Image Conversion Parameter Calculation Using Random Pixel Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image alignment methods face challenges in accurately calculating conversion parameters, particularly when feature points are difficult to extract or when high computational complexity leads to biased pixel values, especially with increasing image sizes and periodicity in pixel values.
Innovation Solution
An image conversion parameter calculation device that randomly selects a predetermined number of pixels from a template image for processing, using a combination of algorithms such as the compositional and inverse algorithms to derive conversion parameters, thereby reducing computational dependence on image size and preventing biased pixel values.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If region-based method is used to calculate conversion parameter, then image alignment accuracy is improved, but computational complexity increases significantly
Solution Approach 1:
The patent extracts only the necessary pixel information from the template image by randomly selecting a predetermined number of pixels, rather than processing all pixels. This extraction approach reduces computational complexity while maintaining alignment accuracy because the selected pixels are sufficient to calculate meaningful conversion parameters.
Solution Approach 2:
The patent applies partial action by processing only a subset of pixels (predetermined number) from the template image rather than the complete image. This partial processing is sufficient to achieve accurate conversion parameter calculation, avoiding the excessive computational burden of processing all pixels.
2Productivity
If spatially thinning out pixels is used to reduce processing amount, then computational complexity is reduced, but pixel value bias occurs
Solution Approach 1:
The patent introduces asymmetry through random selection of pixels, which breaks the regular patterns and periodicities that cause bias in spatially thinning methods. By randomly selecting pixels rather than systematically thinning them, the method avoids the systematic bias while maintaining reduced processing complexity.
Solution Approach 2:
The patent uses random sampling to create a representative copy of pixel information from the template image. This sampled copy maintains the essential characteristics of the original image data while reducing the amount of data to process, avoiding the bias introduced by systematic thinning methods.
3Measurement precision
If all pixels in notice area are used for conversion parameter estimation, then measurement precision is improved, but processing amount increases with image size
Solution Approach 1:
The patent extracts only a predetermined number of pixels from the notice area for conversion parameter estimation, rather than using all pixels. This extraction maintains measurement precision by selecting representative pixels while significantly reducing the quantity of data processed, making the processing amount independent of image size.
Solution Approach 2:
The patent changes the parameter of pixel selection from using all pixels to using a fixed predetermined number of pixels. This parameter change decouples the processing amount from image size, as the number of processed pixels remains constant regardless of the total image dimensions.
Data Source
AI summary
An image conversion parameter calculation device accurately calculates a conversion parameter for image alignment with a processing amount that does not depend on the size of an image to be aligned. A pixel selection element randomly selects pixels from a predetermined number or pixels of not more than the predetermined number from a first image. A parameter derivation element derives a conversion parameter by performing processing for pixels selected by the pixel selection element, the conversion parameter being a parameter for converting, to the first image, a second image that is subject to image alignment with the first image.