Thermal Sensor Deconvolution via Cluster-Adaptive Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing methods fail to effectively remove blurring caused by lens positioning inaccuracies in optical detector devices, leading to suboptimal deconvolution results in images captured by thermal sensor arrays.
Innovation Solution
A method that generates an intensity distribution model to calculate weights for neighboring pixels based on intra-cluster and inter-cluster distances, applying these weights to deconvolve images and generate kernels for each pixel index in a cluster pattern, thereby constructing a processed image that mitigates blurring.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If a standard deconvolution kernel is applied to mitigate blurring, then some blurring is reduced, but residual blurring remains in the image
Solution Approach 1:
The patent applies different kernel weights to different spatial locations based on the cluster pattern. Pixels within clusters receive different weighting than pixels between clusters, creating a location-adaptive deconvolution approach that better matches the actual optical blur distribution caused by lens positioning errors.
Solution Approach 2:
The patent segments the pixel array into repeating cluster patterns, where each cluster contains a specific arrangement of active and inactive pixels. This segmentation allows the deconvolution process to account for the non-uniform sampling pattern and apply appropriate weighting to each pixel based on its cluster position.
2Measurement precision
If lens positioning is adjusted during manufacture to improve focus, then image sharpness improves, but manufacturing complexity and cost increase
Solution Approach 1:
The patent extracts and compensates for the blurring effect mathematically through deconvolution processing, rather than attempting to physically adjust lens positioning during manufacture. The cluster pattern design intentionally creates a known, repeatable blur pattern that can be reversed through signal processing.
Solution Approach 2:
The patent changes the approach from physical parameter adjustment (lens positioning) to digital parameter processing (kernel weights and deconvolution algorithms). By modifying the digital processing parameters rather than physical manufacturing parameters, the solution achieves sharpness without increased manufacturing complexity.
3Ease of manufacture
If the array of sensing pixels uses a regular grid pattern, then manufacturing is simplified, but blurring from lens positioning errors cannot be effectively corrected
Solution Approach 1:
The patent introduces an asymmetric cluster pattern within the regular grid, where pixels are arranged in non-uniform groups (e.g., 2x2 clusters with specific active/inactive patterns). This asymmetric arrangement within clusters provides the diversity needed for accurate deconvolution while maintaining the overall regular grid structure for ease of manufacture.
Solution Approach 2:
The patent nests the cluster pattern within the larger regular grid array. Each cluster is a self-contained pattern of pixels that repeats throughout the array, creating a hierarchical structure where the simple regular grid contains more complex cluster patterns, combining manufacturing simplicity with deconvolution effectiveness.
Data Source
AI summary
A method of digitally processing an image comprises: generating an intensity distribution model (170) in respect of at least a portion of the array of sensing pixels (102) of a detector device (100). The array of sensing pixels comprises clusters of pixels. A pixel (140) from the array of sensing pixels is then selected (202) and a first distance and a second distance from the selected pixel to a first neighbouring pixel (142) and a second neighbouring pixel (144), respectively, are determined (402) and the intensity distribution model (170) referenced (406) by the first distance is used to calculate a first weight and a second weight to apply to the first and second neighbouring pixels, respectively. The first distance comprises an intra-cluster distance and the second distance comprises an inter-cluster distance, the intra-cluster distance being different from the inter-cluster distance. The first weight is applied (214) to the first neighbouring pixel (142) and the second weight is applied (214) to the second neighbouring pixel (144).