Scanning and imaging identify wood flaws

April 1, 2001
To improve wood production yields, university researchers are imaging, analyzing, and modeling color scans of sliced logs for flaws and blemishes.
To improve wood production yields, university researchers are imaging, analyzing, and modeling color scans of sliced logs for flaws and blemishes.

By Joe Hallett,Contributing Editor

Forestry, timber, and wood products constitute a major industry in the United States. In support of that industry, several universities are investigating growing, harvesting, and cutting trees into boards, plywood, millwork, and other materials. At the same time, forestry researchers are looking at optical signatures and computed-tomography (CT) scans of wood to increase production yields. Several university projects also are being directed at small-diameter logs that are more readily available but are more difficult to break down into usable material that is free of defects and blemishes. These projects are being carried out at Virginia Tech (Blacksburg, VA), Oregon State University (Corvallis, OR), and Louisiana State University (Baton Rouge, LA), among others.

Multispectral imaging to identify surface defects and CT scanning to examine interior defects are two promising tools that are under study at the Oregon State University forest-products department. This college laboratory was established in the mid-1980s to perform basic research into analyzing images of wood surfaces. Under the direction of James Funck, the lab's current studies include multispectral analysis of surface defects and blemishes and interior modeling based on color scans of sliced logs. These studies are trying to determine log suitability for the cut-stock (millwork) industry, which produces small, clear wooden parts.

Funck explains, "Small logs are driving our work in trying to get higher value assigned to timber. We're looking at 6- to 10-in.-diameter logs that are not straight. These logs are hard to use and cut. We look at log-breakdown models, such as where to make cuts based on the estimated value of dimensioned lumber. But now we're also looking at a new model where the driving [market] force is the price of cut stock. We're also looking at logs on a screen to find clear areas. We've concentrated on three species that are primarily common to the Pacific Northwest (Douglas Fir, Ponderosa Pine, and Red Alder), but we have worked with others as well."

Click here to enlarge image

FIGURE 1. In the Oregon State University imaging laboratory setup, a Pulnix M-7CM CCD gray-scale camera moves along a track and views wood samples through an eight-position filter wheel. Each image looks for differences in very fine features in the wood samples. The Targa image-capture board can put four frames in color on the computer screen for side-by-side comparisons.

According to Funck, there are many unresolved issues related to scanning. "The variability of wood surfaces is so great that transferring the skills of human inspectors to vision systems has been a major problem. The industry wants tighter tolerances. We're trying to identify the problems. Then it's up to the equipment people to come up with final implementations," he adds.

The projects are being conducted with relatively modest budgets, but by careful selection of imaging tools and in-house development of specialized software, researchers have been able to make important contributions to understanding the new technologies. "We are funded from an industry-paid harvest tax," says Funck, "and also by federal and private research grants.

FIGURE 1. In the Oregon State University imaging laboratory setup, a Pulnix M-7CM CCD gray-scale camera moves along a track and views wood samples through an eight-position filter wheel. Each image looks for differences in very fine features in the wood samples. The Targa image-capture board can put four frames in color on the computer screen for side-by-side comparisons.
Click here to enlarge image

"Our only compromises have been based on cost, which tie us to old Targa (now Pinnacle Systems) image-capture boards and related software," says Funck. "The boards run under a Windows 98-based PC. All the equipment is networked to an internal data server. This setup saves data in our version of the Targa format using 32-Mbyte removable drives with multiple backups for redundancy. Each drive is mirrored twice, with backups stored off-site."

Two different laboratory setups are used for the multispectral imaging tests and the color scanning for log defects. Although the two setups have many common elements, the systems are physically separate, using a half-dozen Targa boards.

"The Targa boards aren't limiting," says Funck, "but they are obsolete and can't be replaced. So, we have invested time in custom C++ programming. Changing to the newest versions of Targa boards would mean significant changes to our software," he reports.

Because human inspectors work with speed and accuracy despite the fact that every piece of wood is different, Funck reasoned that sensitivity to color might be a key factor. "A human inspector works faster with color, so our imaging work has been color-based. In the early 1990s, we realized that we had pushed the standard red-green-blue (RGB) color space to its limits with standard algorithms. So now we're looking at four or more colors in the 400- to 1100-nm range. When using more than six to eight channels, however, the gains are minimal. Our goal was to use stock optical filters, since custom ones were too expensive," he adds.

FIGURE 2. Multispectral images of a wood surface show different spatial signatures that can be used to identify defects under a surface feature classification. Each slice of filter-segmented data represents what the camera sees in a 50-nm band centered on several wavelengths. The illustrated defect is called white speck.
Click here to enlarge image

In the college's imaging laboratory setup, a Pulnix America Inc. (Sunnyvale, CA) M-7CM CCD gray-scale camera views wood samples through an eight-position filter wheel (see Fig. 1). Each image consists of 256 x 200, 0.2-mm pixels. "We are looking at differences in very fine features," says Funck. "The Targa board can put four frames in color on the computer screen for side-by-side comparisons."

There have been some problems with the setup that Funck expects could be corrected by a true multispectral camera replacing the filter-wheel system. "We've had difficulty with image registration because of differing depths of field for the images. And lighting is an important issue. A multispectral camera would have specially designed optics and would take care of registration and intensity differentiation in images in different planes."

In a soon-to-be-released study, Funck and his associates report on multispectral imaging using bands of color within the spectral range of a CCD camera for mapping various types of wood defects (see Fig. 2). "The beauty of our work," says Funck, "is that segmentation and classification of defects are done at the same time. Species can be differentiated, which is hard for a human inspector, especially for white woods."

Computed tomography, long known for medical uses, has also gained use in nonmedical, noninvasive procedures, such as airline safety inspections. Now CT scanning holds promise as a means to analyze a wooden log before initial cutting takes place. "Eventually, log scanning will be tied to CT scans," says Funck. "For now it is just used for developing log-breakdown algorithms. We are working with others who do CT scanning."

FIGURE 3. Multiple views of a wood sample show the various segmentation levels on an image of veneer. These views are obtained by a color camera using traditional segmentation algorithms.
Click here to enlarge image

Explains Funck, "Normally we scan regular lumber and veneer, but we use sliced logs for checking internal features. We cut logs into thin strips and scan each side. Then we map and recreate the log (in the computer) from which we make a log-breakdown model. We get dimensions; use shape, color, and texture to identify defects; and segment the image into blobs of similar-looking pixels. Then, we produce a new image with features based upon analysis, such as pixels above threshold and clusters, that can be used for classification of defects such as tight knots, loose knots, pitch streaks, and stains. However, annular rings are hard to separate from defects. We've tried different segmentation algorithms. All are application-specific for dry veneers, cut stock, and so forth, and varying levels of defects are allowable," he says (see Fig. 3).

The study uses relatively low-cost components to conduct the wood research. Starting with tube-based cameras, the latest iteration of the laboratory equipment uses a Hitachi Denshi America Ltd. (Woodbury, NY) HV-C20 three-chip RGB CCD camera and halogen lights from Thermo Oriel (Stratford, CT). The camera is mounted on an overhead track where it is driven under software control over the length of a log. It provides a 256 x 200-pixel image at between 15 and 30 pixels/in. Since the goal is to determine the internal characteristics that might suggest a particular cutting path for greatest yield, logs are sawed into slabs about 1 in. thick.

After all the slabs have been scanned, a three-dimensional image of the original log is reconstructed that shows the relationship of surface imperfections to unusable areas within the log. More important, it allows the identification of clear regions where the wood is suitable for precision cutting. Algorithms that result from this work will be extended to benefit other researchers and to firms such as Inovec (Eugene, OR), which are trying to adapt high-speed CT scanning to the log-breakdown process.

Researcher Philip A. Araman at Virginia Tech says, "Others have been interested (in CT scanning of logs), but InVision Technologies Inc. (Newark, CA), through its newly formed company WoodVision, has been the main player in North America. We have worked with other researchers but they are not trying to create a commercial machine-vision system because of high cost. Our research is continuing."

Industry sources indicate that the advantages of CT scanning will be seen first on small-diameter logs, with the possibility of a "breakthrough" this year. Speed—which relates to process cost—is seen as the main limiting factor, and it becomes a bigger problem as log diameter is increased.

Says Jeff Franklin, Inovec marketing vice president, "The speed is high enough to address the needs of the wood-products industry. Now, CT scanning can get a dense amount of information—even grain structure—prior to opening up the log. It can change the ways that the industry buys and trades logs."

Multispectral imaging and computed-tomography scanning to examine exterior and interior defects of wooden planks are under study at Oregon State University. The university's laboratory is analyzing color scans of sliced logs to determine log suitability for the cutstock (millwork) industry.


Hitachi Denshi America Ltd.
Woodbury, NY 11797

Inovec Inc.
Eugene, OR 97402

InVision Technologies Inc.
Newark, CA 94560

Louisiana State University
School of Forestry, Wildlife & Fisheries
Baton Rouge, LA 70803

Oregon State University
Forest Products Department
Corvallis, OR 97331

Pinnacle Systems
Mountain View, CA 94043

Pulnix America Inc.
Sunnyvale, CA 94089

Thermo Oriel
Stratford, CT 06615

Virginia Tech
USDA Forest Service/Southern Research Station
Blacksburg, VA 24060

Voice Your Opinion

To join the conversation, and become an exclusive member of Vision Systems Design, create an account today!