Towards a robust nipple detector

Training_setIncreases in raw computing power and novel algorithmic techniques have enabled outstanding advances in image processing since 1999, when Drs. Forsyth and Fleck developed the first computerised system for Automatic Detection of Human Nudes.

Now, researchers at the Institute for Infocomm Research, I²R (pronounced as i-squared-r), which is part of A*STAR, Singapore, have refined the idea of simple body-detection algorithms towards what they call ‘Organ Level Detection’. Nipples in particular.

Dr. Yue Wang, and colleagues have developed a method for automatic nipple detection using shape and statistical skin colour information.

“First the statistical information of skin color relation between nipple and region surrounding nipple has to be extracted, and then this information is used to filter off the non-nipple region detected in last stage.”

They used a carefully tweaked version of the now-classic AdaBoost algorithm which had been ‘trained’ on a representative dataset. Unfortunately though, despite its refinements, the new computer code is still quite a way away from the goal of 100% reliability. In a trial with 980 test images which featured 348 nipples, 263 were successfully identified – but 85 were missed. And there were 170 false detections (including navels and eyes etc). The team is currently working on firming up the software – and is at the same time developing an entirely new approach. “… the private part of human body must be detected as well. A different model has to be constructed for this purpose.”

See: Automatic Nipple Detection Using Shape and Statistical Skin Color Information (S. Boll et al. (Eds.): MultiMedia Modelling 2010, Lecture Notes in Computer Science 5916, pp. 644–649, 2010.) Here is further detail from the study:

n-templates

Question : Is  A*STAR  an example of  RAS syndrome  like PIN number and LCD display?