An image processing pipeline using coupled oscillators

Image Processing Pipelines (IPPs) are systems that iteratively modify an input image into a series of abstractions with the goal of identifying the objects in the image. There have been many proposed IPPs, implemented in both hardware and software, but by and large these designs have been based on Boolean logic based computation and arithmetic. Studies have shown that Boolean computers are hitting a theoretical ceiling on their performance in terms of transistor size, energy consumption/heat dissipation, clock rates, and by extension, computation time. Due to these issues, researchers have proposed using non-Boolean approaches, where possible, for various computations in common algorithms. One of the emerging technologies in the field of non-Boolean computation is the use of coupled oscillators. A proposed use of coupled oscillators is for pattern matching, which can be interpreted as a high-dimensional distance measurement. Using an approach based on using coupled oscillators as a computational primitive, this work utilizes the benefits gained from this new computational paradigm to gain performance in terms of both speed and power with respect to a benchmark IPP, without decreasing its image identification accuracy.