CVB Flex Inspect - Flexible print and pattern inspection software
Designed especially for the inspection of flat materials CVB Flex Inspect uses a new technique which overcomes the traditional limitations of comparing the sample under test to a "golden template" of a single known good part. Based on latest developments from the field of automatic learning, it can be used for the inspection of flat, printed objects, watermarks or patterned textiles.
In the real world, product variations in such materials are frequently to be expected and many products can simply not be manufactured, reproduced or printed with the necessary fidelity to be tested against a "golden template" of a single known good part for quality assurance purposes. Rather than using a single golden template, CVB Flex Inspect uses the latest in machine learning research to build a complex appearance model of good examples that is then applied to the inspected part to give a difference image just like a traditional golden template approach.
- Silk-screen printing, which can lead to deformations and artefacts
- Printing on flexible media, which introduces local deformations
- Watermarks, which are naturally variable
- Textiles, with material flex and varying thread patterns
- Pad printing
CVB Flex Inspect includes a deformable template alignment tool to allow easy alignment of the test image to the template model. This simplifies use by dealing with gross changes in the position and scale of the objects without the need to manually select alignment markers. For applications where this automatic alignment is insufficient CVB Flex Inspect can accept prealigned images provi ding design flexibility for the developer.
During a training phase the appearance model captures the permitted variations in the training set, building a model of the variable characteristics of the products. A unique feature of the tool is the ability to visualise the acceptable distortion of the product using "imagination" mode. By visualising a range of possible images the user has confidence in what the system has learnt and what it will accept. Adding additional individual samples to the template model is easy and fast, allowing samples that are rejected incorrectly to be added to the model on-thefly without taking the system off line.
At runtime, new images are compared to the appearance model and a synthetic golden template is generated within the bounds of variation. Deviations are returned as a difference image for sub sequent morphological analysis using traditional techniques. The difference images can be passed to other algorithms for analysis, or the inbuilt blob analysis tool can be used to classify defects. Deviations that are not classed as defects can be added into the training set at runtime without interruption. A dynamic mode is also available where the algorithm learns continuously and adapts to gradual changes over time.
- Easy user experience
Visualising a range of possible images the user has confidence in what the system has learnt.
- Simple training
Train the appearance variations expected in your production process with good samples. No negative samples or error images are required.
Enables verification of products that cannot be easily matched to a 'golden template'
Adapts to gradual product changes over time
Visualisation and generation of synthetic images of product variations