEG Loss Happens

After transformation, coefficients are divided by quantisation values and rounded. Small coefficients can become zero. Fine distinctions collapse into the same stored value.

This rounding is irreversible. Once several possible original values map to one quantised value, the decoder cannot know which exact source value existed before compression.

The Quality Slider Controls a Family of Trade-Offs

A higher JPEG quality setting generally uses gentler quantisation and produces a larger file. A lower setting quantises more aggressively and produces a smaller file with more visible loss.

But a quality value of 80 in one program is not guaranteed to equal 80 in another. Applications use different tables, scaling rules and chroma settings. The number is an interface choice, not a universal scientific unit.

Smooth Areas Compress Extremely Well

A clear sky or blurred background changes slowly across space, so many high-frequency coefficients are already small. JPEG can represent these regions efficiently with little visible damage.

This is why photographic blur can reduce file size. Simpler spatial structure requires fewer important coefficients.

Fine Texture Is Expensive

Grass, hair, fabric, leaves, gravel and fine noise contain rapid local changes. Preserving them requires many non-zero high-frequency coefficients. Aggressive compression therefore either keeps a large file or begins to simplify those textures.

A distant forest can slowly turn into painterly mush even while large edges remain convincing.

Noise Is Hard to Compress

Random variation lacks the repeated smooth structure compression systems exploit. A noisy image can therefore require more data at a given quality than a clean image.

Denoising before export may reduce file size because the image becomes statistically simpler. But Article 29, Noise, reminds us that denoising can also erase real texture.

Blocking Appears When Neighbouring Blocks Stop Agreeing

At strong compression, each block may reconstruct slightly differently from its neighbour. Smooth gradients can develop faint 8-pixel boundaries, especially in skies, skin and dark areas.

The original scene had no square grid. The squares are a property of the encoding architecture becoming visible.

Ringing Appears Around Strong Edges

When high-frequency coefficients are removed or coarsely quantised, sharp transitions can reconstruct with faint oscillating light and dark bands. Lettering and high-contrast silhouettes may develop halos.

This can interact badly with Sharpening, which may amplify compression artefacts along with legitimate edges.

Mosquito Noise Crawls Around Detail

Fine high-contrast structure can develop speckled or shimmering artefacts around its edges after strong compression. The effect is sometimes described as mosquito noise because it clusters around boundaries.

Again, the compressor did not discover new texture. It introduced an artefact while approximating the original pattern.

Repeated JPEG Saves Are Not Neutral

Open a JPEG, edit it and save again as JPEG. The second encoder does not return to the camera’s original sensor data. It compresses the already decoded and altered first-generation image.

New quantisation can round values again. Cropping or shifting by a non-block-aligned amount can change how structures fall across the new 8 × 8 grid. Over generations, errors accumulate.

Repeated Saving Does Not Always Destroy an Image Immediately

A high-quality JPEG may survive several generations without obvious damage at normal viewing size. The important point is not that every re-save is catastrophic. It is that the process is not mathematically lossless.

Visual similarity does not guarantee numerical identity.

Screenshots Create a New Compression History

A screenshot may capture a resized, tone-mapped and compressed version displayed inside another interface. Saving that screenshot introduces a new pixel grid and possibly a new compression format.

This is why Versions treats original, export, screenshot and repost as different evidence objects even when viewers call them “the same photo.”

Metadata Can Change Independently of Pixel Compression

A service may recompress an image and strip EXIF. Another may retain some metadata while changing pixels. A third may preserve the JPEG bitstream while renaming the file.

Metadata and JPEG compression therefore describe different layers of the object.

JPEG Can Be Excellent for Delivery

Lossy does not mean bad. A well-chosen JPEG setting can make a photograph many times smaller while remaining visually excellent at its intended display size. That reduces storage, page weight and transfer time.

The engineering success of JPEG is precisely that it often removes information viewers do not notice under ordinary conditions.

But Delivery Files Should Not Automatically Become Masters

A compressed web derivative may be perfect for reading an article yet weak for future heavy editing, forensic inspection, large printing or scientific measurement.

The master should preserve more information than the delivery version when future uses are uncertain.

Compression Can Change Apparent Evidence

At low quality, a tiny object can merge with its background. A thin line can disappear. Fine text can become unreadable. A compression halo can appear beside a high-contrast edge and be mistaken for a real outline.

This is why claims based on tiny details should use the highest-quality source available. Visible Evidence becomes version-dependent.

Compression Is a Form of Lossy Representation

The broader How Compression Works route explains why representation systems remove redundancy and sometimes precision to reduce cost.

JPEG is the photographic specimen: a system that looks at an image, changes the representation domain, coarsens selected distinctions and trusts perception to forgive the missing information.

Scientific and Archival Work Needs the Source Path

A JPEG can be useful for documentation and communication, but if pixel-level quantitative analysis matters, the compression history should be known. Lossless or minimally processed source data may be required depending on the measurement job.

The key is not banning JPEG. It is matching the format and quality to the evidence contract.

The JPEG Audit

1. Source: camera original, edited master or previous JPEG?
2. Dimensions: was the image resized before encoding?
3. Chroma: was colour resolution subsampled?
4. Quality: how aggressive was quantisation?
5. Texture: are fine hair, grass or fabric being simplified?
6. Edges: are ringing or mosquito artefacts appearing?
7. Blocks: are square boundaries visible in smooth regions?
8. Generation: how many lossy saves or platform recompressions occurred?
9. Metadata: did provenance information survive separately?
10. Purpose: fast delivery, editing master, archive or measurement?

Photography Laboratory 1: Quality Ladder

Export one detailed photograph at several JPEG quality settings. Compare file size and inspect sky, skin, text, foliage and hard edges at normal size and enlarged view.

Photography Laboratory 2: Generation Loss

Starting from a copy of a master, save a JPEG repeatedly at moderate quality. Compare generation 1, 5 and 10. Keep the original master untouched.

Photography Laboratory 3: Photograph Versus Graphic

Compare JPEG compression on a natural photograph and on an image with small coloured text and hard flat shapes. Notice why the graphic often reveals chroma and ringing artefacts sooner.

For Primary Readers

Save one photograph at high and very low quality. Ask what looks the same from far away and what disappears when you zoom in.

For Secondary Readers

Connect JPEG to frequency representation and rounding. Explain why small transform coefficients can be discarded while broad image structure survives.

For Advanced Readers

Analyse block-based discrete cosine transforms, quantisation matrices, chroma subsampling and entropy coding. Distinguish rate–distortion optimisation from perceptual quality and study how generational recompression interacts with block alignment and prior quantisation.

Common Misconceptions

• “JPEG just makes the file smaller without changing the picture.” Lossy JPEG changes reconstructed pixel values.
• “Same dimensions means same detail.” Compression can remove detail without changing width or height.
• “Quality 80 is a universal standard.” Different encoders interpret quality settings differently.
• “One JPEG save ruins the photograph.” High-quality encoding can be visually excellent; the key point is that the process is not lossless.
• “If I cannot see compression, no information was lost.” Perceptual similarity and numerical identity are different claims.

Frequently Asked Questions

Why do JPEG blocks often appear as squares?

Traditional JPEG processes small blocks independently, so aggressive quantisation can make differences between neighbouring blocks visible.

Does opening and closing a JPEG reduce quality?

No. Merely decoding and viewing it does not create new loss. Re-encoding and saving it again can.

Is JPEG good enough for websites?

Often yes for photographic delivery when quality and dimensions are chosen appropriately. The master should still be preserved separately when future high-quality use matters.

Final Thought: Looking the Same Is Not Being the Same

JPEG succeeds because human perception does not inspect every numerical distinction equally. It can remove a great deal before the photograph looks obviously broken.

The compressed photograph may preserve the message while changing the evidence object that carries it.