Method for optimizing observed image classification criterion and image classification apparatus
View Patent ↗A first object is to use both ADC (automatic defect classification) and MDC (manual defect classification) and reduce the amount of MDC operation. A second object is to prevent a DOI (defect of interest) from being missed. The first object is achieved by displaying judgment information on a screen. The judgment information is necessary when part of the classification is performed by ADC and part of the classification is performed by MDC and used to judge which classification is used, ADC or MDC. In the display operation, ADC classification results and MDC classification results are also displayed in the form of matrix. Further, a missed DOI rate is calculated for each classification threshold used in the defect classification and displayed on the screen.
1. A defect review apparatus having a function of capturing an image of a sample under observation and extracting an area corresponding to each defect from the image as a defect image, the defect review apparatus comprising:
an electronic optical system that images the sample under observation and outputs the acquired image as an image signal;
a defect detection processor that extracts the defect image;
an automatic defect classification processor that automatically classifies the defects into a plurality of categories; and
screen display means for displaying judgment information used to judge which category among the categories into which the automatic defect classification processor has automatically classified the defects needs to undergo manpower-based manual classification;
wherein:
the automatic defect classification processor contains a plurality of pieces of threshold information based on which the defect classification is performed, and
pieces of the judgement information for the plurality of thresholds are calculated based on the automatic defect classification results obtained for the plurality of thresholds.
2. The defect review apparatus according to claim 1 , further comprising:
a confusion matrix that displays results of the manual classification of the defects and results of the automatic defect classification of the defects in the form of matrix, where the plurality of categories corresponding to rows and columns of the matrix is displayed on the screen display means.
3. The defect review apparatus according to claim 2 , wherein:
the defect review apparatus further comprises storage means for storing the classification results provided from the automatic defect classification processor and the results of the manpower-based manual classification of the defect images having undergone the automatic defect classification, and
the automatic defect classification processor uses the automatic defect classification results and the manual classification results to produce not only a confusion matrix in which the number of defects classified by the automatic defect classification in each of the categories and the number of defects classified by the manual classification in each of the categories are arranged on a category basis, but also the judgment information and
displays the confusion matrix and the judgment information on the screen display means.
4. The defect review apparatus according to claim 2 , wherein:
the judgment information is the purity or the correctness rate of the number of defects in each of the categories.
5. The defect review apparatus according to claim 4 , wherein:
the purity is displayed in an enhanced manner in the confusion matrix.
6. The defect review apparatus according to claim 1 , wherein:
the judgment information is a category that requires manpower-based manual classification, a rate representing how many defects does not require the manpower-based manual classification, and the rate at which a defect of interest is missed.