What now after MPEG?

It is fairly obvious now that whenever some film is made such as home movies etc, the amount of tapes can pile up very quickly because of more storage requirements. Imagine the amount of storage required by businesses and small firms that need to capture every second of the day for security purposes, the requirements increase considerably. Although transmitting full-motion, full-colour video at 30 frames per second(fps) is like 40 books per second, it can be quite tough sending this kind of data over networks and requires hugh bandwidth.

However, new techniques in Wavelet video compression are starting to make video capture and transmission much more palatable. MPEG achieves compression rates between 30:1 and 100:1 while wavelets compress video at rates up to 300:1. This means that wavelet compressors can eliminate more data while maintaining video quality, resulting in less data storage and transmission over thin telephone lines (designed initially to carry words at a couple of words per second) or Internet. Wavelets compress video faster than other compression engines, producing less delay between compression and viewing. Wavelets also give higher compression while preserving the basic quality of the signal.

Currently, companies like Summus and InfinOp Inc., of Denton, Texas and Compression Engines LLC of Houston are making headway with wavelet video compression. With the increased demand and use of multimedia, video wavelet technology may well become the next great utility. At the moment, those dealing with videoconferencing applications, video on demand and video security are most likely to be interested in this kind of technology.
Summus have already announced their second-generation video compressor that officials say is better than twice as good as its original model. Although still-image compressors based on wavelet technology are available, the main focus now is on video.
 

MPEG compared to WAVELETS

MPEG like its cousin JPEG, is based on the Discrete Cosine Transform (DCT) which is a mathematical formula that separates images and video into 64-bit blocks and then compresses those blocks. Wavelets on the other hand, compress pixels in a continuous stream. Although the maximum possible compression rate for MPEG is about 100:1, most implementations are somewhere between 30:1 and 60:1. The reason for this is that our eyes are very sensitive to horizontal and vertical lines, and the 'blocking' compression used by MPEG results in "little teeny tiles that make the image look like it has some disease,.." said Mr Fisher from InfinOP company in Texas. But, lower compression ratios help to reduce the effect.

On the other hand, the continuous pixel compression technique used by wavelets allows them to drop more visual information and still retain the overall image. The result is like a watercolour painting where some of the edges run together. At low resolutions, wavelet-compressed images look soft and muted. But viewers are less distressed by that than by jagged tiles. Although wavelets are implemented with less cost than MPEG, on the downside, wavelets are a bit slower than MPEG.

Summus, for example, have found a way to sharpen the resolution of a particular item within a compressed image so the eye focuses on that. For practical applications, the user could sharpen a hand moving over a whiteboard to give students a clear picture of the teacher's notes. InfinOp have developed  two wavelet compressors, one for video on demand and another for real-time video. The former, LSVideoN allows users to compress video for storage on a server and the latter LSVideoR compresses and displays video on the fly for applications in teleconferencing, security monitoring and emergency cameras used in ambulances. The company has also achieved compression ratios of 250:1 with video clips that have a dark background and not a lot of motion, which could be ideal for surveillance in dark hallways, teleconferencing from a wood-paneled  boardroom or in one-to-one videoconferencing between users at their desks.

Nevertheless, some people do not see wavelets completely taking over MPEG since MPEG has been around for quite some time now and has maintained a good reputation and proved useful to many. Indeed, the JPEG 2000 compliance committee is already making way for wavelet image compression in the updated standard it plans to release at the turn of the century. Most experts expect the MPEG committee to do likewise with wavelet video compression. It is not wrong to say that standards would only help the wavelet market since without them users cannot use a compressor from one company and a decompressor from another. For now, companies such as Summus, Compression Engines and InfinOp (as well as their customers) must make do with proprietary solutions.
 
 
SUMMUS Technology

Summus's Wavelet Image (WI) is a digital image compression format whose quality versus compression ratio is the best available today. Speed is not sacrificed for WI's high quality. Summus's WI Compressor and Decompressor are faster than both JPEG and fractals. In addition, WI offers access to and control of image information that is not possible in other image formats. Some of these are Regional Focusing, Embedded Image Enhancements and Flexible Progressive Decompression. Summus's Wavelet has proven to be superior both in speed and quality to JPEG, fractals and other wavelet based compression methods by several independent studies. The foundation here is Speed, Quality and Flexibility.

The following results show how Summus Wavelet compression format performs in terms of Speed alongside JPEG and fractals :
 

The following results show how Summus Wavelet compression format performs in terms of Quality alongside JPEG and fractals :
 

 

A set of samples were used and tested by Summus themselves which yielded some results that show the speed and quality gained using their technology. A test image was used (640x480, 24-bit) and run on a 486 and Pentium-90 computers. The results are summarised below and truly show the capabilities of the Summus invention.
 

Although the techniques are not mentioned, that have been used to produce the above results, the results above do show the capabilities and benefits of the Summus Wavelet Image Format (WI) and clearly indicate that this could be the new standard if MPEG drowns.
 
 

4U2C WAVELET IMAGE COMPRESSOR - by SUMMUS Technology

The 4U2C Wavelet Image Compressor by Summus Technologies uses the latest Wavelet technology which enables images to be zoomed at very high percentages giving good quality at the same time. Files containing image data can be compressed upto a ratio of 1000:1 and can be stored in a very small storage space. Obviously the quality will be lost but compared to other technologies, this one achieves higher compression ratios until the quality is really bad.
Below is some information about 4U2C and some of the things it specialises in:

What Is Progressive Decompression
Progressive Decompression is the ability to decompress a Wavelet Image in parts. The advantage of this is that intermediate images can be displayed at any time during decompression.  This allows quick preliminary viewing of large Wavelet Images.  This is particularly useful for viewing images over the Internet.

Example

The Wavelet Image in this example is 4500 bytes.  It was progressively decompressed in 1500 byte increments.  See how each 1500 bytes of the WI file improves image resolution.
 

For instance, suppose the file size of a Wavelet Image is 20,000 bytes:
 
 

a buffer size of 2000 gives 10 intermediate displays.

Decompressing a Wavelet Image progressively does not slow down decompression.  New WI image data builds upon decompressed data without having to re-process it.

Wavelet Focussing

Wavelet Focusing (sometimes called user defined region of interest) allows selected image regions to maintain higher visual quality than non-selected regions.  You can use focusing to create better looking images at a higher compression ratio.

Example

See the effects of focusing and no focusing:

(no focusing - highly compressed without focusing)            (with focusing - highly compressed with focus box)

Notes

Focusing does not slow down compression or decompression.

Focus regions can overlap.  Overlapped regions will not be doubled focused.

There is no limit on the number of focus regions.

The high quality in the focus region is maintained at the expense of the surrounding non-focused region.
 

Wavelet Magnification

Wavelet Magnification enlarges images:

* without increasing WI image file size
* without introducing pixel replication block artifacts
* without increasing image compression time.

Wavelet Magnification reduces the time it takes to send larger sized images over the Internet.

Example

If a 240 x 320 image generates a 7600 byte WI file size, then wavelet magnifying the image to 480 x 640 at the same compression settings will also generate a 7600 byte WI file size.
See below:
 

It takes the same amount of time to send both images over the Internet.

Notes

Images cannot be resized smaller than 16 x 16 or larger than 4096 x 4096.

Each click on the magnification control increases or decreases the image size 2 times.

The software was tested by using a self-captured image and then compressing it several times at various compression ratios. The file size was observed and the data recorded and presented as a bar chart.
Details of the original image can be seen below along with the image itself. The bar chart is shown also.
 

In terms of quality, for ratios of 20%, 40%, 60% upto 240% the image was recognisable and had fairly good quality but at 300%, the image became smudged and blurry after which further increments in ratio were stopped.

A mean-opinion-score was taken from a number of colleagues who judged the quality of the image at various ratios. These results can be seen after the bar chart. (File-size in bytes vs Compression Ratio) 
 

There were 20 volunteers who judged the image quality for various compression ratios. The rating was out of 5 and the mean score is taken at each point. The outcome of this experiment can be seen below: