Content-Based Image Retrieval for Traditional Indonesian Woven Fabric Images Using a Modified Convolutional Neural...

Area of Science:

• Computer Science
• Textile Arts
• Cultural Heritage

Background:

• Content-based image retrieval (CBIR) systems can support Indonesian traditional woven fabric artisans and trade.
• Developing effective CBIR systems is challenging due to limited datasets and the need to consider unique fabric characteristics simultaneously.

Purpose of the Study:

• To create the TenunIkatNet dataset, a specialized collection of Indonesian ikat woven fabric images.
• To develop and evaluate a modified convolutional neural network (MCNN) for efficient and accurate retrieval of these fabrics.

Main Methods:

• Collected 4800 images across 120 classes of Indonesian ikat woven fabrics.
• Captured images under various conditions (perpendicularly, different backgrounds, utilized forms).
• Employed a modified convolutional neural network (MCNN) for feature extraction and image retrieval.

Main Results:

• The TenunIkatNet dataset comprises 120 classes and 4800 images.
• The MCNN model demonstrated superior performance compared to established pretrained CNN models.
• Achieved high retrieval accuracies: 99.96% (top-5), 99.88% (top-10), 99.50% (top-20), and 97.60% (top-50).

Conclusions:

• The developed TenunIkatNet dataset and MCNN provide an effective solution for Indonesian ikat woven fabric retrieval.
• This system can significantly benefit artisans, cultural preservation, and trade promotion efforts.
• The MCNN's performance highlights its potential for specialized image retrieval tasks.