Combined query embroidery image retrieval based on enhanced CNN and blend transformer

Area of Science:

• Computer Science
• Art History
• Digital Humanities

Background:

• Embroidery images are rich in historical data and represent a significant art form.
• Efficient retrieval of specific embroidery images is a challenge due to the complexity of image features.
• Convolutional Neural Networks (CNNs) excel at feature extraction but often overlook global context in textured images.

Purpose of the Study:

• To develop an efficient combination query retrieval method for embroidery images.
• To address the limitations of existing methods in capturing both local and global contextual information.
• To enhance the feature representation richness for improved embroidery image retrieval.

Main Methods:

• Proposed Blend-Transformer with Group External Attention (GEA) to integrate multi-dimensional features and capture local/global context.
• Introduced Enhanced CNN with Shuffle Attention (SA) to regroup and reaggregate CNN-extracted features for richer representations.
• Validated the approach on TCE-S and ICR2020 datasets.

Main Results:

• The proposed algorithm demonstrated excellent performance in embroidery image retrieval tasks.
• The Blend-Transformer effectively integrated local and global context, overcoming CNN limitations.
• Enhanced CNN improved the richness of feature information, leading to better retrieval outcomes.

Conclusions:

• The developed method significantly advances embroidery image retrieval capabilities.
• This research offers a new perspective for the preservation and study of embroidery art.
• The findings contribute to the field of image retrieval for cultural heritage artifacts.