Image Measurement for Quantification of Deterioration
Wood is widely used for structural purposes. Its strength gives it the ability to transmit loads. Structural analysis of the material is based on several factors, including the cross-sectional area of individual members. Members that have lost cross-sectional area through decay or removal (e.g., drill holes, notches) may be compromised based on both the extent of the reduction and its location in the load transmitting assembly. Radioscopy is a way of assisting the field investigator in identifying and possibly quantifying such loss. The ability to do so can contribute to subsequent analysis and recommendations for remediation.
The visual analysis and representation of decay in wood proved to be more challenging. Figure 73a depicts higher density wood (with more solid wood) as darker, and lower density (with more decayed wood) as lighter, but without much contrast. Simple contrast manipulation, using Adobe® Photoshop® (Figure 73b) yielded an erroneous pattern (possibly due to the hotspot problem previously discussed). However, an unsharp mask (Figure 74) combined with overlaying a color mask (Figure 73c) based on substituting a transparent mask of contrasting pseudo-colors makes the image somewhat more intelligible. The drawback of this approach was the necessity of multiple transformations, and the time and effort required in each. Clearly, a better approach was needed.
Quantification of Deterioration
In the earlier discussion of deterioration we showed that it was relatively easy to identify the presence of deterioration and loss of section. In an effort to try to quantify loss of section due to decay or insect damage, it was decided to begin with a simple case, the stepped block seen in Figure 39. This was a known situation with discrete boundaries separating the sections, which ranged in thickness from 10 inches to 2.5 inches. Radiographs were taken using a standard source-to-object distance of 24 inches. Successive radiographs were taken with different pulse numbers, ranging from 16 to 100.
To find a user-friendly method to quantify the range of tones in these radiographs, we elected to use a readily available software package, Adobe® Photoshop® Elements. This package (along with most other photographic- enhancement software) has a histogram function that identifies the range of tones in a radiograph by graphing the number of pixels at each color intensity level (which are grey tones only for the radiograph). The number of pixels is represented on the vertical axis and the horizontal axis ranges from the darkest values (at zero) to the brightest values (255 for this package). For a given radiograph, the thickest section (10 inches) thus had the lowest range of histogram values, while for a specific thickness, the mean histogram value increased as the number of pulses increased.
One of the advantages of using the histogram function to quantify decay is that it can be used with the Rectangular Marquee Tool. Without selecting any portion of the radiograph, the histogram function displays the histogram of the entire radiograph. But each section can be outlined with the Rectangular Marquee Tool, and then the histogram function can be employed. For each radiograph, this was done, using the entire section at 100, 75, 50 and 25 percent of the initial cross section, and taking the mean and median values for each histogram. The results are displayed in Figure 75.
This established that there was an excellent correlation (R2 values were all 0.99 or above) between the mean value and the percent of remaining cross section at each pulse level. However, this information was useful only for one species, using one particular setup and only for these thicknesses. It was felt that in a typical field situation, a more realistic method to determine loss of section might include an identification of intact wood versus decayed wood and a comparison between them. To this end, the ratio of the median histogram value for sound wood (the 10-inch section) to the median histogram value for other thicknesses was calculated, and plotted against the percent of remaining section. This is shown in Figure 76.
The plots of the histogram function, when using the ratio of the reading at 100 percent to the reading taken at 75, 50 and 25 percent all collapsed to essentially one function. This is important in that the number of pulses are not a factor when trying to determine loss of section in a timber as long as ratios are used, rather than raw values.
However, there were questions about the applicability of this graph to other situations. The primary concern was for the hot spot effect. This effect is clearly visible in Figure 40, the colorized version of the stepped block. It was also apparent when the histogram function was used on smaller sub-sections across a portion of the radiograph representing only one thickness, with higher values found closer to the center of the hot spot. It is possible that the graph above might be more linear, since the 10-inch section and 2.5-inch section of the step block were on the outer edges of the radiograph, and hence had median histogram values that were lower than would be expected if they were in the center of the radiograph due to the hot spot.
To address this problem, the radiographs were examined visually, to determine the probable center of the hot spot. It was then decided to use only the top 15 percent of the 7.5-inch and 5-inch sections, which appeared to have approximately the same relationship to the center of the hot spot as the entire portion of the 10-inch and 2.5-inch sections. To make this a relatively uniform measurement, the grid function was superimposed on the radiograph prior to measurement. The results of this analysis are shown in Figure 77. While there is still some curvature in the graph, it does appear to partially correct for the hot spot problem, with all histogram ratios following approximately the same trend line. To completely eliminate the hot spot problem, additional investigation would be required. The masking of the hot spot discussed above would not necessarily correct the histogram appropriately.
To validate the potential relationship, several radiographs were used. For the initial set of radiographs, another ratio was calculated. This was the ratio of the median histogram values for the 7.5-inch and 5-inch sections, using only small sub-sections of each near the center of the radiograph (and thus the center of the hot spot). This ratio, which represents 1/3 loss of section if the whole timber was 7.5 inches thick, was done for all radiographs taken, with pulse values ranging from 16 to 100. As expected, there was a very small range of values, with the average value shown in Figure 78 (labeled 67% Average), plotted against the averages of all the lines plotted in Figure 77. Further, measurements were taken on the radiograph of the decayed log in Figure 43, and of the timber with simulated termite damage, shown in Figures 14 and 15 (labeled TD 1 to TD 4).
Finally, a linear function, showing what might be the effect if the hot spot problem was completely eliminated was plotted as the heavy black line on Figure 78. The equation for this line is:
- HR is the histogram ratio (the median value of the histogram for “intact” wood divided by the median value of the histogram for “damaged” wood)
- RS% is the percent of remaining section
Rearranged to solve for the percent of remaining section, the equation is:
This research implies that a simple histogram function might be reliable to estimate loss of section, if small adjacent sections (within the same area of the hot spot) were used to compare intact to damaged wood. However, it would require more research to identify the impact of setup geometry, different species, different x-ray systems, different software packages, and other miscellaneous factors on the ratio. The final equation might prove to be simply the histogram ratio multiplied by 100 percent.