Image Transformation¶
Image Transformations¶
Image transformations change the domain (pixel coordinates) rather than the range (pixel intensities). Transformations include translation, rotation, scaling, affine warps, and projective warps. See the detailed notes in the companion section for mathematical foundations, homogeneous coordinates, and code demos.
Image Morphing¶
Image morphing combines warping (geometric transformation) with cross-dissolving (intensity blending) to create smooth transitions between two images. Key steps:
Define corresponding feature points/lines in source and target images
Compute intermediate warps for each frame
Cross-dissolve pixel intensities at each intermediate step
Panorama¶
Panorama stitching aligns and blends multiple overlapping images into a single wide field-of-view composite:
Detect and match features across image pairs
Estimate homography (projective transformation) between images
Warp images into a common coordinate frame
Blend seams to produce the final panorama
High Dynamic Range (HDR)¶
HDR imaging captures and merges multiple exposures of the same scene to recover the full range of irradiance values:
Capture bracketed exposures (short → long)
Estimate camera response curve
Reconstruct radiance map
Apply tone mapping to display on standard monitors
Stereo¶
Stereo vision uses two images from slightly different viewpoints to estimate depth:
Find corresponding points between left and right images
Compute disparity (pixel offset) — inversely proportional to depth
Generate a depth map from disparity values
Photosynth¶
Photosynth (Microsoft) reconstructs 3D scenes from large collections of unstructured photographs using structure-from-motion techniques.
Extrinsic Camera Parameters¶
Extrinsic parameters describe the camera’s position and orientation in world coordinates:
Rotation matrix R (3×3): camera orientation
Translation vector t (3×1): camera position
Combined as a 3×4 matrix [R | t]
Intrinsic Camera Parameters¶
Intrinsic parameters describe the camera’s internal geometry:
Focal length (fx, fy)
Principal point (cx, cy) — where optical axis meets image plane
Skew coefficient
Encoded in the 3×3 camera matrix K
Camera Calibration¶
Camera calibration recovers intrinsic and extrinsic parameters from images of known calibration targets (e.g., checkerboard patterns) using techniques like Zhang’s method.