Autostitching Panoramic Images

Python application for stitching panoramic images.

Description

Implements stitching arbitrary images either through manually defined features or through SIFT feature matching based on the Harris corner detector, as described in Brown et Al. Stitching is accomplished using RANSAC to compute a robust homography between the set(s) of image correspondence, then warping each image toward a chosen reference image. Finally, multiresolution blending is accomplished using Laplacian pyramids. The application also supports image rectification.

Usage

Requirements are listed in requirements.txt. Application is run from main.py:

usage: main.py [-h] -r REF [-w WARP [WARP ...]] -m
               {rectify,manual_mosaic,autostitch} [-v]

optional arguments:
  -h, --help            show this help message and exit
  -r REF, --ref REF     reference image path
  -w WARP [WARP ...], --warp WARP [WARP ...]
                        warp image path(s)
  -m {rectify,manual_mosaic,autostitch}, --mode {rectify,manual_mosaic,autostitch}
                        choose between image rectification, manual
                        correspondence mosaicing and autostitched mosaicing
  -v, --verbose         log file reads/writes to stdout & visualize
                        intermediate autostitching outputs

Modes

• rectify: requires only -r REF
• manual_mosaic: requires -r REF and at least one -w WARP, supports a single global feature set
• autostitching: requires -r REF and at least one -w WARP

Global filepaths for intermediate and final outputs are set in utils.py.

Note: Click same coordinate twice to end manual shape labelling.

Examples

Rectification

Hallway	Hallway Rectified

Mosaicing

Building Left	Building Right	Building Mosaic
Room Left	Room Right	Room Mosaic