17984755. SYSTEM AND METHOD FOR TRAINING OF NOISE MODEL USING NOISY SIGNAL PAIRS simplified abstract (SAMSUNG ELECTRONICS CO., LTD.)
Contents
- 1 SYSTEM AND METHOD FOR TRAINING OF NOISE MODEL USING NOISY SIGNAL PAIRS
- 1.1 Organization Name
- 1.2 Inventor(s)
- 1.3 SYSTEM AND METHOD FOR TRAINING OF NOISE MODEL USING NOISY SIGNAL PAIRS - A simplified explanation of the abstract
- 1.4 Simplified Explanation
- 1.5 Abstract
- 1.6 Bullet Points
- 1.7 Potential Applications
- 1.8 Problems Solved
- 1.9 Benefits
- 1.10 Original Abstract Submitted
SYSTEM AND METHOD FOR TRAINING OF NOISE MODEL USING NOISY SIGNAL PAIRS
Organization Name
Inventor(s)
Marcus Anthony Brubaker of Toronto (CA)
Michael Scott Brown of Toronto (CA)
SYSTEM AND METHOD FOR TRAINING OF NOISE MODEL USING NOISY SIGNAL PAIRS - A simplified explanation of the abstract
This abstract first appeared for US patent application 17984755 titled 'SYSTEM AND METHOD FOR TRAINING OF NOISE MODEL USING NOISY SIGNAL PAIRS
Simplified Explanation
Abstract
A noise model is trained to simulate noise introduced by a capture device. This is done by using a denoiser and a training data set of pairs of noisy signals. The noise model and denoiser are trained iteratively using a loss function that considers both the first and second denoised signals. This approach avoids the need for obtaining clean signals and prevents convergence on undesired training results.
Bullet Points
- The patent application describes a method for training a noise model to simulate noise introduced by a capture device.
- The method uses a denoiser and a training data set of pairs of noisy signals.
- Each iteration of training involves obtaining denoised signals and optimizing a loss function.
- The loss function considers both the first and second denoised signals to train both the noise model and the denoiser.
- The use of noisy samples avoids the complexities of obtaining clean signals.
- The use of "cross-sample" loss functions prevents convergence on undesired training results without complex regularization.
Potential Applications
- This technology can be applied in the field of image and video processing to simulate noise introduced by capture devices.
- It can be used in the development and testing of denoising algorithms and techniques.
- The noise model can be utilized in computer graphics and animation to add realistic noise effects to virtual scenes.
Problems Solved
- The technology solves the problem of obtaining clean signals for training a noise model.
- It addresses the issue of convergence on undesired training results without the need for complex regularization techniques.
- The method provides a more efficient and practical way to simulate noise introduced by capture devices.
Benefits
- The use of noisy samples simplifies the training process by eliminating the need for clean signals.
- The "cross-sample" loss functions ensure better convergence during training without the need for complex regularization.
- The technology enables the development and testing of denoising algorithms and techniques in a more realistic and practical manner.
Original Abstract Submitted
A noise model is iteratively trained to simulate introduction of noise by a capture device, by use of a denoiser and a training data set of pairs of noisy signals. First and second noisy signals of each pair are independently sampled by the capture device from source information corresponding to the pair. Each iteration of training obtains first and second denoised signals from respective noisy signals, then optimizes at least one loss function which sums first and second terms to train both the noise model and the denoiser, where the first term is based on the first denoised signal and the second noisy signal, and the second term is based on the second denoised signal and the first noisy signal. By using noisy samples, the complexities of obtaining “clean” signals are avoided. By using “cross-sample” loss functions, convergence on undesired training results is avoided without complex regularization.
