Sparsense is developing novel algorithms (SparseCode) for sparse sensing, as well as its own data processing hardware (SparseBox) optimized to run SparseCode algorithms.

Sparse sensing builds on the fact that most of the physical, real world signals (e.g. images, spectra, sound, temperature, etc.) are compressible. For effective compression the building blocks or “components” of the signal must be defined. Sparsense’s technology makes it possible to discover these components in a signal. Once the components are known, Sparsense can generate a sparse code for a signal or measurement. Sparse coding is beneficial, because
- fewer measurements are sufficient to reconstruct the entire signal,
- sparse code filters out noise, as unstructured noise is not reconstructed,
- sparse code is a compression of the signal.

Sparsense’s tests to date, performed on artificial databases and real life image processing tasks, have confirmed that SparseCode algorithms are more scalable and faster than current leading sparse coding techniques such as L1 Magic or factorization methods. They have also shown excellent potential in the processing of noisy data.

...Copyright: Sparsense 2008
Photography: John Pitcher...