Holographic Object Detection via Sparse Convolutional Dictionary Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing methods for object detection in holograms, particularly in high-cell-concentration samples, fail due to non-linear diffraction processes and require costly and limited training data, with traditional methods being unreliable and necessitating a computationally expensive autofocus step.
Innovation Solution
A method that jointly reconstructs and counts objects in holograms using a sparse convolutional model, learning complex-valued templates and approximating the wavefront propagation, allowing for robust detection even with approximate focal depth knowledge, thereby reducing the need for autofocus.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If traditional object detection methods are applied directly to holograms, then the method is simple to implement, but the detection accuracy fails in high-cell-concentration samples due to non-linear diffraction processes
Solution Approach 1:
The patent introduces a learned template dictionary as an intermediary between the hologram and object detection. The template dictionary, learned from training data, encodes the expected patterns of objects in holographic images. During detection, these templates are correlated with the hologram to identify objects, effectively mediating the detection process and enabling accurate object identification even in high-concentration samples where direct methods fail.
2Measurement precision
If a two-step process (reconstruction then detection) is used, then object detection can be performed on reconstructed images, but the computational cost increases due to the need for autofocus and image reconstruction
Solution Approach 1:
The patent merges the image reconstruction step and object detection step into a single integrated process. Instead of first reconstructing the image and then detecting objects, the template-based detection is applied directly to the hologram in the Fourier domain. This combination eliminates the need for separate autofocus and reconstruction operations, significantly reducing computational expense while maintaining detection accuracy.
3Reliability
If autofocus is performed to compensate for focal depth errors, then detection robustness improves, but the computational time and expense increase significantly
Solution Approach 1:
The patent performs preliminary action by learning the template dictionary in advance during a training phase. The templates are pre-adapted to the specific imaging conditions and object types. During actual detection, these pre-learned templates are directly applied without needing autofocus or iterative optimization, making the detection process both robust and computationally efficient. The heavy computational work is done beforehand rather than during real-time detection.
Applied Scientific Principles
This section explains which scientific principles are used to turn an abstract innovation direction into a practical engineering solution.
Function Achieved in This Case
Enables accurate detection of cells in holograms, including high-cell-concentration samples, with improved robustness to focal depth errors and reduced computational expense, as demonstrated by successful counting of white blood cells in holographic images.
Implementation Method 1
as the signals from individual cells propagate from the object plane to the hologram plane, they interact through a non-linear diffraction process
Data Source
AI summary
Systems and methods for detecting objects in a holographic image are provided. The techniques include obtaining a holographic image having one or more objects depicted therein. A set of object templates is obtained. The set of object templates represents objects to be detected in the holographic image. One or more objects are detected in the holographic image using the set of object templates by iteratively computing a phase (θ) of the optical wavefront at the hologram plane, background illumination (μ). and encoding coefficients (A) for the set of object templates, until converged.