Awecom, Inc.

University of California

Emersys, Inc.

Christopher Hillar

Kilian Koepsell

Ram Mehta

Charles Garfinkle

Laurence Howard Tecott

Jascha Sohl-Dickstein

Alexander Terekhov

Christopher Priebe

Felix Effenberger

Darren Rhea

An image processing method includes selecting an image in fixed storage of a computer and loading the selected image into memory of the computer. The method further includes representing the loaded image by a processor of the computer in the memory as an initial two-dimensional array of pixel values. Thereafter, the initial two-dimensional array of pixel values may be transformed into a hierarchy of progressively axially decremented two-dimensional arrays of signs and a pair of one-dimensional values for each 2&#xd7;2 array of signs amongst the decremented two-dimensional arrays of signs. Finally, each of the two-dimensional arrays of signs and each pair of one-dimensional values may be stored in the fixed storage as a compressed form of the selected image.

Resilient image compression and decompression

Semantically-segmented video compression includes loading a frame of video imagery, determining a context for the frame, selecting an object of interest for the context and identifying within the frame, a portion of the frame of the imagery deemed to be of greater interest than other portions of the frame of video imagery based upon the object of interest of the context determined for the video imagery, and an object, that has been pre-specified to be of importance to the context. A hybrid compression of the video imagery is then performed that includes both a higher quality compression of the portion of the video imagery determined to be a region of higher interest that produces a minimization of loss during decompression and also a compression of the other portions of the video imagery determined not to be a region of higher interest that produces more loss during decompression.

Semantically segmented video image compression

Methods for classifying subjects and analyzing the relation between subject classification and multiple features of the subjects are provided. Some embodiments of the disclosure provide processes that use clustering analysis as an unsupervised machine learning technique to classify subjects based on multiple features. Some embodiments may associate features of the subjects with a categorical variable on which subject groups are based. This association between the features and the categorical variable of interest (or subject groups) can be obtained by finding featured-based clusters that have similar members as the subject groups. Systems and computer program products implementing the methods are also disclosed.

Unsupervised high-dimensional behavioral data classifier

An invention based on learning a discrete recurrent neural network for a given signal domain is described. In one implementation to the domain of visual images, the method can be used to efficiently compress digital photographs and to devise a new perceptual distortion measure between images that well-matches data collected from a human psychophysics experiment. Other applications of the invention include unsupervised detection of recurrent patterns in high-dimensional data and Shannon-optimal error-correcting coding from few training examples.

Self-organizing discrete recurrent network digital image codec

Growing community of inventors

San Francisco, CA, United States of America

Christopher Hillar