Nowadays, people pay more attention to the quality of diamonds, rather than only focusing on their beauty and appearance. Clarity, color, cut and carat weight are the four most important criteria globally used to assess diamond quality. These four criteria to assess a diamond are currently manually evaluated by professional appraisers. However, it takes a long time to train an appraiser, and the job is very physically demanding and the grading results from different appraisers are sometimes inconsistent. Therefore, there is a high demand in the gemological industry to replace manual diamond evaluation methods with an automated system.

Many gemological laboratories around the world have been working on automated diamond detection. Several approaches based on machine vision and me...[ Read more ]

Nowadays, people pay more attention to the quality of diamonds, rather than only focusing on their beauty and appearance. Clarity, color, cut and carat weight are the four most important criteria globally used to assess diamond quality. These four criteria to assess a diamond are currently manually evaluated by professional appraisers. However, it takes a long time to train an appraiser, and the job is very physically demanding and the grading results from different appraisers are sometimes inconsistent. Therefore, there is a high demand in the gemological industry to replace manual diamond evaluation methods with an automated system.

Many gemological laboratories around the world have been working on automated diamond detection. Several approaches based on machine vision and methods for improving image processing have been proposed. However, their experimental results only match up to 75% of manual detection, which does not meet the expectations of automatic detection in gemological industry. One of the main scientific challenges is improving the signal to noise ratio of the captured diamond images, so that different types of signals of a diamond can be clearly acquired and distributed into different areas to avoid overlapping. Another challenge is how to find the image features of each type of signals, such as inclusions and reflections, in order to accurately extract the regions of interest and distinguish each type of signals of a diamond image.

Nevertheless, research work conducted so far has inspired us to think about alternative ways to deal with the above-mentioned problems. In this thesis, the problems are attempted to be solved from a physical perspective. The physical laws that govern diamond light propagation are discovered based on the analysis of diamond material features and optical properties of natural diamonds. Based on the physical laws, the physical principle in diamond image formation is obtained.

The obtained physical principle is applied to design a new diamond image acquisition system, which can control the light distribution on the image plane by optimizing the parameters and relative position of each optical component. Thus, different types of signals of the inspected diamond can be successfully distributed into different image regions without overlap, so that the signal to noise ratio of the captured diamond image improves significantly. It simultaneously captures two images of an inspected diamond—from the bottom and the side of the diamond. The bottom image shows the inclusions (clarity) and light leakage (cut) of a diamond. The side images can be used to evaluate diamond color and measure the proportional parameters of diamond cut.

After acquiring the diamond images, an inclusion extraction approach is developed to accurately separate the regions of interest and then to extract the inclusion regions, which is key to the success of the clarity automatic detection. Two main findings about the diamond optical properties are built into the image processing to identify the image features of individual regions and distinguish between diamond inclusions and reflections. One is the wide discrepancy between the absorption rate of the pure diamond and that of inclusions. The other is the irregular texture and shape of the diamond inclusions which cause light to propagate through the inclusions differently than it does through the pure diamond. The effectiveness and robustness of diamond inclusion extraction system are experimentally verified. The experimental results match 92% of the gemologists’ manual detection results.

In addition, an automatic approach of diamond color measurement is developed to accurately detect the minor color difference between diamonds in different grade. The analysis of human color perception makes sure the diamond color image processed exactly the way that people observe color. In order to detect the tiny traces of color and the presence of a faint yellow or brown tint, the hue channel referring to a pure spectral color and the RGB channels are analyzed after the data are corrected and adjusted to match human color perception. The experimental results match 96% of the gemologists’ manual detection results. The whole system is also verified to be significantly less sensitive to noise than existing approaches and unaffected by the fluctuations in illumination.

Moreover, diamond cut is detected by calculated diamond proportional parameters and extracted leakage light regions. As for diamond proportional calculation, the side pattern of a diamond including pavilion, girdle and crown portions are reconstructed by extracted diamond edges. Key parameters of diamond proportions and facets are calculated and then compared with diamond ideal-cut parameters to find the cutting errors. The other method for diamond cut detection is developed to test the total area and brightness of all light leakage regions, which considers the effect of slight nuances of faceting and clarity on diamond cut. These two methods for diamond cut evaluation are experimentally verified and measurement precision is higher than that of gemological industry.

Key words: Machine vision, image acquisition system, image processing, image formation, diamond quality evaluation, feature extraction, color perception.