University of California

Jeng-Hau Lin

Rajesh Kumar Gupta

Zeyu Chen

Yunfan Yang

Zhuowen Tu

Disclosed are devices, systems and methods for providing efficient implementations for pattern recognition. Embodiments of the disclosed technology use local binary comparisons and random projection in place of conventional convolution operations. Some embodiments of the disclosed technology provide an important means to improve memory and speed efficiency that is particularly suited for small footprint devices and hardware accelerators. Other embodiments compare a binary pattern to one or more portions of an input image to determine a plurality of binary results, and generate an estimate of the feature based on the plurality of binary results. Yet other embodiments generate an final image, comprising a final estimate of the feature, by randomly selecting one or more portions from each of a plurality of initial images, wherein each of the plurality of initial images comprises an initial estimate of the feature.

Growing community of inventors

La Jolla, CA, United States of America

Jeng-Hau Lin