Deep Image Compression via Primitive Sample Merging
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Deep images, which include multiple samples per pixel, require large storage space, making them impractical for use in computer graphics due to their size, despite offering advantages in high-fidelity image compositing and post-rendering processes.
Innovation Solution
A method of compressing deep images by combining samples associated with the same primitives using identifiers, allowing for significant storage savings by reducing the number of samples per pixel, with each primitive represented by a single sample, and utilizing a running alpha value for merging samples.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If deep images with multiple samples per pixel are used for high-fidelity image compositing, then image quality and compositing capability are improved, but storage space requirements increase significantly
Solution Approach 1:
The patent combines multiple samples associated with the same primitive into a single merged sample. This is achieved by identifying samples with matching primitive identifiers and merging their alpha values using the formula: merged_alpha = 1 - (1 - alpha1) * (1 - alpha2) * ... * (1 - alphaN). The color values are similarly combined. This merging process reduces the number of samples stored while preserving the composite transparency information needed for high-fidelity compositing operations.
Solution Approach 2:
The patent transforms the representation of multiple samples by changing how alpha values are stored and processed. Instead of storing separate alpha values for each sample, the system merges them into a single alpha value that represents the combined transparency effect. This parameter transformation enables significant storage reduction while maintaining the ability to perform accurate compositing operations.
2Reliability
If multiple samples per pixel are stored for each primitive, then compositing accuracy is improved, but the number of samples and file size increase
Solution Approach 1:
The patent merges multiple samples into a single representative sample for each primitive by combining their alpha values using the running alpha formula. This reduces the number of samples from potentially hundreds or thousands per pixel to a manageable number, while the merged alpha value preserves the cumulative transparency information needed for accurate compositing.
Solution Approach 2:
The patent extracts only the essential information needed for compositing by merging samples and storing only the combined alpha value and associated color information. This extraction process removes redundant data while retaining the critical transparency and color information required for accurate deep compositing operations.
Data Source
AI summary
A method of compressing a deep image representation may include receiving a deep image, where the deep image may include multiple pixels, and where each pixel in the deep image may include multiple samples. The method may also include compressing the deep image by combining samples in each pixel that are associated with the same primitives. This process may be repeated on a pixel-by-pixel basis. Some embodiments may use primitive IDs to match pixels to primitives through the rendering and compositing process.