Adaptive Image Filtering Using Power Spectrum and SNR Selection
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image processing techniques lack filters that can precisely act on arbitrary input images, leading to difficulties in adjusting image quality and requiring excessive time to process large images due to inefficient expansion methods.
Innovation Solution
A method and apparatus that calculate and select filters based on the power spectrum and signal-to-noise ratio of input signals, creating a third filter by summing a first filter adapted to the power spectrum and a second filter adapted to the signal-to-noise ratio, allowing for efficient convolution and processing of input signals.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If wavelet algorithm is used to expand input image into image elements of plural resolutions, then image processing can be performed, but processing time becomes excessively long for large images
Solution Approach 1:
The patent segments the filter design process into two independent parts: (1) designing a filter for a reference image with typical characteristics, and (2) adapting this filter to different input images by adjusting parameters based on their statistical properties (variance, contrast, etc.). This avoids the need to process each input image through the full wavelet expansion procedure, significantly reducing processing time while maintaining adaptability.
Solution Approach 2:
The patent performs preliminary filter design and optimization on a reference image before actual processing. The filter parameters are pre-calculated and stored based on the reference image's characteristics. When processing actual input images, only parameter selection and adjustment are needed, not the full filter design process, thereby reducing processing time.
2Productivity
If fixed parameter filters are used for image processing, then processing speed is maintained, but the filters cannot precisely act on arbitrary input images with varying characteristics
Solution Approach 1:
The patent makes the filter parameters dynamic by establishing relationships between filter parameters and input image statistical characteristics (variance, contrast, mean intensity). The filter automatically adapts to different input images by selecting and adjusting parameters based on these characteristics, maintaining both speed and adaptability.
Solution Approach 2:
The patent changes filter parameters based on input image characteristics. Specifically, it adjusts parameters such as filter size, standard deviation, and threshold values according to the variance, contrast, and intensity of the input image, enabling the same filter structure to adapt to different images without sacrificing processing speed.
3Ease of manufacture
If filter parameters are adjusted by assuming input images, then some image processing can be performed, but the parameters cannot be precisely optimized for actual varying input images
Solution Approach 1:
The patent uses a reference image to create a template filter that captures typical image characteristics. This copied filter structure is then adapted to actual input images through parameter adjustment based on statistical comparisons, maintaining simplicity while improving precision.
Solution Approach 2:
The patent introduces feedback mechanisms where filter parameters are adjusted based on the actual statistical properties of input images compared to the reference image. This feedback loop ensures parameters are optimized for each specific input image while maintaining the overall simplicity of the filter implementation framework.
Data Source
AI summary
Image signal processing method and apparatus having a first filter storage portion in which first filters are correlatively stored; a second filter storage portion in which second filters are correlatively stored; a first filter selection portion for selecting a first filter based on the power spectrum of the input image; a second filter selection portion for selecting a second filter based on the SIN (signal-to-noise ratio) of the input image; a third filter creation portion for creating a third filter by summing up the first and second filters; and a convolutional processing portion for convolving the input image using the created third filter.