Image Noise Estimation for Stitched High Dynamic Range Photos

The patent divides the image data into multiple sets, where each set is associated with different exposure values. Instead of using a single noise value for the entire stitched image, the system calculates separate noise values for each set based on its specific exposure characteristics. This segmentation allows accurate noise estimation for high dynamic range images while maintaining manageable processing complexity through organized data structures.

The patent applies local quality by determining noise values that are specific to each set of image data rather than using a uniform noise value across the entire image. Each set's noise value is calculated based on its own exposure value and pixel intensity values, ensuring that the noise estimation is locally optimized for each exposure condition while still being part of the overall stitched image processing system.

The patent changes the approach from calculating noise for each pixel individually to calculating a representative noise value for each set of image data based on exposure parameters and aggregated pixel intensity values. This parameter change reduces computational complexity while maintaining accuracy by leveraging the statistical properties of noise across multiple pixels within the same exposure conditions.

The system uses the pixel intensity values from the image data itself to determine the noise values, rather than requiring external noise measurements or complex calibration procedures. The noise estimation is performed self-service style by utilizing the inherent statistical properties of the captured image data, including the distribution of pixel intensities within each set.