Garment classification method based on image recognition

By combining foreground segmentation and dual-path feature extraction with metadata information, the problems of complex background interference and local feature loss were solved, achieving high-precision clothing classification and improving the classification accuracy and user experience of e-commerce platforms.

CN120997810APending Publication Date: 2025-11-21QUANZHOU NORMAL UNIV
0 Cites 1 Cited by

Patent Information

Application Number
CN202510951005.0
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-07-10
Publication Date
2025-11-21

AI Technical Summary

Technical Problem

Existing technologies cannot effectively handle complex background interference in clothing classification, ignore local key point features, and lack effective feature fusion and decision-making mechanisms, resulting in low classification accuracy and poor fine-grained performance.

Method used

Foreground segmentation is used to generate clothing foreground masks. Dual-path feature extraction is combined to generate global texture and local key point feature vectors. Metadata information is used for preliminary classification, and weighted fusion and category decision tree models are used for fine-grained classification. A manual review module is introduced to ensure accuracy.

Benefits of technology

It improves the accuracy and reliability of clothing classification, enhances the effect of fine-grained classification, and ensures the stability of classification results and user experience.

✦ Generated by Eureka AI based on patent content.
Patent Text Reader

Abstract

The invention provides a clothing classification method based on image recognition, which comprises the following steps: controlling an image acquisition device to obtain a to-be-classified clothing image in real time, and if complex background interference is detected, performing foreground segmentation processing on the image to generate a clothing foreground mask; performing dual-path feature extraction on the image to generate a global texture feature vector and a local key point feature vector; performing preliminary classification according to the image metadata information to generate first category probability distribution; combining the clothing foreground mask, the global texture feature vector and the local key point feature vector to generate second category probability distribution, and performing weighted fusion on the second category probability distribution and the first category probability distribution to obtain fusion category probability distribution; and performing fine-grained classification according to the fusion category probability distribution, generating a final garment category label, and controlling a display terminal to display or a database to store a classification result. The garment classification accuracy can be improved, the accuracy of the classification result is ensured, and the overall performance and the user experience of an e-commerce platform garment classification system are improved.
Need to check novelty before this filing date? Find Prior Art