Image Blending via Gain Compensation and Kernel Weighting
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current image stitching technologies are inefficient for real-time processing, particularly in video applications, as they are either too large, generate excessive heat, or require offline processing to achieve undetectable results, making them unsuitable for integration into cameras.
Innovation Solution
A filtering apparatus and method that combines pixels from multiple images by applying weights to present pixels, mapping the arrangement of present and non-present pixels, and using kernel-based weighting to generate new pixels, along with gain compensation and alpha masking to harmonize image qualities across overlapping regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Manufacturing precision
If multiple images are stitched together offline to achieve high-quality seamless blending, then image quality is improved, but processing time increases and real-time performance is lost
Solution Approach 1:
The patent segments the image stitching process into distinct functional modules: overlay module for positioning images, blend module for combining overlapping regions, and filter module for processing pixel data. This segmentation enables parallel processing and optimization of each module independently, achieving real-time performance while maintaining quality
Solution Approach 2:
The patent performs preliminary actions by pre-calculating blend weights and filter kernels before actual stitching operations. The system pre-processes image data to identify overlapping regions and pre-computes blending parameters, which significantly reduces processing time during real-time video stitching while preserving image quality
2Productivity
If real-time stitching systems are designed to process video images quickly, then processing speed is improved, but system size increases and heat generation becomes excessive
Solution Approach 1:
The patent extracts the core stitching functionality into a dedicated filter apparatus that can be integrated into camera systems. By separating the stitching function from general-purpose processors and implementing it as a specialized module with dedicated hardware for pixel filtering and blending, the system achieves real-time processing without requiring large complex systems
Solution Approach 2:
The patent implements dynamic processing where the filter apparatus adapts its operation based on input characteristics. The system dynamically adjusts processing parameters and activates only the necessary processing pipelines based on scene complexity, maintaining high processing speed while reducing power consumption and heat generation during operation
3Manufacturing precision
If traditional filtering methods are used to combine pixels from multiple images, then image blending is achieved, but computational overhead increases and real-time processing becomes difficult
Solution Approach 1:
The patent changes the parameters of the filtering operation by using simplified weight calculation methods and optimized kernel functions. Instead of complex multi-step filtering algorithms, the system uses parameterized filter models that can be evaluated efficiently, reducing computational overhead while maintaining blending quality suitable for real-time video processing
Data Source
AI summary
An apparatus for combining multiple images to form a blended image, configured to identify regions of overlap: (i) in a first image and in a second image, corresponding to where those first and second images will overlap each other in the blended image; and (ii) in the first image and in a third image, corresponding to where those first and third images will overlap each other in the blended image, identify an image quality associated with each region of overlap, determine a gain for each image that, when applied to the image as a whole, will minimise a sum of: (i) a difference between the image qualities associated with the regions of overlap in the first and second images; and (ii) a difference between the image qualities associated with the regions of overlap in the first and third images and apply the respective gains to the first, second and third images.