Ever thought that the zebra’s black-and-white striped markings resemble a biological barcode? Well now a team of US computer scientists and biologists have come up with a scanner, allowing them to identify individual animals from a single still photo.
The system, dubbed StripeSpotter, only requires a small amount of human input. Users draw a rectangle around the zebra’s side, then this part of the image is automatically sliced into a number of horizontal bands and each pixel is made fully black or fully white, creating a low-resolution version of the zebra’s stripes.
Each band is then encoded as a StripeString, a sequence of coloured blocks with particular lengths – for example, white for two blocks, black for three, white for one – and the collection of StripeStrings forms a StripeCode, the zebra equivalent of a barcode.
When a zebra has been entered into the database and given a StripeCode, the researchers match another picture of the same animal by comparing the StripeStrings of the new and original images. Each image will generate a different set of StripeStrings, but the underlying ratios of black and white should remain similar.
By finding the StripeCode with the most similar StripeStrings in the database, the system is able to accurately identify the correct animal. Other existing zebra identification systems are less accurate, more complex, and require a greater level of manual input from the user.
The zebra scanner is not the first algorithm developed to identify animals in the field – there are also systems for tagging turtles, penguins and rhinos. This new system could also be applied to tigers and giraffes, or any animal with large markings in a small number of distinctive colours. The researchers will present their work at the International Conference on Multimedia Retrieval in Trento, Italy later this month.