Skip to main content

Vanishing Point Detector

Project description

Automated Rectification of Image

Implements the modified version of the following paper:

Chaudhury, Krishnendu, Stephen DiVerdi, and Sergey Ioffe. "Auto-rectification of user photos." 2014 IEEE International Conference on Image Processing (ICIP). IEEE, 2014.

Modifcation note: Instead of finding edge direction using structural tensor and its eigenvectors as in paper, I have used more reliable canny edge detection and probabalistic hough line transform.

Results

Input image:

Input Image

After rectification:

Rectified Image

How it works

First, compute list of 'edgelets'. An edgelet is a tuple of edge location, edge direction and edge strength.

edgelets1 = compute_edgelets(image)
vis_edgelets(image, edgelets1) # Visualize the edgelets

Edgelets

Next, find dominant vanishing point using ransac algorithm. In our case it turns out to be horizontal.

vp1 = ransac_vanishing_point(edgelets1, num_ransac_iter=2000, 
                             threshold_inlier=5)
vp1 = reestimate_model(vp1, edgelets1, threshold_reestimate=5)
vis_model(image, vp1) # Visualize the vanishing point model

Horizontal Vanishing Point

Remove the inliers for horizontal vanishing point. Vertical lines should now be dominant. Recompute the vanishing point using ransac should give us vertical vanishing point.

edgelets2 = remove_inliers(vp1, edgelets1, 10)
vp2 = ransac_vanishing_point(edgelets2, num_ransac_iter=2000,
                             threshold_inlier=5)
vp2 = reestimate_model(vp2, edgelets2, threshold_reestimate=5)
vis_model(image, vp2) # Visualize the vanishing point model

Vertical Vanishing Point

Finally, compute homography and warp the image so that we have a fronto parellel view with orthogonal axes:

warped_img = compute_homography_and_warp(image, vp1, vp2,
                                         clip_factor=clip_factor)

Rectified Image

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

pyvpd-0.1.1.tar.gz (7.5 kB view hashes)

Uploaded Source

Built Distribution

pyvpd-0.1.1-py3-none-any.whl (7.6 kB view hashes)

Uploaded Python 3

Supported by

AWS AWS Cloud computing and Security Sponsor Datadog Datadog Monitoring Fastly Fastly CDN Google Google Download Analytics Microsoft Microsoft PSF Sponsor Pingdom Pingdom Monitoring Sentry Sentry Error logging StatusPage StatusPage Status page