How do you build an HDR image from bracketed exposures?

WHAT THIS TESTS The interviewer checks whether you treat HDR as physically recovering scene radiance across a wide range, then compressing it for display, rather than naively blending JPEGs.

A GOOD ANSWER COVERS First align the frames; even on a tripod, sub-pixel shifts and ghosting from any motion must be handled. Next recover the Camera Response Function, the nonlinear mapping from scene radiance times exposure time to recorded pixel value, using a method like Debevec and Malik that solves for the inverse response from corresponding pixels across exposures. Invert the CRF to linearize each image into radiance, divide by exposure time, then merge into one radiance map using a weighted average that gives more weight to well-exposed mid-tone pixels and down-weights saturated or noisy dark pixels. The result is a floating-point HDR radiance map with far more range than 8-bit can show. Finally apply tone mapping to compress that range into displayable values, preserving local contrast and detail.

COMMON WRONG ANSWERS Averaging pixel values directly, which is wrong because pixels are nonlinearly encoded. Forgetting alignment and ghost removal. Conflating the radiance map with the final viewable image, or skipping tone mapping entirely.

LIKELY FOLLOW-UPS Global versus local tone mapping operators, ghost removal for moving subjects, why RAW data may let you skip CRF estimation, and how exposure weighting functions are chosen.

ONE CONCRETE EXAMPLE Given three shots at one over sixty, one over fifteen, and one quarter second, you linearize each via the recovered CRF, scale by exposure, and merge so the bright window comes mostly from the fast exposure and the dark interior from the slow one, then tone map so both window and room are visible on screen.