Apparel Image Classification via Body Segmentation and Shape Matching
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current methods for searching clothing items through images are inefficient due to the deformability of clothing, which makes it difficult for reverse image search algorithms to provide accurate results, and many retail listings lack comprehensive product attributes, making it hard for consumers to find what they need.
Innovation Solution
A method and server that classify apparel items by comparing an input image to a database of reference shapes, segmenting the image into body and apparel regions, generating a geometric model of the body, and computing matching scores for reference shapes to select the best match.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of operation
If reverse image search algorithms are used to search for clothing items, then image-based querying capability is provided, but accuracy deteriorates due to clothing deformability across different camera angles, body shapes, and lighting conditions
Solution Approach 1:
The patent segments the clothing item from the body in the input image by identifying body keypoints and generating a body mask, then extracting only the clothing region. This segmentation isolates the deformable clothing from the variable body pose, enabling accurate shape representation independent of body shape and camera angle variations.
Solution Approach 2:
The patent creates a simplified geometric representation (copy) of the clothing item's shape by generating a binary mask and computing shape descriptors. This abstract geometric copy captures the essential shape characteristics while eliminating details affected by lighting, camera angle, and body pose variations.
2Measurement precision
If comprehensive product attributes are required for effective word-based searching, then search precision improves, but ease of operation deteriorates due to consumers not being familiar with technical terminology and the burden of providing detailed information
Solution Approach 1:
The system performs automatic image analysis and classification, extracting clothing shape characteristics and generating search queries without requiring user input of technical attributes. The system serves itself by automatically processing the uploaded image to identify and categorize clothing items based on their geometric shapes.
Solution Approach 2:
The patent replaces the manual process of users inputting text-based product attributes with an automated computer vision system that analyzes images and extracts shape features. This substitution eliminates the need for users to understand or input technical terminology while maintaining high search precision.
3Ease of operation
If traditional reverse image search is used, then image upload is simple, but computing resources are wasted in futile attempts to locate specific clothing items due to inaccurate matching
Solution Approach 1:
The patent performs preliminary processing of the input image by segmenting the clothing from the body and generating shape descriptors before comparison. This preliminary action pre-computes the essential shape features, enabling more efficient and accurate matching that reduces futile comparisons and computing resource waste.
Solution Approach 2:
The patent transforms the image data into shape descriptor parameters that capture the geometric characteristics of clothing. By changing the representation from raw pixel data to abstract shape parameters, the system enables more efficient comparison and matching, reducing computational waste while maintaining accuracy.
Data Source
AI summary
Many clothing listings, particularly in the secondhand market, lack comprehensive or standardized information about the product's attributes, making it difficult for consumers to find what they need. A method and server for classifying images depicting apparel items based on reference shapes is provided. Images depicting apparel are classified based on reference shapes and a geometrical model of the body. A method and system for querying images based on the classification is further provided.