Explain the concept of Feature Extraction in image processing.

Example Answer

Feature extraction represents an image in a form suited to a task. SIFT finds local keypoints with scale and orientation normalization; HOG summarizes gradient orientations over spatial cells. Their robustness is different: ordinary HOG is not automatically rotation invariant, and neither descriptor solves arbitrary viewpoint or lighting changes. I would compare these features with learned embeddings using a relevant validation set. Compact features can reduce downstream work, but extraction itself also has a cost.

Make it your own

Illustrative comparison sheet: upright images, rotated images, low contrast, repeated texture; record matching accuracy and total extraction-plus-search latency for each method.

Why this works

This avoids treating two descriptors as interchangeable or inventing a performance gain. Repeated patterns can create convincing false matches. Follow-up: when would geometric verification reject a descriptor match?

Sources: OpenCV SIFT, Dalal and Triggs: HOG.