Scientists from the ATR Computational Neuroscience Laboratories have developed a new brain analysis technology that can “extract” images directly from the human brain (visual cortex) using software analyzing multivoxel patterns of fMRI 2 seconds single volume scans.
They succeeded in catching the visual cortex signals and then reconstructing the presented 10×10 patches black and white images (contrast patterns)!

In their experiment, the researchers first trained their system by “recording” individual brain patterns on known 400 different still images, then showed people the 6 letters in the word “neuron” and finally succeeded in reconstructing the presented letters.

By combining the outputs of local decoders that predicted local contrasts of multiple scales, we were able to reconstruct a large variety of images using only several hundred random images to train the reconstruction model.

Note: they “used” only 2 subjects. Obviously, the whole system is subject specific (I think!) ... training on one subject could not resolve images seen by another subject.

It was the first time in the world that it was possible to visualise what people see directly from the brain activity.

More interesting are attempts to reconstruct subjective states that are elicited without sensory stimulation, such as visual imagery, illusions, and dreams.

Neuron article [Visual Image Reconstruction from Human Brain Activity using a Combination of Multiscale Local Image Decoders]
ArsTechnica article