Wednesday, July 01, 2009

PPP – Product Placement in Photographs

A New Product Idea for Google


User-submitted photos are image-analyzed into their constituent objects. Generic objects (bottles, cars, watches) are selectively (and somewhat inconspicuously) replaced by product-specific iconic versions (e.g. Coca-cola bottle, Tesla car, Rolex watch). The manufacturer pays Google for this service. The user can opt for the generic versions on payment of a small subscription.

Product Description

Holiday photos are frequently unsatisfactory. The resolution is too low, the digital zoom has removed all detail, there is motion blur – and 2D is so flat. All of this is fixable by algorithms.

STEP 1: User submits a .jpg or similar to Google Images.

STEP 2: Google applies a palette of image-processing algorithms to deblur, identify object edges, resolve ambiguity through contextual knowledge and inference resulting in the creation of a 3D scene description. This probably looks like an XML file.

In many cases the resolution will identify a manufacturer-specific entity, e.g. a jeep. However, in other cases there will just be a generic match such as a bottle, a car, a journal.

The business opportunity is to replace the generic object description by product placement of a distinctive version emblematic of a particular company, which could include adding a logo. Google would charge manufacturers for this service in a variant of their current business model.

STEP 3: The image description is then served back to the user where it can be viewed/rendered via a Google image viewer. This could include viewing at any resolution and 3D rotation.

If the user objects to branded products appearing within their picture (the point would be to make product placement somewhat unobtrusive) then they could avoid advertising by making a payment for the service.


To accurately reconstruct a scene from an image requires contextual knowledge. The Google service should include an interactive function whereby users can correct the results, corrections which could be used to tune the knowledge/inference engine.


All claims to this idea I freely cede to Google as I’d like to use this service. Ask me also about fixing video, especially from low-resolution camera-phones (and CCTV).


Conceived Wednesday 24th June in the Dordogne.