Feature detector vs feature descriptor.

WHAT THIS TESTS The interviewer checks whether you separate finding keypoints from characterizing them, a foundational distinction underlying matching, stitching, and SLAM.

A GOOD ANSWER COVERS A feature detector finds interest points: locations in an image that are distinctive and repeatable across viewpoint, scale, or lighting changes, such as corners, blobs, or edges. It answers where the notable points are. A feature descriptor takes a detected keypoint and its surrounding patch and encodes that local appearance into a compact, often invariant, vector. It answers what the neighborhood looks like, so that the same physical point can be recognized and matched in another image by comparing descriptor vectors. Detection localizes; description characterizes for matching. Many classic pipelines combine both, and some algorithms provide both stages while others specialize. The descriptor's invariance to rotation, scale, and illumination is what makes robust matching possible, while the detector's repeatability ensures the same points are found again.

COMMON WRONG ANSWERS Using the terms interchangeably. Thinking a detector outputs a matchable signature, or that a descriptor finds keypoints. Believing every method does only one role.

LIKELY FOLLOW-UPS Which properties make a good detector versus a good descriptor, how matching uses descriptors with a ratio test, binary versus float descriptors, and where learned methods fit.

ONE CONCRETE EXAMPLE The Harris or FAST detector flags a sharp corner of a window as a keypoint, the where; the SIFT or ORB descriptor then summarizes the gradient or intensity pattern around that corner into a vector, the what, so the identical corner can be located in a second photo by matching descriptors.