The University of Auckland
2 files

Pictorial Representation of cP systems implementation of basic Loopy Belief Propagation for Stereo Matching

posted on 2020-09-24, 05:33 authored by James CooperJames Cooper
High-level overview of the cP systems Belief Propagation algorithm presented in a paper by me. From top to bottom: Pixel input values are received from the camera source and converted to appropriate cP systems cells (represented by the blue connectors passing into the black converters). These are then passed into the Data Cost Cell, which computes the data costs for the entire system at once, before distributing the costs to the Messaging Cells, which represent independent processing elements that each compute the final value for a single pixel in the disparity map. The Messaging Cells then exchange messages and update their cost estimates appropriately, before sending their final disparity estimates to the output cell for conversion back to pixel values and output to the receiving device.

Both formats are the same image, they are both included for your convenience, and because figshare apparently doesn't render .eps files in-browser but .eps seems to be the required vector graphics format for many publications. I wanted to include SVG, but for some reason I couldn't seem to produce a good SVG version of this image, so you get the pdf version that was automatically generated by Ghostscript when I ran pdflatex.

This image is unlikely to be of any real relevance to anyone else, but I am posting it up here to ensure that it can be re-used if needed.



University of Auckland