Fractional Pixel Replication for Non-Integer Image Scaling
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing image scaling methods, such as nearest neighbor and bilinear interpolation, often result in artifacts when scaling images with non-integer factors, particularly affecting transparent regions and leading to halo effects.
Innovation Solution
A fractional pixel replication (FPR) algorithm that uses a weighted average of input pixel values, including transparent areas, and applies a translucency mask to smoothly transition foreground and background colors, effectively handling non-integer scaling factors and preserving transparent regions.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Productivity
If nearest neighbor pixel replication algorithm is used for scaling, then the scaling process is simple and fast, but artifacts and aliasing occur in the output image
Solution Approach 1:
The patent changes the scaling parameter from integer-based pixel replication to fractional pixel positioning. By representing pixel positions as fractions (numerator/denominator) and using weighted averaging based on these fractional positions, the method achieves higher quality scaling while maintaining computational efficiency through the specific algorithm structure.
Solution Approach 2:
The patent replaces the mechanical pixel replication process with a mathematical weighted averaging system. Instead of physically copying pixels, the system calculates output pixel values by weighting contributions from multiple input pixels based on their fractional distances, eliminating aliasing artifacts while maintaining processing efficiency.
2Manufacturing precision
If bilinear or bicubic interpolation methods are used to reduce artifacts, then image quality improves, but processing complexity and computational requirements increase significantly
Solution Approach 1:
The patent transforms the interpolation problem by changing from continuous coordinate space to fractional pixel space. By expressing all calculations in terms of fractions with common denominators, the method achieves bilinear/bicubic quality results using simpler integer-based weighted averaging operations, reducing computational complexity while maintaining image quality.
Solution Approach 2:
The patent segments the scaling process into distinct phases: fractional position calculation, weighted contribution determination, and cumulative sum accumulation. This segmentation allows each phase to be optimized independently, reducing overall processing complexity while achieving high-quality results through systematic breakdown of the interpolation task.
3Productivity
If standard scaling algorithms are used on images with transparent regions, then scaling is performed efficiently, but halo effects and artifacts appear around transparent areas
Solution Approach 1:
The patent applies local quality by treating transparent and opaque regions differently during the weighted averaging process. By detecting transparency information and adjusting the blending weights accordingly, the method preserves sharp edges in transparent regions while maintaining smooth transitions in opaque areas, eliminating halo effects without sacrificing overall processing efficiency.
Solution Approach 2:
The patent introduces transparency information as an intermediary factor that mediates the blending calculation. This intermediary allows the algorithm to selectively apply different weighting strategies based on regional transparency properties, preventing artifact formation in transparent areas while maintaining efficient processing through unified algorithmic treatment.
Data Source
AI summary
Non integer scaling of images to reduce artifacts is presented herein. One embodiment includes determining a scaling resolution based on a pixel resolution of the image and a pixel resolution of an output device (i.e., factors of the scaling resolution). The image is converted to the scaling resolution to change (e.g., increase) the pixels of the image by the first factor. A grid is generated to scale the converted image. The grid has a number of sections defined according to the second factor times the pixel resolution of an output device. The converted image is sectioned according to the grid. Each section of the grid includes an integer number of pixels of the converted image. The color values of the pixels of the converted image are averaged within each section of the grid to compute a single color value for each section of the grid and scale the image.