Arabic Captioning for Images of Clothing Using Deep Learning

Area of Science:

• Computer Vision
• Natural Language Processing
• Artificial Intelligence

Background:

• Automated item descriptions are crucial for e-commerce, especially for large clothing inventories.
• Existing image captioning research heavily favors English, leaving a gap for Arabic language applications.
• Deep learning models, combining visual and textual analysis, are standard for image captioning.

Purpose of the Study:

• To develop and evaluate a deep learning model for Arabic fashion image captioning.
• To create a new Arabic dataset for clothing image captioning named ArabicFashionData.
• To enhance caption quality by incorporating classified image attributes.

Main Methods:

• Utilized deep learning with an image model for visual analysis and a language model for caption generation.
• Developed a novel Arabic dataset (ArabicFashionData) for clothing images.
• Integrated an attention mechanism and classified clothing attributes as input to the model's decoder.

Main Results:

• Achieved a BLEU-1 score of 88.52 for Arabic fashion image captioning.
• Demonstrated the effectiveness of an attribute-based approach in enhancing caption quality.
• The developed model is the first of its kind for Arabic fashion image captioning.

Conclusions:

• The attribute-based image captioning model shows significant promise for Arabic language applications.
• Further improvements are expected with larger datasets.
• This work lays the foundation for advanced Arabic image understanding in e-commerce.