The Digital Press

Researchers created a vast database measuring similarity between sub-regions of Hubble Space Telescope images using human comparisons from a citizen science project. The resulting data helps evaluate and improve image search algorithms by capturing detailed visual similarities.

Why it matters: Human-reviewed similarity data enhances accuracy assessment of computer vision algorithms for astronomical images.

The big picture: Combining citizen science and paid crowdwork yields detailed, reliable similarity metrics reflecting subtle image features.

Stunning stat: Nearly 850,000 human comparison measurements contributed to building the similarity distance matrix.

Quick takeaway: Collective human judgment matches expert accuracy in capturing complex visual similarities like morphology and texture.