# Image Blending ## Feathering Do a gradual transition between the images. However, you still might have some artefacts ![[feathering.jpg]] How do decide good window size? - Smooth but not ghosted - Empirical evaluation ![[feathering-optimal.jpg]] ## Multi-band Pyramid blending Idea is to create a Laplacian pyramid, blend each level, like [[Scale-Invariant Feature Transform (SIFT)]] 1. Compute Laplacian pyramid 2. Compute Gaussian pyramid on _weight_ image 3. Blend Laplacians using Gaussian blurred weights - At low frequencies, blend slowly - At high frequencies, blend quickly 4. Reconstruct the final image ![[laplacian pyramid.jpg]] Blending at different scales ![[blend-laplacian.jpg]] Result: ![[pyramidblending-result.jpg]] ## Choosing seams ### Planar Blending (Simple method) - Assign each pixel to image with nearest center - Create a mask: - Smooth boundaries ( "feathering"): - Composite ![[planar-blending.jpg]] ### Dynamic program method - We dont want hard boundaries around images - Object recognition based ![[dynamic porgram blending.jpg]] ## Gain compensation Simple gain adjustment - Compute average RGB intensity of each image in overlapping region - Normalize intensities by ratio of averages - Still might have artefacts, can apply laplacian blending ![[gain-compensation.jpg]] ## De-Ghosting ![[deghosting.jpg]] --- ## References