Bottle Attribute Extraction Using Segmentation and Multi-Model Vision

Resolve Bottlenecks,
Find Innovative Solutions
Generate Solutions

Solution Overview

Problem

Existing image recognition systems fail to provide sufficient detailed features for beverage bottles to support occasion brand-price-packaging-channel analysis, and existing CNN architectures struggle to detect and segment multiple objects in complex situations like multiple objects in an image.

Innovation Solution

A computer-implemented method using a CNN architecture that includes a shape model trained on shape labeled objects in images to identify shapes, combined with a Mask RCNN model for instance segmentation, followed by a series of bootstrapped models to generate fine-grained features such as shape, color, and design elements of objects like bottles.

Engineering Contradictions & Design Principles

VSEngineering Contradiction Analysis

1Loss of information

If a CNN architecture is used for image recognition, then high-order semantic information can be extracted, but sufficient detailed features for beverage bottles cannot be provided

Engineering Contradiction:
Improvedetailed featuresVSAvoidmodel architecture
Core Design Contradiction:
Loss of informationVSDevice complexity

Solution Approach 1:

The patent segments the image processing task into multiple specialized models: a segmentation model that divides the image into multiple segments, and multiple attribute extraction models that process different segments to extract different attributes (shape, color, design elements). This segmentation approach enables detailed feature extraction without requiring a single overly complex model.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The patent transitions from a single-model approach to a multi-model architecture, adding the dimension of model specialization. By creating separate attribute extraction models for different attributes (shape, color, design elements), the system achieves comprehensive detailed feature extraction while maintaining manageable complexity through modular design.

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

2Measurement precision

If a single model is used for object detection, then the system is simple, but multiple objects in complex situations cannot be detected and segmented accurately

Engineering Contradiction:
Improveobject segmentation accuracyVSAvoidmodel architecture
Core Design Contradiction:
Measurement precisionVSDevice complexity

Solution Approach 1:

The patent introduces a segmentation model that divides the image into multiple segments, enabling precise identification and separation of multiple objects. This segmentation step is followed by attribute extraction models that process individual segments, achieving high measurement precision for multiple objects in complex situations.

Inventive Principle:
Principle #1Segmentation

Solution Approach 2:

The attribute extraction models are designed to be universal and can process different types of objects and attributes through a common framework. The models extract multiple attributes (shape, color, design elements) from segmented regions, providing multi-functionality that handles various object detection and analysis tasks.

Inventive Principle:
Principle #6Universality (Multi-functionality)

3Loss of information

If detailed attribute extraction is performed on segmented objects, then fine-grained features are obtained, but processing time and computational resources increase

Engineering Contradiction:
Improvefine-grained attributesVSAvoidprocessing time
Core Design Contradiction:
Loss of informationVSLoss of time

Solution Approach 1:

The patent performs preliminary segmentation of the image into multiple segments before attribute extraction. This preliminary action organizes the input data in a way that facilitates efficient processing by subsequent attribute extraction models, reducing the computational burden of detailed attribute analysis.

Inventive Principle:
Principle #10Preliminary action

Solution Approach 2:

By dividing the image into segments and processing each segment with specialized attribute extraction models, the system achieves efficient parallel processing. This segmentation approach reduces the overall processing time compared to analyzing the entire image with a single complex model, while still extracting comprehensive fine-grained attributes.

Inventive Principle:
Principle #1Segmentation

Data Source

PatentEP4405912B1Visual attribute expansion via multiple machine learning models
Publication Date: 2026.03.11 MICROSOFT TECHNOLOGY LICENSING LLC
  • EP4405912B1 patent drawingFigure 1
  • EP4405912B1 patent drawingFigure 2
  • EP4405912B1 patent drawingFigure 3

AI summary

A computer implemented method includes receiving an image that includes a type of object, segmenting the object into multiple segments via a trained segmentation machine learning model, and inputting the segments into multiple different attribute extraction models to extract different types of attributes from each of the multiple segments.