Parallel Phase Retrieval for Diffraction Imaging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
The existing methods for generating high-resolution image data are computationally intensive, requiring significant processing time due to the need for repeated iterations in the phase retrieval algorithm, which can take up to 14 minutes for a single high-resolution image using a PC with an Intel Core2 2GHz CPU.
Innovation Solution
The method employs parallel processing using two or more computational units to compute intermediate object functions, which are then combined to form an estimate of the object function, minimizing disparities between object function estimates for individual regions and reducing computation time by partitioning aperture positions into regions and assigning them to computational units for simultaneous processing.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If repeated iteration of the phase retrieval algorithm is performed to achieve high-resolution image data, then image quality and resolution are improved, but processing time increases significantly (up to 14 minutes per iteration)
Solution Approach 1:
The patent divides the computational domain into multiple regions and assigns different computational units to process each region simultaneously. The object function is segmented into multiple partial object functions, each computed in parallel by different computational units, then combined to form the complete object function. This segmentation enables parallel processing that reduces total computation time while maintaining image resolution quality.
Solution Approach 2:
The patent introduces a temporal dimension to parallel processing by implementing multiple iterations of the phase retrieval algorithm simultaneously across different computational units. Each computational unit performs iterations on different regions, and the results are combined iteratively. This dimensional approach to parallelization reduces the sequential processing time from 14 minutes to approximately 5 seconds per iteration.
2Productivity
If parallel processing is implemented to reduce computation time, then processing speed is improved, but system complexity increases due to the need for multiple computational units and coordination mechanisms
Solution Approach 1:
The computational domain is segmented into multiple regions, each handled by a separate computational unit. This segmentation allows independent processing while maintaining overall system coordination through the combination of partial results. The segmentation principle enables parallel processing without requiring complex inter-dependencies between all computational units.
Solution Approach 2:
Multiple partial object functions computed by different computational units are merged through a combination process to form the complete object function. This merging mechanism simplifies the overall system architecture by providing a straightforward aggregation approach rather than requiring complex synchronization and coordination between all parallel units.
3Loss of time
If aperture positions are partitioned into regions and assigned to computational units, then computation time is reduced through simultaneous processing, but the complexity of coordinating and combining results increases
Solution Approach 1:
The aperture positions are segmented into distinct regions, with each region assigned to a specific computational unit. This segmentation enables simultaneous processing of different aperture positions without requiring complex coordination between all units, as each unit independently processes its assigned region.
Solution Approach 2:
The patent implements feedback mechanisms where computational units use previous estimates of the object function to guide their computations. This feedback approach simplifies coordination by providing a common reference framework that aligns the work of different computational units, reducing the complexity of result combination.
Data Source
AI summary
Embodiments of the present invention provide a computer-based method for providing image data of a region of a target object (31), comprising the steps of estimating (706) a first intermediate object function indicating at least one characteristic of a first sub-region (501, 503) of said region of the target object, estimating (706) a second intermediate object function indicating at least one characteristic of a second sub-region (502, 504) of said region of the target object, and combining (707) the first and second intermediate object functions to form a combined object function indicating at least one characteristic of the region of the target object (31).