Feature

Sharper Images

Image processing opens up new vistas for researchers and added advantages for manufacturers.

By Errol Wood

Image processing techniques are providing researchers with a greater understanding of everything from large-scale weather patterns to how to make better potato chips. Mathematical manipulations and image enhancement procedures are used in image processing to glean more information from data than is conventionally available.

It is a key technology for remote sensing of the Earth's resources using satellite images, and for industrial inspection systems and robotic control. Medical imaging and microscopy make use of image processing techniques to learn more about the human body, how it functions and the things that attack it.

In New Zealand, image processing work is being carried out by a large number of organisations, and for a diverse range of applications. Canterbury University has gained an international reputation for its work, largely under the aegis of the late Professor Richard Bates, who died last year. His students continue to work on image reconstruction, finding ways of improving the quality of images which have been degraded by blurring.

At Victoria University, meteorologist Dr Jim McGregor has compiled a sequence of infrared satellite images that depicts the weather in the New Zealand region during September 1990. The images have been strung together in an animated video, showing the dynamics of low pressure systems and southerly fronts.

"Because animations highlight features not readily seen in still images, animated sequences have great potential as teaching aids, and have proved to be a very useful research tool for studying vast quantities of data," says McGregor.

Weather Replay

McGregor has found that varying the playback of the video allows different features to be examined in the complex swirl of atmospheric motion. Backward loops can be useful in enabling weather systems to be traced back to their source.

The first animation McGregor tried used film [see "The Weather -- The Movie, Nov 91], but video production has proved cheaper and simpler. Satellite data on magnetic tape was converted for use with an Amiga 1000 computer and processed to produce individual images. These were merged with a coastal outline to mark where New Zealand and Australia lay under the cloud pattern, and a clock and date feature added. The month of data fitted on the Amiga's 170 Mbyte hard disk. From there, a connection with a Super-VHS video recorder allowed animated sequences to be stored on video tape. A display rate of twenty-five frames per second was needed to provide smooth animation.

McGregor has begun work on producing a year's worth of satellite images, from July 1990 to July 1991, showing long-term weather patterns as well as shorter, local variations.

Another application of infrared satellite imagery is in the study of large-scale sea circulation.

Qing Wu of DSIR Physical Sciences is using sophisticated methods for deriving sea surface temperature and velocity data from time-lapsed satellite images. He's been concentrating, in particular, on an "anticyclonic eddy" system that persists near Cook Strait off the east coast of New Zealand. The mathematically manipulated satellite data corresponds closely to local measurements taken by a geophysical survey vessel.

The information obtained from such work is very useful for fisheries resource allocation. Different fish species prefer different temperatures. By being able to estimate temperature from satellite data, fisheries managers will have a more accurate understanding of the placement of the country's fish resources.

Stereo Sights

Just as human stereoscopic vision enables people to judge distance, a sequence of satellite images taken from slightly different viewpoints can be used to give height information. This can in turn be used to build up topographical maps of the Earth.

Such work is being done in DSIR's Image Processing Group, although the varying angles of satellite images, and the movement of the satellites themselves, can make such a procedure difficult and produce distortions. This problem is being studied by DSIR's Stephen McNeill. He is working on developing methods for correcting images to produce accurate stereoscopic pairs.

The images come from the SPOT satellite which passes regularly over New Zealand. The camera on board the satellite takes nine seconds or so to scan a 60 by 60 kilometre scene. In that time, McNeill says, the satellite wobbles slightly, causing the image's centre to move. While the satellite's actual movement is tiny, such wobbling can cause the image's centre to be moved 10-20 metres, marring stereoscopic matching.

By calculating the errors involved and correcting for them, greater accuracy can be achieved. The project has concentrated on satellite data, but McNeill says that the software developed for this can be used with aircraft or close range imaging situations.

Industrial Imagery

Image processing can also be used for down-to-earth purposes. Researchers at DSIR Crop Research have been looking at the problems foodstuff manufacturers encounter in producing a product of consistent quality from a raw material that is inherently variable.

Graeme Coles of DSIR Crop Research has been using image processing to provide objective measures of important properties such as colour, texture or size. In potato chip manufacture, reactions of certain sugars in the frying process tend to darken the chips. Manufacturers aim to minimise this darkening -- the lighter the crisp, the better, because artificial colouring can be added in a controlled way.

Conventionally, manufacturers check on the darkening by frying a few slices from potato samples and using a colorimeter to estimate the depth of the brown colouration. Image analysis can speed up this process and result in batches of more evenly coloured chips.

"To a consumer, the product is the individual crisp. The variation in crisp colour found in a sample is of considerable importance," Coles says.

In flour manufacture, it is important to have uniformity in the size of wheat seeds, otherwise some grains are ground more thoroughly than others in milling. Efficiency declines rapidly if there is significant variation in grain size, Coles notes, and often 10-15% of the raw material has to be discarded as unfit for milling.

"Image analysis is easily able to produce size data on large numbers of individual wheat seeds in a short time," Coles says. The data can then be used to select the best seed and the most appropriate management techniques to reduce size variability, and to reduce the need to tune milling equipment for various sizes.

A joint project between Streat Instruments and the Wool Research Organisation of New Zealand involves the development of an image analysis system for monitoring the colour of the wool in a scouring plant. In scouring, it is vital that wool achieves its maximum whiteness, as this affects the price and market of the wool.

New instrumentation to assist in process optimisation has been developed by Dr Bryan Thomas, a WRONZ researcher on secondment to Streat. Automating colour analysis removes the problems of subjective, human-based decisions, and can provide the scourer with a tool for continuously checking the effectiveness and consistency of the washing process.

Apple Automation

Research student Brent Foster, of Massey University, has designed an image analysis system for apple inspection, in particular to detect stem end split common in the popular Gala, Royal Gala and Fuji varieties. Such splitting is a major cause of concern for apple growers, as it presents a source of decay.

The New Zealand Apple and Pear Marketing Board quality standards allow a maximum of 2% of apples in a box to contain splits. If a sample shows a greater proportion than this, the entire batch has to be checked and repacked.

"Automated visual inspection systems offer greater consistency than human graders, since a machine does not suffer from fatigue or boredom," notes Foster.

Machine vision does not require the same level or degree of lighting needed by a human checker. Foster used diffuse lighting and infrared wavelengths to improve the visibility of splits, particularly in the striped Gala and Fuji apples. Stem end splits radiate out from the centre of the stem cavity, and can be detected as a long vertical feature on a processed image.

Massey Professor Bob Hodgson believes that New Zealand can make wide use of image processing techniques and machine vision technology to competitive advantage.

The skill base and specific industries where the technology could be usefully employed are available, he maintains. What is now required is some coordination of effort to ensure that the technical information is disseminated and limited resources are most efficiently utilised.

He hopes to sees the establishment of a Machine Vision Users and Suppliers group which could advise on a wide variety of applications, and a consortium of allied interests making the best use of image processing and machine vision technologies.

The dairy industry could make effective use of such technologies in cheese inspections and milk powder quality assessments. Forestry could inspect timber, boards and paper, and the wool industry look to carpet texture analysis and fibre inspection.

"[This] technology has matured and could now be profitably applied to improve New Zealand's position in international trade," Hodgson says positively.

Errol Wood is principal scientist with the Wool Research Organisation of New Zealand, Lincoln.