Researchers at Arizona State University have developed a computationally simple, low-memory, and inherently parallelizable system for image recovery. The proposed invention learns multiple weak sparse models and appropriately combines them to recover the image from its degraded version. The weak models used replace the inherently complex optimization with operations that involve only simple comparisons, and computation of correlations. Obtaining the “dictionary” is also simplified and can be as simple as randomly choosing a subset of training samples. By appropriately weighting the contribution of each of the weak models, the combination can provide better image recovery compared to a single sophisticated sparse model. This approach may be applied in real-time applications, and in mobile platforms that have low computational resources.
- Digital Image Editing Software
- Mobile Phone Camera Software
- Mobile Phone Image Editing Applications
- Medical Imaging
Benefits and Advantages
- Highly Scalable – Proposed system is highly scalable as it is parallel-friendly.
- Real Time Applications – By storing the look-up tables in the GPU memory and performing all computations in the GPU, the full potential of GPUs can be harnessed, thereby enabling real-time image recovery and enhancement.
- High fidelity recovery – Recovered images are of high-fidelity, since a weighted combination of multiple models is used.
For more information about the inventor(s) and their research, please see
Dr. Andreas Spanias' directory webpage