Industrialized production of jewelry uses large quantities of small diamonds of 3mm and below. Traditional evaluation of the diamond stone’s quality is by naked human eyes with the help of a magnifying glass based on criteria known as 4Cs, which are carat, cut, color and clarity. However, the process can be biased by human judgment due to inconsistent grading standard and physical fatigue. Therefore, it is worthwhile to automate the evaluation process. Among the 4Cs criteria, the automation of evaluation of carat, cut and color is realized, only automatic diamond clarity grading lacks a systematic study. Hence, a fully automatic diamond clarity grading system is very desirable and would have a huge market demand.

This thesis proposes a systematic framework for automatic diamond clarity grading for small diamonds and development of an effective algorithm for this task. Image processing and pattern recognition techniques are employed to tackle the problem, including image segmentation and classification. In the image segmentation stage, diamond is first extracted from the raw input image by the thresholding technique. It is then divided into separated regions by the diamond frame model and edge detection. In the classification stage, feature descriptors are extracted from each region and then classified as impurity or non-impurity by a learned classifier. The clarity scale of the diamond can be determined based on the identified impurity regions.

This thesis contributes in several aspects. First, it contributes to the development of a systematic framework for automatic diamond clarity grading. Experiments are conducted to demonstrate the effectiveness of the algorithm. Second, it proposes efficient descriptions of impurity characteristics. Last, the image segmentation technique for extracting the diamond frame model can be used for evaluating diamond cutting quality.