Document image retrieval through word shape coding.

Area of Science:

• Computer Science
• Information Retrieval
• Image Processing

Background:

• Traditional document image retrieval often relies on Optical Character Recognition (OCR), which can be error-prone and computationally intensive.
• Existing methods may struggle with document degradation, variations in fonts, or complex layouts.
• There is a need for robust and efficient techniques to search document images directly based on their visual content.

Purpose of the Study:

• To develop a novel document image retrieval technique that does not require OCR.
• To enable searching of document images using a content-based approach.
• To create a system that is efficient, fast, and resilient to document degradation.

Main Methods:

• A new word shape coding scheme is proposed to represent document content.
• Word images are annotated using topological shape features like ascenders, descenders, holes, and water reservoirs.
• Document images are retrieved using query keywords or a query document image based on the generated word shape codes.

Main Results:

• The proposed technique successfully retrieves document images without relying on OCR.
• Experimental results demonstrate the speed and efficiency of the word shape coding scheme.
• The method shows tolerance to various types of document degradation, maintaining retrieval accuracy.

Conclusions:

• The developed word shape coding technique offers a viable alternative to OCR-based document image retrieval.
• The approach is effective for searching document images based on visual content and topological features.
• The technique provides a fast, efficient, and robust solution for document image retrieval, even in the presence of degradation.