A Combined Corner and Edge Detector¶
Why this mattered¶
Harris and Stephens reframed low-level feature detection around the local auto-correlation, asking how image intensity changes under small shifts in two directions. This made “cornerness” a geometric property of local image structure rather than a hand-coded response to particular edge templates. Edges, flat regions, and corners could be distinguished by the eigenvalue structure of the second-moment matrix: one strong direction indicated an edge, two strong directions indicated a corner. The practical Harris response made this idea computationally simple enough for real image sequences.
The shift mattered because it produced repeatable, local, trackable points in natural imagery. Before robust corner detectors, feature tracking for 3D interpretation and motion estimation was limited by unstable or ambiguous edge measurements. After Harris-Stephens, vision systems could reliably select distinctive image points that survived small viewpoint, illumination, and motion changes, enabling stronger pipelines for stereo matching, structure from motion, visual odometry, mosaicing, and camera calibration.
The paper also established a template for later local-feature breakthroughs: detect stable interest points first, then match or track them across images. SIFT, SURF, ORB, KLT-style tracking, and many SLAM systems inherited this separation of detection, localization, and correspondence, even when they replaced the detector or added scale and rotation invariance. In that sense, the Harris detector was not just a better corner measure; it helped make sparse, repeatable image features a central primitive of modern computer vision.
Abstract¶
Consistency of image edge filtering is of prime importance for 3D interpretation of image sequences using feature tracking algorithms. To cater for image regions containing texture and isolated features, a combined corner and edge detector based on the local auto-correlation function is utilised, and it is shown to perform with good consistency on natural imagery.
Related¶
- enables → Distinctive Image Features from Scale-Invariant Keypoints — The Harris corner detector established repeatable local interest points that SIFT extended with scale-invariant keypoint detection and descriptors.
- cite ← Distinctive Image Features from Scale-Invariant Keypoints — Lowe's SIFT uses Harris and Stephens' corner-detection idea as an antecedent for identifying stable local image features.
Sources¶
- DOI: https://doi.org/10.5244/c.2.23
- OpenAlex: https://openalex.org/W2111308925