System and method for detecting defects on imaged items
Embodiments of the invention provide a machine learning based detection system, in which defects can be detected even if the system was not trained for these defects and even in items that were not used to train the system, thereby offering an inherently flexible detection system.
1. A method for obtaining a prediction for an imaged item, the method comprising:
receiving a variable number of images of a same-type item of a known class and an image of the same-type item of an unknown class;
calculating a predetermined size representation of the received images by comparing the received images to extract attributes from the received images, and using the attributes to produce the predetermined size representation;
inputting the predetermined size representation to a machine learning predictor to obtain a prediction regarding the item of the unknown class; and
controlling a device based on the prediction.
2. The method of claim 1 wherein the prediction comprises detecting a defect on the item of the unknown class.
3. The method of claim 1 wherein the item of known class comprises a defect-free item.
4. The method of claim 3 comprising:
determining that a number of images that fully represent the defect-free item, were received prior to calculating the predetermined size representation.
5. The method of claim 4 wherein determining that a number of images that fully represent the defect-free item, were received, comprises analyzing information regarding 2D and 3D shape of the defect-free item.
6. The method of claim 4 comprising:
determining that a number of images of a defect-free item that fully represent the defect free item, were received;
adding the image of the item of unknown class; and
calculating the predetermined size representation by using the images of the defect-free item and the image of the unknown class.
7. The method of claim 1 comprising extracting attributes from the images based on similarity of the attributes between images.
8. The method of claim 1 wherein the calculating comprises using machine learning techniques.
9. The method of claim 1 wherein the variable number of images of a same-type item of a known class, comprises several images.
10. The method of claim 1 wherein the machine learning predictor is trained on images of items of a first type, and wherein the item of known class and the item of unknown class are of a second type which is different than the first type.
11. A system for visual inspection of a production line, the system comprising:
a processor in communication with a camera, the processor to:
receive a variable number of images of a same-type item of a known class and an image of an item of an unknown class;
calculate a predetermined size representation of the received images by comparing the received images to extract attributes from the received images, and using the attributes to produce the predetermined size representation; and
use the predetermined size representation in a machine learning process to obtain a prediction regarding the item of unknown class.
12. The system of claim 11 comprising controlling, based on the prediction, a signal processing system to control the production line.
13. The system of claim 11 wherein the prediction comprises detecting a defect on the item of unknown class.
14. The system of claim 11 wherein the same-type item of a known class comprises a defect-free item.
15. The system of claim 14 wherein the processor is to determine that a number of images that fully represent the defect-free item, were received prior to calculating the predetermined size representation.
16. The system of claim 11 wherein the machine learning process was trained on images of items of a first type, and wherein the item of known class and the item of unknown class are of a second type which is different than the first type.