A team of researchers from Pittsburgh’s Carnegie Mellon University have created sets of eyeglasses that can prevent wearers from being identified by facial recognition systems, or even fool the technology into identifying them as completely unrelated individuals.
In their paper, Accessorize to a Crime: Real and Stealthy Attacks on State-of-the-Art Face Recognition, presented at the 2016 Computer and Communications Security conference, the researchers present their system for what they describe as “physically realisable” and “inconspicuous” attacks on facial biometric systems, which are designed to exclusively identify a particular individual.
Half of US adults are recorded in police facial recognition databases, study says
The attack works by taking advantage of differences in how humans and computers understand faces. By selectively changing pixels in an image, it’s possible to leave the human-comprehensible facial image largely unchanged, while flummoxing a facial recognition system trying to categorise the person in the picture.
Research Paper: Accessorize to a Crime: Real and Stealthy Attacks on State-of-the-Art Face Recognition
Sharif, Bhagavatula, Bauer, Reiter
In September 2016, engineers at Yahoo released a machine learning model trained to detect “NSFW” images. The acronym NSFW, “not safe for work”, is net slang for images with typically nudity or graphic sexual content. Recently (October 2016) PHD research Gabriel Goh has released experiments applying a means of generating images from such models.
Detecting offensive / adult images is an important problem which researchers have tackled for decades. With the evolution of computer vision and deep learning the algorithms have matured and we are now able to classify an image as not suitable for work with greater precision.
Defining NSFW material is subjective and the task of identifying these images is non-trivial. Moreover, what may be objectionable in one context can be suitable in another. For this reason, the model we describe below focuses only on one type of NSFW content: pornographic images. The identification of NSFW sketches, cartoons, text, images of graphic violence, or other types of unsuitable content is not addressed with this model.
Since images and user generated content dominate the internet today, filtering nudity and other not suitable for work images becomes an important problem. In this repository we opensource a Caffe deep neural network for preliminary filtering of NSFW images.
What makes an image NSFW, according to Yahoo? I explore this question with a clever new visualization technique by Nguyen et al.. Like Google’s Deep Dream, this visualization trick works by maximally activating certain neurons of the classifier. Unlike deep dream, we optimize these activations by performing descent on a parameterization of the manifold of natural images. This parametrization takes the form of a Generative Network, trained adversarially on an unrelated dataset of natural images.
In How One Sees, Siegfried Zielinski and Franziska Latell trace the genealogy of vision. Quoting the philosopher Abu Nasr Al-Farabi (870-950), the authors write:
Optics teach “according to the true circumstances of what is looked at to find the matter, the quantity, form, position and order and the other things which of the things is where the gaze can be mistaken […] That is done with an object via an instrument, which serves to direct the gaze in such a way that it does not err”.[…] Al-Farabi assumes an active ray of sight; it carries light from inside the eye or human body and is beamed through the eye at external objects to scan them for perception. This divine scanner is one of the foundations of the Platonic world-view.
The image above is a sketch illustrating the theories of Ibn al-Haytham, astrophysicist, inventor of the camera obscura and keen moon-gazer for whom the light is emitted from luminescent objects and find their way to the retina. Al-Haytham’s model contradicts the active ray of sight’s model and will influence medieval perspectivists and aid the development of perspectival representation.
How One Sees, Siegfried Zielinski and Franziska Latell in Variantology 4: On Deep Time Relations of Arts, Sciences and Technologies In the Arabic-Islamic World and Beyond (Kunstwissenschaftliche Bibliothek), 2011.
Very interested in the situological and situographical development of topology. It will be necessary to stay rapidly informed of all the scientific conclusions about this — and to adapt or détourn them. The primary force of our position is to intervene therein as an artistic activity (with a game of gestures raised to the dignity of art) whereas the former dominant tendency was toward objective observation.
Guy Debord in letter to Asger Jorn, July 6 1960
Guy Debord: Correspondence: The Foundation of the Situationist International (June 1957 — August 1960). Translated by Stuart Kendall and John McHale. Semiotext(e), 2009.
Binford, O.T. & Nevatia, R (1977) Description and Recognition of Curved Objects, Artificial Intelligence 8(1):77-98.
In this experiment [Hochberg & Brooks, 1962], a human baby was raised until the age of 19 months under the constant supervision of his parents who avoided exposing the child to line-drawings or two-dimensional pictures of any kind. Although the baby accidentally had opportunities to glance at some pictures on a few occasions, at no point was the content of a picture ever named to him or was other attention drawn to it. All of the baby’s playthings were chosen so that they had solid coloring and no two-dimensional patterning of any kind. Finally, at the age of 19 months the child was shown some line-drawings for the first time, including those illustrated in Figure 1-3. The child was immediately to recognize objects in these drawings with no reported difficulty, and performed equally well when identifying the contents of black-and-white photographs.
Found in: Lowe, D.G. (1985) Perceptual Organization and Visual Recognition , http://www.dtic.mil/dtic/tr/fulltext/u2/a150826.pdf
“In a footnote Pickard reports that in lab experiments with students playing the computer game Doom, signs of stress came less with the appearance of a new deadly enemy (the intended site of emotional affect) than during times when students were experiencing difficulty configuring the software.”
in Suchman, L. (2007) Human-Machine Reconfigurations. Cambridge University Press, New York, p215.
Portrait heads of the emperor Octavius Augustus.
“The Joslyn portrait displays clear indications that it was recut from another person’s likeness, limiting the sculptor by dissimilarities to Augustus’ features in the original subject. Recut portraits in general exhibit asymmetries, undercutting, and unnatural planar features where a chisel or other implement has been used to remove undesirable elements. Here, for example, a shallow trough on the forehead is a result of cutting back longer locks in the original. Great care was taken to emulate Augustus’ signature fork-and-pincer feature higher up on the forehead, using the technique of undercutting to create the illusion of volume and definition, which in the process created the trough. […] Careful examination of the nape of the neck further reveals the work of the chisel in removing excess hair. […]