ThinWave is a wavelet-based approach to lossy compression of images and audio signals
ThinWave is a wavelet-based approach to lossy compression of images and audio signals. It shares some features of DCT (Discrete Cosine Transform) approaches to compression in that it transforms the signal to a different basis of representation when encoding. DCT techniques perform a type of discrete Fourier transform on a signal, which maps it into an orthogonal basis of sine functions- i.e. a space where signals are represented by the sum of their constituent frequencies. In general, to map any real signal into frequency space, it is necessary to map the signal into a space of complex exponentials to preserve phase information. Thus compression schemes based DCT, such as JPEG, have to do many small transformations on small (8X8 is typical) pixel blocks so that the phase information lost is implicitly recovered to some degree by the agreed on ordering of these blocks.
A wavelet transform also maps a signal to a function space, but instead of sines and cosines, the basis functions are , function that unlike sine and cosines, have compact support, that is they are zero almost everywhere. This "local" nature of wavelets makes it possible for the wavelet transforms to store the phase information far more compactly, allowing the encoding of far larger blocks of data, such as an entire picture or audio track, than the DCT methods allow. The "block" nature of DCT coding is especially apparent in motion video because the eye tends to be very good at picking out these stationary blocks in a motion picture.
ThinWave uses orthogonal wavelets, in a Mallat pyramidal scheme, to produce a wavelet transform of the signal. This results in a representation of the signal where most coefficients are nearly zero. At this point the signal still contains all the original information- that is, it could be inverse wavelet transformed and the exact original would be recovered. To achieve higher compression rates however, every lossy scheme uses some sort of quantization scheme to turn the floating- point numbers that represent the transform (be it wavelet or DCT) into integer values.
ThinWave uses our variant of a scalar Lloyd-Max quantizer (also known as optimal-alphabet-constrained-quantization) for this purpose. This step is where data is lost, resulting in "lossy" compression. After quantization, RLE coding is applied to the now, very sparse (almost all zeroes) signal, followed by an entropic compression scheme such as Huffman or arithmetic coding. Our RLE-Entropy coder scheme uses a variant of the public domain FBI's AFIS (Automated Fingerprint Identification Standard). This system which is able to store grayscale fingerprint images at compression levels of 20:1 in an automated database and yet be able to please the experts that look over these images in minute detail. Not only is wavelet compression good for storing compressed fingerprints, but because of the multi-resolution nature of wavelet transforms, they are a good way to encode any signals that may be archived in a searchable database.
ThinWave is a set of algorithms, application programming interfaces (APIs) and applets that collectively support wavelet based audio and image compression functionality. These tools can be integrated into customer applications as needed to provide state-of-the-art compression of sound and picture with restoration fidelity better than any other method available today. By exploiting WavePack customers can significantly cut bandwidth costs for propagation of audio and image data.
The Company's implementation of wavelet audio compression methods allows compression down to 1.5kbps per full duplex voice channel with excellent subjective quality. ThinWave's unmatched efficiency in transmitting audio data can play an important role in the deployment of wireless IP services, particularly for wireless VoIP, and can serve a crucial role today in improving the performance of existing VoIP deployments, which typically occupy 5-7kbps of bandwidth per channel. The Company intends to license this technology to OEM and VoIP network services.
The Company also believes there may a significant opportunity to exploit ThinWave and DSP technology to dramatically shrink the occupied bandwidth of two-way radio products.
Also in the ThinWave suite are image compression methods that allow compression of complex image data to levels of 50:1, with less complex scene data allowing levels to 200:1. ThinWave image compression technology is particularly competent in dealing with scene data presenting many fine details, and performs significantly better in this respect than JPEG2000. ThinWave can reduce a 7-megabyte image file to 155 kilobytes and then restore it with details intact. Such compression levels allow, for instance, 40 24 bit SVGA full screen images to be stored on a single floppy disk. Again, the Company intends to license ThinWave image compression technology through OEMs to content providers and others who need an unbeatable combination of highly aggressive compression and very faithful reproduction of image data.
The company intends to vend licenses to DLL and other library forms of the ThinWave algorithms, as well as to license actual source code where terms permit.
BroadIP Networks, Inc., 7828 Flager Circle, Manassass, VA 20109. Tel: 800-555-1141; Fax: 703-288-3565.