Bystander Removal With Attached Objects Using Aggregated Bounding Boxes

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image editing methods struggle to effectively remove bystanders attached to objects, leading to unsatisfactory results due to misidentification of pixels associated with both the bystander and the object, which can confuse image classification and editing processes.

Innovation Solution

A computer-implemented method involving a machine-learning model to generate an aggregated box encompassing the bystander and attached objects, followed by segmenting and inpainting the bystander mask with background pixels, using techniques like intersection over union ratio and trained neural networks to enhance precision.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Measurement precision

If existing image editing methods are used to remove bystanders, then the bystander removal process is simple, but the removal precision deteriorates when the bystander is attached to an object

Engineering Contradiction:
Improvebystander removal precisionVSAvoidprocessing complexity
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent segments the image processing into distinct modules: a detector that identifies bystanders and attached objects separately, a classifier that distinguishes between them, and an inpainting system that processes them differently. This segmentation allows precise removal of bystanders while preserving attached objects, resolving the contradiction between simple processing and precise removal.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent introduces an intermediary classification system that analyzes the relationship between detected objects and bystanders. This intermediary layer (the classifier module) mediates between the detection stage and the inpainting stage, enabling precise identification of attached objects versus independent bystanders, thereby improving removal precision without overwhelming complexity.

Inventive Principle:
Principle #24Intermediary (Mediator)

2Productivity

If pixels associated with both bystander and object are identified for removal, then the removal process is straightforward, but the image quality deteriorates due to loss of object information

Engineering Contradiction:
Improveremoval efficiencyVSAvoidimage quality
Core Design Contradiction:
ProductivityVSReliability

Solution Approach 1:

The patent applies local quality by treating different regions differently: the detector and classifier modules analyze local pixel characteristics to distinguish between bystander pixels (which should be removed) and object pixels (which should be preserved). The inpainting system then applies different processing to different regions, maintaining object information while removing bystander pixels, thus improving image quality while maintaining removal efficiency.

Inventive Principle:
Principle #3Local quality

Solution Approach 2:

The patent extracts only the necessary information for removal (bystander pixels) while leaving the object information intact. The classification system extracts and separates the bystander components from the attached objects, allowing the inpainting process to remove only the bystander pixels while preserving the object pixels, thereby maintaining image quality and productivity.

Inventive Principle:
Principle #2Taking out (Extraction)

3Manufacturing precision

If a single bounding box is used for bystander detection, then the detection process is simple, but the segmentation accuracy deteriorates when objects are attached to the bystander

Engineering Contradiction:
Improvesegmentation accuracyVSAvoiddetection complexity
Core Design Contradiction:
Manufacturing precisionVSDevice complexity

Solution Approach 1:

The patent segments the detection process into multiple stages: first detecting the bystander with a bounding box, then detecting attached objects separately, and finally integrating these detections through the classification system. This segmentation enables high segmentation accuracy by processing bystander and object regions independently before combining them, avoiding the limitations of a single bounding box approach.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transitions from a single-dimensional bounding box approach to a multi-dimensional detection framework that incorporates spatial relationships, object-bystander connectivity, and pixel-level classification. This dimensional expansion allows the system to accurately segment attached objects from bystanders while maintaining detection simplicity through automated multi-stage processing.

Inventive Principle:
Principle #17Another dimension (Dimensionality change)

Data Source

PatentEP4320585B1Bystander and attached object removal
Publication Date: 2026.03.18 GOOGLE LLC
  • EP4320585B1 patent drawingFigure 1
  • EP4320585B1 patent drawingFigure 2
  • EP4320585B1 patent drawingFigure 3A

AI summary

A media application detects a bystander in an initial image. The media application generates a bystander box that includes the bystander, wherein all pixels for the bystander are within the bystander box. The media application generates localizer boxes that encompass the bystander and one or more objects that are attached to the bystander. The media application aggregates the bystander box and one or more of the localizer boxes to form an aggregated box. The media application applies a segmenter to the initial image, based on the aggregated box, to segment the bystander and the one or more objects from the initial image to generate a bystander mask, wherein the bystander mask includes a subset of pixels within the aggregated box. The media application generates an inpainted image that replaces all pixels within the bystander mask with pixels that match a background in the initial image.