The HeavyChef Project

Googles continued push for improving search took another step forward last week when two of its scientists presented a research paper in China. The research outlined the future of image search online.

In simple terms, images are currently indexed by the content that surrounds them, the page itself and the alt tags used to describe them. But Googles so called VisualRank (Page Rank for Product Image Search) means us humans will no longer be needed to help describe what the image is. It uses visual recognition to identify what is in an image in order to categorise it for search.

Visual recognition software is something that has been developed greatly over the past few years but it does throw up the question of “How effective can it be for search?”

To find out, I investigated how good human beings actually are at deciphering images. With the thought that, if we struggle on some image recognition, how good could a machine be?

I use the term “human being” loosely as my test group were 6 members of the World Wide Creative team. Below are the images used and the descriptions they came up with.

Descriptions:
Bird
Bird
Bird from Chenobyl
Frog butterfly
Entry into worth1000.com photoshop competition
Manipulated picture of bird frog and butterfly

Descriptions:
Rhino
Elephant
Dustbin with street person
Mans head in fire jet
Your transmission is being monitored
Guy with head up a engine component that resembles a traffic one

Descriptions:
squirrel rat
kangaroo
mamouse
squirrel rodent
squirrel
museum display of prehistoric rat

Description:
Gremlin
Toy
Frog Boat
ugly man on boat
popeye on drugs
cartoon drawing of shek on drugs

Descriptions:
Dead Rat
Roadkill
Dead possum
your government working for you
Yellow Lines
Roadkill

Descriptions:
Baldy
Bald guy
Surprise
Wide Eyes
Baby
Dave Duarte

Extremely diverse tagging by humans! So would Googles’ visual recognition be any better? Bearing in mind search engine optimisers already manipulate the alt tags for their own gain anyway, should the question rather be: “Could visual recognition software be any worse at providing a good search result?”