© 2020, Roman Juranek ijuranek@fit.vutbr.cz
Minimalistic library for image perspective correction. The library prvides API for automated identification of converging line groups and image transform computation from automatic identification of two orthogonal vanishing points. The library does not warp the image, this is in the hands of the user.
Development of this software was funded by TACR project TH04010394, Progressive Image Processing Algorithms.
The library requires no external dependencies (apart from standard library and OpenMP runtime). Internally we use Eigen for representation of images and linear algebra. Example app uses OpenCV for image IO and warping.
Input image
Automatically detected lines by find_line_segment_groups. The function accepts float image buffer, its dimmensions and stride between rows and returns allocated array of LineSegment structures which can by modified (filtering, user input, etc.).
The library provides corners for image transformation with function compute_rectification_transform. The function accept line segments and returns ImageTransform structure with new image corners.
top_left: -60.5308,-13.6642
top_right: 1041.69,-54.4891
bottom_left: 55.0519,649.464
bottom_right 969.276,621.39
There are more than one strategy how to calculate the transform - user can specify what should happen in horizontal/vertical direction. The options are - rectify, rotate (by horizontal or vertical angle) or do nothing. There are 15 meaningful combinations suitable for different use cases. The responsibilit of user is then to use the corners to calculate the final image.
Here is an example of warped image calcualted by cv::warpPerspective (see autorectify.cpp for code).
