Duplication Test: Same Pixels, Different Tags



Add to Favorites
Report Abuse


This was a test to determine what would happen if the synther were given three exact copies of the same image.

I say exact copies, but I only refer to the pixels being identical. In order to have the synther treat them as three individual objects, I applied a different alphanumeric tag to each of the three. Editing the metadata is, in fact, the only way that I know of to give the exact same visual data to the synther multiple times without it detecting the copies as clones and thus not bothering to treat them as something to be compared with the original.

Logic told me that by giving the synther three copies of the exact same pixels, each feature found would be found in the exact same fashion in all three instances. When compared to each other, every triple match should then result in a point for the pointcloud, and therefore, although the synther only uses a 1.5 megapixel copy of any image that it is given for image feature detection and scene reconstruction, we should see a dense plane of points - one for every single image feature found within a single 1.5 megapixel copy of our original, rather than only those which successfully matched images from other perspectives.

This self verification of image features is to be avoided wherever possible, in fact the normal synthing process is designed to do just that. According to my understanding, many more image features are found in every photo than will ever be correctly identified in photos from other perspectives, ergo our pointclouds are not as dense as they would be if even every feature found in every 1.5 megapixel version of our images were retained in the final reconstruction. The reason is that any feature found in only one image cannot have its three dimensional location triangulated. Because of this, any feature whose position cannot be solved for is useless to have intersecting the correctly positioned points with flat planes of points. It would simply reduce the accuracy and desireability of our pointcloud.

As it stands, having given the synther exactly three copies of exactly the same pixels, which should result in identical 1.5 megapixel versions of themselves. I see an array of points here, but nothing like as many as I would expect to be detected in this image, even a 1.5 megapixel version of this photo.
Synthy 100%
Views 54
Favorites 0
Photos 3
Date Created 2/12/2010

Related Photosynths


You need to be Signed In to add a comment.
New to Photosynth? Sign Up for a free account.