A woman walks to an automated teller machine (ATM) on a bright afternoon. As she punches in her PIN code, she does not notice a car pull up to the curb behind her. When she reaches for her withdrawal, a masked man jumps out of the car, grabs her money and her purse, rips the gold chain off her neck, jumps back in the car and speeds off.
The whole sad episode is captured by the ATM's video surveillance camera, but until recently, this crook would have got away - his face was covered; ATM cameras are usually fixed-focus in order to be able to document who is using the machine, meaning that the car's licence plate would almost certainly be out of range. To make matters worse, the surveillance video is often digitised and compressed (using MPEG or Motion-JPEG) and transmitted to a central location. These pixel-oriented, DCT-derived compression techniques compress the video by dividing the image up into blocks, usually 8 pixels square. So when a frame from this ATM video is magnified, it would likely break up into the notorious PEG mosaic of blocks.
However, our crook had the misfortune of mugging his victim in front of an ATM equipped with a video system exploiting Wavelet transform-based compression. Wavelet, of course, deals with compression on a frame-by-frame-rather than pixel-by-pixel-basis and delivers compression ratios of up to 100:1 with little loss of detail after decompression.
The bank security officers use a visual search feature built on the same technology to locate a frame showing the getaway car. Expansion software enlarges the frame until they can see the licence plate. They then apply de-noising software to filter out noise due to dirt on the camera lens. De-blurring and edge-enhancement software bring the licence plate number into focus, providing police with enough information to apprehend the mugger while he still has the lady's purse and gold chain.
De-blurring aside, all of the technologies described above are now being marketed by Kallix, a start-up that has been a pioneer in the practical application of Wavelet technology to the security industry and has recently introduced its first product, the DSMS-100, a low-cost, PC-based turnkey digital security management system. The DSMS-100 compresses and stores video from four NTSC inputs on its internal hard drive. Kallix's turn on Wavelet compression is at the heart of the product line - and of this turn on a fast-standardising technology.
Security goes digital
The use of CCTV for security has been increasing for many years as the cost of cameras, switches and VCRs have dropped and the cost of loss and litigation has grown. Security industry consultant Steve Kaufer from Inter/Action Associates (Palm Springs, CA) points to some specific reasons for the notable ongoing move from analog VCR to digital storage of these security application images:
The VCRs are relatively expensive to buy and maintain, Kaufer says, "especially when you have a large bank of them and 24 h monitoring. The picture quality is not all that good to begin with, and as the tapes and the heads get older, the quality gets worse. Even with a new VCR and a fresh tape, the image deteriorates quickly when you try to zoom in on part of the image. It can be difficult to recognise a face or read a licence plate number on an enlarged image."
By now, many companies are developing digital video systems specifically for security applications; Kallix appears to be one of the few, thus far, dedicated to using Wavelet transform technology to do the job. "We had to develop our own Wavelet compression chip." Robin Ong, Kallix's President, told us. "As far as we know, we are only the second company with a Wavelet chip."
"People in the digital world don't realise how exciting this feature is to people who have had to use video tapes." Kaufer pointed out "Finding a particular image on a video tape can be a frustrating and time-consuming task, even if you know the exact time that the event you want to look at took place."
Wavelet variations
Wavelet-based compression offers higher compression ratios and higher resolution than conventional MPEG and JPEG compression techniques. Here, Kallix adds its own 'spin' on Wavelet implementation.
As Dr Angel DeCagama, Kallix's Chief Scientist put it, "The standard approach to image compression using Wavelets is to discard (from a human perception standpoint) the least significant coefficients of the Wavelet-transformed signals after application of a set of Wavelet filters - a low-frequency filter and a high-frequency filter that are orthogonal to each other, corresponding to a specific mother Wavelet.
"Reconstruction of the original signal is accomplished by the inverse Wavelet transform using the available coefficients. The more coefficients are discarded, the higher the compression ratio, but the quality of the reconstructed signal is also more degraded. However, if the discarded coefficients are intelligently chosen, such degradation is essentially imperceptible up to a point and very graceful after that, thereby achieving compression ratios for a given desired quality well beyond what can be achieved with any other technique."
There is no industry standard for Wavelet compression, per se. In other words, you must decompress the images using the same proprietary algorithms used during the compression. (The JPEG 2000 Compliance Committee is working to include Wavelet image compression in the updated standard that they hope to release within a year.)
DeCagama spelled out this firm's approach in this way: "Kallix's approach is based on a patent-pending mathematical technique that allows the reconstruction of the low frequency part of a given level from a high-quality estimate of the low-frequency part of the next lower level without actual knowledge of the high-frequency part," he said.
"When this concept is applied to the highest level of transformation, the discarding of all the high-frequency coefficients results in an image (the low-frequency part) which is one-fourth the size of the original image. Standard Wavelet compression techniques can then be applied to further reduce the image. In other words, for image compression, Kallix's approach results in a reduction of data by at least a factor of four prior to applying to compression."
"Wu added that Kallix images retain excellent image quality, even at 100:1 compression: "This becomes important in applications where the customer may want to expand the image significantly."
100:1 video compression
Ed Wu, Vice President of Product Development for Kallix, said that the algorithm DeCagama describes is now implemented in a single chip that can perform realtime compression of four simultaneous NTSC inputs, achieving maximum compression ratio of 100:1 by combining frame compression with video compression.
"First we compress each frame using our proprietary Wavelet algorithms." Wu says. "This provides 4:1 compression. Then we use Wavelet transforms to compare each frame to the previous frame and only store the changes. This gives us up to an additional 25:1 compression."
100:1 compression means that a full-colour, full-size video frame (two 704 x 480 pixel rasters [fields] with 16 bits per pixel) is reduced from 675,8 to 6,8 kbyte. The 15 fps video typically used for CCTV in the security industry requires about 100 kbyte/s bandwidth. Of course, bandwidth can be substantially reduced by using half or quarter-size video frames and fewer bits per pixel.
Wu added that Kallix images retain excellent image quality, even at 100:1 compression: "This becomes important in applications where the customer may want to expand the image significantly." Users can select lower compression ratios with resulting higher image quality dynamically under manual or program control.
Search, authenticate, clean
The realtime search engine lets users locate any frame by time and date. It has a slide bar that provides fast (or slow) frame-by-frame search, forwards or backwards, of any desired time period. It also permits realtime playback.
"People in the digital world don't realise how exciting this feature is to people who have had to use video tapes." Kaufer pointed out, "Finding a particular image on a video tape can be a frustrating and time-consuming task, even if you know the exact time that the event you want to look at took place."
Kaufer said that the one issue limiting the adoption of digital imaging for security applications is authentication. "Everyone knows that digital images can be manipulated. How do you prove in court that your digital image is authentic and not modified in any way?"
Ong added that Wavelet-based compression is inherently authentic: "You can edit the uncompressed image, but you cannot edit the compressed image on disk." He explained that since Wavelet transforms compress the entire frame, any change - no matter how trivial - would make it impossible to decompress the image.
One of the more remarkable features of Kallix software is the ability to filter out noise due to dirt or dust on the lens.
"Wavelet transforms let us detect the relative depth of the image." Wu said, adding that unlike other de-noising techniques, the Kallix software can distinguish between specs due to dirt on the lens and specs which are part of the image. "If you scanned a winter photograph, our de-noising software could remove white specs due to dirt on the scanner glass without touching a single snowflake."
Image expansion
Kallix image expansion software can expand an NTSC/PAL video frame up to 64 times without loss of resolution, according to Ong. Wu emphasised that this is different from pixel-based software zoom offered by Windows and many graphics processing packages, which merely increase the size of the pixels, resulting in grainy images.
"We use the information in the Wavelet transforms to expand the entire image, or a selected part of the image, to produce high-resolution expansions," Wu said. "The pixels remain the same size. We just use more pixels in the expanded image." He added that when Kallix introduces their digital camera this summer, it will be possible to expand frames hundreds of times without loss of resolution.
De-blur software
Kallix expects to add de-blur ('focusing') software during the course of this year. This is an important feature since so many security cameras are fixed-focus. "Even when a camera is well-focused, frame expansion tends to exaggerate any misfocus," Ong stated.
For details contact Kallix on tel: (091) 978 392 9642, fax: (091) 978 392 5898, or e-mail: [email protected]
© Technews Publishing (Pty) Ltd. | All Rights Reserved.
printer friendly version