UtopiaCompression Corporation

Pattern-Driven Modeling and Compression

Image & Video Modeling and Compression

According to UC, compression should be concerned with novel ways of representing visual patterns (simple and complex) using a minimal set of extracted features. This view requires application of computational geometry and computational intelligence, in particular Machine Learning (ML), to extract primitive visual patterns unique to the class and pertinent to compression. The codec is then trained on image representatives of the class to generate a knowledge base of such primitive patterns so that at runtime coarse grain segments of the image can be accurately modeled without further decomposing the segments. This approach should give rise to significant improvement in compression performance. An adaptive pattern-driven approach mines pertinent features belonging to patterns in a class, which upon training can better model patterns.

UC’s pattern-driven codec, upon trained on an image repertoire from a class of images having similar characteristics delivers superior and faithful compression performance. Focusing on the extraction/modeling of primitive patterns and their features, the pattern-based codec is designed to be adaptive to class characteristics of patterns. Adaptation to patterns defining image classes leads to:

Significantly enhanced compression performance (achieving high compression ratios for extremely high reconstruction qualities in an efficient manner)
Generate embedded security unique to the imagery

UC has:

Developed intelligent tri-partite filtering technologies for achieving pattern-driven image compression
Evaluated pattern-driven technologies against current data-driven technologies
Demonstrated the core intelligence based compression technology using different classes of imagery such as SAR, Infra-Red (IR), aerial and others.

Application of UC’s compression technologies to three DoD programs: JASSM, Predator and F/A 22

The efficiency and effectiveness of UC’s innovative compression technologies has been demonstrated for a number of Government sponsored programs.

JASSM (Joint Air-to-Surface Standoff Missile) – Transmit IR images to shooter via command/control data link for last-moment re-targeting/abort, bomb hit indication, and maritime interdiction
MQ-1/9 Predator – Increase data link capacity by providing greater compression performance than existing technologies
F/A-22 Raptor – Compress ground attack Synthetic Aperture Radar (SAR) image database while maintaining image integrity

JASSM Program

Compression Rate = 109 for UC and JPEG2K Compression ratio around 57 for JPEG

UC’s JASSM edge-modeling based compression technology is tailored to extremely tight bandwidth applications for Battle Damage Assessment (BDA), weapon re-targeting/abort and maritime interdiction. This application can be extended to transmit imagery data to handheld devices used by the Marine Corps in the battlefield. The sharpness of features retained in the UC compressed images allows a man-in-loop to assess a target with high confidence.

Predator Program

Compression results for Aerial Imagery

As a part of our Enhanced Image Capture and Transmission (EICT) program, UC applied its compression algorithm to aerial images shot from predator aircrafts. The images have sky in the upper half of the image which is normally smooth. The lower half is usually ground and higher textured. UC codec performed significantly better than the JPEG and JPEG 2K at high PSNRs for this class of images.

F/A-22 Program

Compression results for SAR imagery via SRAD processed (left) and segmented (right)

UC’s F/A-22 codec makes it possible to store more Synthetic Aperture Radar (SAR) images for precision ground-targeting operations with existing F/A-22 data storage hardware. The F/A-22 codec exemplifies efficient pattern-driven compression of homogeneous areas and texture regions. Since SAR imagery suffers from coherent speckle noise resulted from backscatter interference of randomly positioned scatterers in radar imaging resolution cell, all methods of data compression that rely on exploiting spatial correlation (redundancy) in the image are not immediately applicable. UC has developed a suite of speckle suppression approaches to reconstruct underlying cross-sections before compressing the data. One of these methods producing quasi-optical despeckled image is the Speckle Reducing Anisotropic Diffusion (SRAD) algorithm. Segmentation based speckle reducing method helps obtain optimal compression performance with UC codec.