Is your screen showing you reality or a clever hallucination?
In short: When you zoom into a low-resolution photo, the computer has to invent what isn't there. Modern upscaling doesn't just stretch pixels; it solves an inverse problem by guessing a plausible image that would look like the original if shrunk back down.
Image scaling is fundamentally different depending on the file type. Vector graphics, which use geometric primitives, can be rendered at any resolution without quality loss. In contrast, raster images are discrete signals; resizing them requires resampling or reconstruction. From a signal processing perspective, this is a two-dimensional version of sample-rate conversion, governed by the Nyquist sampling theorem to prevent 'aliasing' artifacts.
Standard methods like bilinear or bicubic interpolation work by calculating intermediate pixel colors. While effective for continuous tones, they can blur sharp edges or fail when downscaling significantly—specifically more than twice the original size—because they begin to sample non-adjacent pixels. To solve this, developers use 'mipmaps' (pre-scaled copies) or 'box sampling,' which ensures every input pixel contributes to the output but is harder to optimize.
The most sophisticated modern approach involves deep convolutional neural networks and machine learning. These tools can add detail and sharpness to historical photographs that never existed in the original capture. However, this creates a cognitive trap: because these systems must 'invent' or 'recreate' details to fill the gaps, they can produce distortions. A 2025 incident involving an AI-upscaled photo of Donald Trump highlighted how these tools can hallucinate features, leading viewers to believe they are seeing original data when they are actually viewing a high-fidelity guess.
Specialized algorithms also exist for specific niches. Pixel art requires preserving stylized visual cues rather than smooth gradients; for this, 'hqx' or edge-directed interpolation (like DCCI) is used to maintain crispness. In gaming, technologies like Nvidia's DLSS use deep learning to upscale images in real-time, while AMD's FSR 1.0 relies on hand-written spatial algorithms to achieve similar results without machine learning.
Opens the Intellect+ feed with this card first. Swipe up for the next one.
Sources
- Summary by Intellect · source: Image scaling
To cite this card, link to https://intellectplus.co.uk/app/science-image-scaling.
How we source, check and correct cards: Editorial standards.
General information, not professional advice. Spotted an error? Flag the card or see corrections and complaints.
Intellect Plus is a visual educational discovery platform for exploring fascinating facts and ideas. Brain Food, Not Brain Rot. What is Intellect Plus?
Related
More in Found by OrchestratorBot · More in Science · Latest · A–Z · All topics
