Image Merging with Blend and Conflict Maps
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Low dynamic range (LDR) imaging struggles to capture scenes with wide brightness ranges, often resulting in loss of detail in both bright and dark regions due to limitations in exposure time, leading to undesirable outcomes where either bright or dark areas are over or underexposed.
Innovation Solution
The method involves capturing multiple images of a scene at various exposure times, aligning and merging them using blend and conflict maps to create a high dynamic range (HDR) image that balances exposure, ensuring details in both bright and dark regions are preserved.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Loss of information
If a single LDR image is captured with fixed exposure time, then the device complexity is low, but the brightness range coverage is limited causing loss of detail in bright or dark regions
Solution Approach 1:
The patent divides the image into multiple tiles (e.g., 4x4 grid) and processes each tile independently by selecting optimal images from different exposure times. This segmentation allows detailed processing of specific regions without overwhelming complexity, as each tile can be handled with simpler algorithms compared to processing the entire image at once.
Solution Approach 2:
Multiple images are captured at different exposure times before the merging process. This preliminary action stores exposure data in advance, allowing the system to select and merge images during processing without requiring complex real-time adjustments. The blend map and conflict map are generated in advance to guide the merging process.
2Illumination intensity
If multiple images at different exposure times are merged, then the brightness range coverage is improved, but the processing time increases
Solution Approach 1:
By dividing the image into tiles and processing them in parallel, the system reduces overall processing time. Multiple tiles can be processed simultaneously using different exposure images, and the blend/conflict maps can be generated efficiently for each tile independently, then combined to produce the final result.
Solution Approach 2:
The patent uses blend maps with continuous values (0-255) to control the weighting between different exposure images, allowing smooth transitions and optimized blending. The conflict maps use threshold-based binary values to identify pixel-level conflicts. These parameter representations enable efficient processing by avoiding complex pixel-by-pixel comparisons across all images.
3Stability of the object's composition
If blend maps with continuous values are used, then the blending smoothness is improved, but the computation complexity increases
Solution Approach 1:
The patent transforms the blend map from a simple binary selection (0 or 1) to a continuous weighting map (0-255), allowing smooth transitions between different exposure images. This continuous parameter approach maintains blending smoothness while the computation remains manageable through tile-based processing and efficient data structures.
Solution Approach 2:
Different regions of the blend map can have different blending characteristics. Areas with high conflict values use stronger blending weights, while low conflict areas use weaker weights. This local quality adjustment optimizes the blending process without requiring complex global computations, as each tile can be processed with appropriate local parameters.
Data Source
AI summary
A base m×n tile, X, of a base image of a scene, and an alternate m×n tile, Y, of an alternate image of the scene may be obtained. An m×n blend map, B, for X and Y may also be obtained. B(i,j) may take on a first value to refer to X(i,j), or a second value to refer to Y(i,j). An m×n conflict map, C, for X and Y may further be obtained. C(i,j) may take on a third value where X(i,j) and Y(i,j) are within a threshold value of one another, or a fourth value where X(i,j) and Y(i,j) are not within the threshold value of one another. Based on B and C, the pixel values of X and Y may be merged to form an m×n tile Z.