"VIMA Filter: Objectionable Image and Website Filter" "An Anatomy of Multimodal Personalized Information Filter," Yi-Leh Wu, et al, VIMA Technologies Technical Paper, April, 2002.

ImageBeagle recognizes images using Perception-Based Image Retrieval (PBIR), which combines artificial intelligence and cognitive psychological concepts to analyze visual images the way a set of human eyes does. Our software can find what a user is looking for -- whether to retrieve, filter, analyze or organize -- faster and more accurately than any other visual image technology on the market. Our patented PBIR process breaks an image down into more than 150 features, much as the human brain does. These image features are then compared in multiple dimensions to determine an image's similarity with one or more other images. Based on a training set of sample images containing both positive and negative examples of pornography, the ImageBeagle learning engine gains the capability to score images as more or less "objectionable," and therefore worthy of review. Earlier content-based image matching approaches have relied on only about four features (color, texture, shape, contrast), which reduces accuracy. These approaches, which rely heavily on skin tone detection, make it nearly impossible to differentiate between art, pornography and medical photos -- all categories that ImageBeagle can distinguish. No image analysis system is 100% accurate in identifying pornography. But ImageBeagle's is the best available, and provides an effective tool for identifying potentially offensive images.

Copyright © 2003, VIMA Technologies, Inc. Click here to read disclaimer and legal notices. Trademarks referenced here are property of their respective owners.