Google boosts sex (identification) performance

Google Boosts Sex Identification Performance

Google researchers believe they can tackle any problem they set their sights on.

A Google research challenge being eyed is: "Boosting Sex Identification Performance"

Who says research is boring? Not Googley research!

In MySpace wants Photobucket? Where is Google! I discuss the importance of Image Search to Google. Efficient Image Search is no easy feat, however, even for the Googleplex. Google researchers are on the image case, however, the sex identification of images, that is.

The most requested images from search engines are those that contain people, Google research suggests. Google people image queries include:

Specific individuals such as celebrities, actors, musicians and politicians, to general queries for adult content and stock photography images.

Given potentially billions of images searched, however, renders it "impossible to manually label all of the content." Google therefore needs filters to "better categorize and recognize the people that appear in images."

Perhaps the most basic people image filter is the determination of the sex of the person in the image. Google reesearch relies on an AdaBoost (Adaptive Boosting) meta-algorithm to improve images search performance by identifying the sex of a person from a low resolution grayscale picture of the face in the image.

Google asserts it achieves 80% accuracy in sex identification with less than 10 pixel comparisons and 90% accuracy with less than 50 pixel comparisons. The performance is an improvement over previous standards:

The best classifiers published to date use Support Vector Machines (SVM); We match their accuracies with as few as 500 comparison operations on a 20 X 20 pixel image. AdaBoost based classifiers achieve over 93% accuracy; these match or surpass the accuracies of the SVM-based classifiers, and yield performance that is 50% faster.

A practical application touted for Google's enhanced "sex identification" within images is a more efficient screening for adult content.

Google appears to be quite tolerant of adult content, however. SEE: Sex trade at Google, not Craigslist