Defect classification method, method of sorting photomask blanks, and method of manufacturing mask blank

A method of classifying a defect recognized on a surface of a to-be-inspected substrate, the method comprising the steps of:

• scanning the surface of the to-be-inspected substrate with inspection light by using a defect inspection device having a confocal optical system (S1);
• detecting reflected light from the to-be-inspected substrate in a two-dimensional manner (S2);
• performing signal processing in a two-dimensional manner on an output signal corresponding to the reflected light (S3) and thereby detecting the defect on the surface of the to-be-inspected substrate (S4);
• performing image processing on an image obtained by the signal processing in the two-dimensional manner and classifying the defect that has been detected (S5), the image processing including:
• a first substep of processing a first inspection image and a second inspection image generated when a defect on a substrate whose surface defect is classified is inspected by the defect inspection device to configure a classification criterion for classifying the defect through machine learning using a predefined parameter as a feature amount;
• and
• a second substep of computing a feature amount from a first inspection image and a second inspection image obtained from the to-be-inspected substrate using the defect inspection device to classify the defect of the to-be-inspected substrate in accordance with the classification criterion, wherein
• computation of the feature amount of a defective section in the first substep and the second substep is performed by obtaining an approximate region where the defect exists in the second inspection image, applying the approximate region to the first inspection image, and limiting a region of image processing to compute a feature amount of a defective section.