Image Histogram Mean SNR Estimation for Multi-Exposure Noise Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Calculating the mean Signal to Noise Ratio (SNR) value for images, especially multi-exposure images, is costly in terms of processing capacity and does not accurately account for varying noise levels and sources across different exposure regions, making it challenging for noise filtering and auto-exposure algorithms.
Innovation Solution
A method that estimates the mean SNR value by using an image histogram to weight and sum SNR values for each bin representing a range of pixel values, dividing by the total number of image pixels, which can be applied to stitched output images from multiple exposures, including multi-exposure HDR images.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If the SNR value is calculated from image data using the standard formula, then the SNR value accurately reflects the signal quality, but the processing capacity required becomes excessively high
Solution Approach 1:
The patent divides the image into multiple bins based on luminance ranges, calculating SNR values for each bin separately rather than for the entire image. This segmentation allows the complex SNR calculation to be distributed across multiple simpler sub-calculations, reducing the overall processing burden while maintaining accuracy through localized analysis of different luminance regions.
Solution Approach 2:
The patent transforms the continuous SNR calculation problem into a discrete bin-based approach by changing the parameter space from individual pixel values to binned luminance ranges. This parameter transformation enables the use of representative values (minimum, maximum, mean) for each bin, significantly reducing computational complexity while preserving the essential SNR characteristics of the image.
2Device complexity
If the average luminance value is used as the expectancy value of the signal, then the calculation is simplified, but the SNR value becomes inaccurate for multi-exposure images with varying noise levels
Solution Approach 1:
The patent applies local quality by calculating SNR values specifically for different luminance regions (bins) rather than using a single global average. Each bin represents a local region of the image with similar luminance characteristics, allowing the SNR calculation to account for local variations in noise levels and signal strength that are present in multi-exposure images.
Solution Approach 2:
The patent performs partial action by calculating SNR values for only certain luminance bins rather than all possible pixel values. By focusing on representative bins that capture the essential luminance ranges of the image, the method achieves sufficient accuracy without the excessive computational cost of analyzing every individual pixel or luminance level.
3Ease of operation
If noise filtering is applied uniformly across the entire image, then the implementation is simple, but the filtering strength does not adapt to the varying noise levels in different exposure regions
Solution Approach 1:
The patent enables local quality by associating different SNR values with different luminance bins, which can then be used to apply adaptive noise filtering strengths to different regions of the image. Regions with lower SNR values (higher noise) can receive stronger filtering, while regions with higher SNR values can receive lighter filtering, optimizing the overall image quality.
Solution Approach 2:
The patent introduces dynamics by making the noise filtering strength variable rather than fixed. The SNR values calculated for different bins provide a dynamic basis for adjusting filtering parameters according to local image characteristics, allowing the filtering process to adapt to varying noise conditions across different exposure regions.
Data Source
AI summary
A method and a camera provide an estimation of a mean signal to noise ratio value for an output image comprising a number of pixels, wherein each pixel of the output image has a pixel value. An image histogram divided into bins, the image histogram having information regarding the distribution of pixel values among the pixels of the output image. Each bin has a set of pixels having pixel values within a predetermined range. For each bin of the image histogram, a signal to noise ratio value of that bin may be attributed and weighted with the number of pixels in the set of pixels of that bin. The weighted signal to noise ratio values of the bins are summed, and the sum may then be divided by the total number of pixels of the output image or the total number of pixels in the bins of the image histogram.