Invention:
This technology is a novel method for phase unwrapping in optical imaging in a more efficient and accurate way. It uses unwrapping algorithms based on a transport of an intensity equation with a convolutional neural network to unwrap the phase. This novel method uses a multi-class classification process and introduces an efficient segmentation network to identify classes. This allows for a more efficient and effective method in removing the ambiguity that occurs in a wrapped phase.
Background:
Systems that use modern and mature algorithms to extract phase signals tend to return phases that involve phase jumps that produce inherently wrapped outputs that result in unusable phases. These unusable phases require the phase discontinuities to be removed by a phase unwrapping algorithm.
While there are many types of technology that involve and require the extraction of a phase signal from an input image, such phase unwrapping is a time-consuming process that makes it challenging for there not to be any errors when it is produced. Depending on the outcome of the process, the entire system can be affected. The outcome can involve thousands of individual phase wraps, false phase wraps caused by noise, or by the phase extraction algorithm itself. Differentiating between these is a complicated process and the errors can be spread throughout the entire image leaving the resulting image unusable. While there have been quite a few methods that have been developed that are used to try and meet this need, there isn’t a fully efficient process available.
This technology is a more efficient process that removes 2Pi ambiguity while unwrapping at a quicker rate. It's also more efficient in reducing unwrapping errors that occurs in the current technology, which allows for the unwrapping process to become more accurate and less time-consuming.
Applications:
- 3D imaging
- Medical diagnostics
- Optical metrology
- Military imaging
Advantages:
- Occurs at a quicker speed than current techniques
- Is not noise sensitive
- Efficient segmentation network
- Preprocess noisy wrapped phases
