Logo Questions Linux Laravel Mysql Ubuntu Git Menu
 

How to identify similar images despite some being cropped?

I have 100 images, all visually similar and I need to search for duplicates. I have an algorithm which can match same colour/pattern/editing etc. but it doesn't support cropping. Meaning if two similar image, one of which is cropped, the result will be different.

I need algorithm which can match two similar images despite of cropping, somehow like tineye works. I got some references but nothing worked.

Here's an example article for reference.

like image 941
125fura Avatar asked Jan 19 '15 07:01

125fura


People also ask

Does Reverse Image Search work if cropped?

If you find a cropped image and wonder what's behind the crop, just reverse search the image. You will be able to find plenty of examples of the original along with other cropped results.

What will happen to the image when you cropped it?

To “crop” an image is to remove or adjust the outside edges of an image (typically a photo) to improve framing or composition, draw a viewer's eye to the image subject, or change the size or aspect ratio. In other words, image cropping is the act of improving a photo or image by removing the unnecessary parts.


1 Answers

I think you are on the right track using hashing for identifying nearly-duplicates. I believe locality sensitive hashing can give you the extra mileage you need. It takes into account the locality of the image features from which it computes the hash key and thus achieves better performance for the task of near-duplicates detection.

like image 155
Shai Avatar answered Oct 03 '22 02:10

Shai