System and method for evaluating impacts of defects in printed images
A system and method evaluate defects in printed images. A target image, which has been captured of a printed image, is processed to identify defects, where present, which do not occur in a source image from which the printed image was generated. A trained classification model predicts a defect class for respective regions of the target image, each of the defect classes being drawn from a predefined set of defect classes. For at least one of the identified defects, a measure of severity of the defect is determined, such as a size of the defect. A decision on the acceptability of the printed image is made, based on the measure of severity of the at least one defect and the predicted defect class of a respective one of the regions in which the defect occurs.
1. A method for evaluating defects in printed images, comprising:
receiving a target image, captured of a printed image;
identifying one or more defects in the target image, where present, which do not occur in a source image from which the printed image was generated;
with a trained classification model, segmenting the source image into regions and predicting a defect class for each of the regions of the source image and respective regions of the target image, each of the defect classes being drawn from a predefined set of at least three defect classes, each of the defect classes being associated with a threshold measure of severity;
for at least one of the identified defects, determining a measure of severity of the defect;
making a decision on an acceptability of the printed image, based on the measure of severity of the at least one defect and the threshold measure of severity associated with the predicted defect class of a respective one of the regions of the target image in which the defect occurs.
2. The method of claim 1 , wherein the receiving of the target image comprises receiving a scanned image of the printed image and aligning the scanned image with the source image to identify the target image.
3. The method of claim 1 , wherein the identifying defects in the target image comprises identifying differences between the target image and the source image.
4. The method of claim 1 , wherein the trained model has been trained on labeled images, labels of the training images each identifying at least a first region of a respective training image and a respective defect class.
5. The method of claim 1 , wherein the defect classes in the set of defect classes are each associated with a defect significance, at least one of the classes having a higher defect significance than a second of the defect classes.
6. The method of claim 1 , wherein there are at least ten defect classes.
7. The method of claim 1 , further comprising associating a measure of confidence with the predicted defect class.
8. The method of claim 1 , wherein the defect classes comprise visual object classes.
9. The method of claim 1 , wherein the trained model comprises a neural network.
10. The method of claim 1 , wherein the determining of the measure of severity of the defect comprises comparing a size of the defect to a threshold size, the threshold size being based on the defect class for the region in which the defect is located.
11. The method of claim 1 , wherein the determining of the measure of severity of the defect comprises identifying a type of the defect, selected from a set of defect types.
12. The method of claim 1 , wherein the making of the decision on the acceptability of the printed image comprises determining whether or not the printed image should be discarded.
13. The method of claim 12 , wherein when the decision is to discard the printed image, performing at least one of:
causing the printed image to be directed to a discharge path of a printer which printed the printed image;
alerting a user that the printed image should be discarded; and
modifying the printer to reduce a probability that a subsequent printed image will include defects.
14. The method of claim 1 , further comprising:
sending the source image to a printer for printing the printed image on print media with a marking material; and
scanning the printed image with a scanner to generate a scanned image comprising the target image.
15. The method of claim 1 , further comprising updating the classification model based on the source image.
16. A system for evaluating defects in printed images,
comprising memory which stores instructions for performing the method of claim 1 and a processor in communication with the memory which executes the instructions.
17. A system for evaluating defects in printed images, comprising:
a printer which renders a source image on print media to generate a printed image;
an image capture device which captures a target image of the printed image;
a classification model trained to segment the source image into regions and predict defect classes for the regions of the source image and corresponding regions of the target image from a set of predefined defect classes, each of the defect classes being associated with a threshold defect size; and
a defect evaluation component which is configured for:
identifying defects in regions of the target image which each exceed the threshold size for the predicted class output by the model for a respective one of the regions in which the defect is located; and
making a determination on an acceptability of the printed image based on at least one of the identified defects.
18. The system of claim 17 , wherein the classification model comprises a neural network.
19. The system of claim 17 , wherein the defect evaluation component includes a model learning component which updates the classification model using the source image.
20. A method for evaluating defects in printed images, comprising:
printing a digital source image on a sheet of print media using marking materials to generate a printed image;
capturing a digital target image of the printed image;
with a trained classification model, predicting a respective class for each of at least one region of the source image, each of the classes being drawn from a set of classes, each of the classes in the set being associated with a respective threshold measure of severity;
identifying at least one defect in the target image, where present, which does not occur in the source image and which meets the threshold measure of severity for a respective region in which the defect is located; and
determining an acceptability of the printed image, based on the identified at least one defect, where present.