GPU Feature Erasure Along Radial Pixel Paths

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image processing technologies struggle to efficiently erase high-contrast features like text and signatures from documents on darker backgrounds without introducing latency due to CPU-GPU memory copying, which is resource-intensive.

Innovation Solution

Utilizing a graphics processing unit (GPU) to set the color value of each pixel along a line outward from the image center to the edge, effectively erasing features by setting them to the lightest color value of sampled pixels, thereby simplifying edge detection and reducing computational burden.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Productivity

If CPU-based image processing is used to erase high-contrast features, then processing can be performed, but latency increases and resource utilization decreases due to CPU-GPU memory copying

Engineering Contradiction:
Improveprocessing speedVSAvoidlatency
Core Design Contradiction:
ProductivityVSLoss of time

Solution Approach 1:

The patent uses the GPU as an intermediary processing unit that performs feature erasure directly on the image data without requiring data transfer between CPU and GPU memory. The GPU shader program acts as the intermediary that implements the erasure algorithm, eliminating the memory copying bottleneck and reducing latency while maintaining high processing throughput.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Productivity

If traditional image processing methods are used to detect and remove features, then features can be erased, but computational burden increases and resource utilization decreases

Engineering Contradiction:
Improveprocessing efficiencyVSAvoidcomputational burden
Core Design Contradiction:
ProductivityVSUse of energy by moving object

Solution Approach 1:

The patent replaces complex mechanical image processing operations with a simplified GPU-based approach. Instead of using traditional CPU-based algorithms for feature detection and removal, the system uses a GPU shader that processes pixels in parallel, substituting the mechanical complexity of sequential processing with the parallel architecture of the GPU, thereby reducing computational burden while improving efficiency.

Inventive Principle:
Principle #28Mechanics substitution (Replace mechanical system)

3Difficulty of detecting and measuring

If high-contrast features are present in the image, then feature detection becomes easier, but feature erasure becomes more difficult without introducing artifacts

Engineering Contradiction:
Improvefeature detection easeVSAvoiderasure complexity
Core Design Contradiction:
Difficulty of detecting and measuringVSDevice complexity

Solution Approach 1:

The patent applies local quality by processing each pixel independently based on its local neighborhood characteristics. The shader program examines the contrast between each pixel and its neighbors to determine whether it belongs to a high-contrast feature, allowing the erasure operation to adapt locally to the image content. This approach simplifies the erasure process by treating each location uniquely based on its local properties rather than requiring global analysis.

Inventive Principle:
Principle #3Local quality

Data Source

PatentEP3716203B1Feature eraser
Publication Date: 2025.08.27 GOOGLE LLC
  • EP3716203B1 patent drawingFigure 1
  • EP3716203B1 patent drawingFigure 2
  • EP3716203B1 patent drawingFigure 3

AI summary

A GPU receives an image comprising an array of pixels. The image depicts features in a field of an object on a background. The features and the background contrasting with the object field, and at least a portion of the object is at the center of the image. In parallel for each particular pixel of a first plurality of the pixels, the GPU sets the color value of the particular pixel to the lightest color value of a second plurality of the pixels substantially along a line outward from the particular pixel toward an edge of the image. The line can be defined by the particular pixel and the image center.