Take two 1920×1080 images. One is a gradient from white to black. One is a photograph of a crowd at a concert. Both contain the same number of pixels—around 2 million.
Compress them both as PNG (lossless). The gradient might be 50KB. The crowd photo might be 4MB—80 times larger.
Same pixel count. Same format. Wildly different file sizes. The reason is entropy.
What Entropy Means for Images
In information theory, entropy measures the amount of unpredictability—or information—in a data sequence. High entropy means data is hard to predict; low entropy means data is highly regular and predictable.
A white gradient has extremely low entropy. Each row of pixels is nearly identical to the previous row. The progression from white to black changes gradually and predictably. A compressor can describe the entire image very efficiently: “start white, decrease brightness by X per pixel, repeat.”
A crowd photo has extremely high entropy. Each pixel is independent from its neighbors. You can’t predict the color of pixel (847, 523) from the pixels around it. The compressor has to describe each pixel individually—or use complex statistical models to eke out any reduction at all.
What Creates High Entropy in an Image
Noise and grain. Camera noise is, in a meaningful sense, random. Random data has maximum entropy by definition—it cannot be compressed. An image shot at ISO 6400 has much higher entropy than the same scene at ISO 100 because noise introduces unpredictability everywhere. This is why applying noise reduction before compression often dramatically reduces file size.
Fine texture. Grass, fabric, tree bark, sand—any surface with fine, irregular texture has high local entropy. The colors change unpredictably at the pixel level even though they look cohesive at human viewing scale.
Complex photographic subjects. A photo of a crowded market, a forest, or a fireworks display has more information than a photo of a white wall, an empty beach, or a solid-color product on a plain background.
High-contrast detail. Sharp edges create sudden pixel value changes—high local entropy—while smooth gradients create gradual, predictable transitions.
What Creates Low Entropy
Solid colors. A logo on a white background. A simple icon. A one-color background. These compress very efficiently because large regions have identical values.
Smooth gradients. The sky in a landscape photo, a blurred background (bokeh), a graphic design element—all change gradually and predictably.
Simple geometric shapes. Vector-sourced content exported to raster (logos, UI screenshots, diagrams) tends to have large uniform areas and predictable edges.
Post-processing that smooths. Heavy noise reduction, blur, posterization, and aggressive JPEG compression reduce entropy—partly by destroying information, partly by making remaining pixel values more predictable.
Why This Matters for Choosing a Compression Format
Lossless formats (PNG, lossless WebP) preserve every pixel exactly. For low-entropy images—logos, screenshots, diagrams, solid-background icons—lossless compression is efficient because the image has few surprises to encode. PNG is better than JPEG for these use cases.
For high-entropy photographic images, lossless compression produces large files because there’s too much information to predict. A lossless PNG of a detailed photograph can be 10–20× larger than a quality-80 JPEG of the same image.
Lossy formats (JPEG, lossy WebP, AVIF) discard some information to achieve higher compression. They’re effective on high-entropy images because the information they discard is often in the noise, fine texture, and imperceptible detail that human vision doesn’t notice much.
The practical rule: Use lossless for low-entropy images (graphics, logos, screenshots). Use lossy for high-entropy images (photographs, video frames, images with heavy texture).
Why Noise Increases File Size
A photographer asking “why is this raw photo 20MB when the output JPEG should be smaller?” is often running into entropy. If the image was shot at high ISO, the noise adds true randomness to every pixel, which the compressor can’t eliminate. The JPEG encoder tries to discard the high-frequency noise as part of compression, but at higher quality settings, some noise survives.
Applying noise reduction before compression—in Lightroom, Photoshop, or a standalone denoiser—removes entropy before it reaches the encoder. The result is often a dramatically smaller file at the same perceived quality level, because the input image has fewer surprises to encode.
This is not a trick. You’re genuinely removing information (noise) that you don’t want. The resulting image has lower entropy, and compression can describe it more efficiently.
Entropy and JPEG Quality Settings
At high JPEG quality settings (90–100%), the encoder discards very little information—the output closely matches the input, including its entropy. At low quality settings (20–50%), the encoder discards aggressively, reducing entropy by replacing fine detail with blocky approximations.
The visual goal is to find the quality setting where you’ve discarded noise and imperceptible detail without discarding structure that the eye uses for perception. For most photographs, quality 75–85% achieves this. Below 70%, entropy reduction starts destroying meaningful detail.
The Bottom Line
Image entropy explains why compression ratios vary so dramatically across different images—and why no single quality setting is right for all images. A high-entropy crowd photo needs a higher quality setting to look acceptable; a low-entropy product shot can be compressed much more aggressively without visible degradation.
When you’re trying to minimize file size, the highest-impact actions are: shooting at lower ISO to reduce noise entropy, applying noise reduction before compression, and using lossy formats for photographs while reserving lossless for graphics.
See for yourself how entropy affects compression. Drop any image into the free Image Compressor and compare the output at different quality settings.