Erasable Photograph Tagging: A Mobile Application Framework Employing Owner's Voice

Zhenfei Zhao, Hao Luo, Hua Zhong, Bian Yang, Zhe‐Ming LuIEICE 2014 · DOI

Abstract

This letter proposes a mobile application framework named erasable photograph tagging (EPT) for photograph annotation and fast retrieval. The smartphone owner's voice is employed as tags and hidden in the host photograph without an extra feature database aided for retrieval. These digitized tags can be erased anytime with no distortion remaining in the recovered photograph.

Introduction

As smartphone photo libraries expand, locating specific images becomes increasingly difficult. Traditional solutions like manual tagging are cumbersome on mobile devices, while Content-Based Image Retrieval (CBIR) systems often rely on external databases that are prone to corruption or synchronization issues. Furthermore, privacy concerns arise when metadata containing sensitive information is permanently attached to shared images.

To address these challenges, we present the Erasable Photograph Tagging (EPT) framework. EPT allows users to tag photos using their voice, embedding the data directly into the image file using reversible data hiding. This approach ensures the metadata travels with the file, requires no external database, and allows for perfect restoration of the original image by "erasing" the tags.

System Framework

The EPT framework operates through two primary phases: Annotation and Retrieval.

Annotation Phase

When a photo is captured, the user provides a vocal tag (e.g., "beach"). The system extracts Mel-frequency cepstral coefficients (MFCCs) from the audio, which are then processed via robust hashing to create a consistent bit-stream, even in the presence of minor noise. This bit-stream is embedded into the image pixels using reversible data hiding. This technique is invisible to the human eye and, crucially, allows the embedded data to be extracted later to return the image to its bit-perfect original state.

Retrieval Phase

To find a specific photo, the user speaks the search term. The system generates a query hash and compares it against the tags embedded within the photos in the album.

The EPT system architecture illustrating the annotation and retrieval workflows.

Experimental Evaluation

The system was tested using a dataset of 100 JPEG photos captured by a smartphone, categorized into five groups: "car," "beach," "animal," "friend," and "smile."

Performance Metrics

The framework demonstrated high visual quality, with annotated photos maintaining an average Peak Signal-to-Noise Ratio (PSNR) exceeding 50 dB, rendering the tags imperceptible. The embedding capacity reached 0.62 bits per pixel (bpp), sufficient for standard voice tags.

Performance metrics for various speech clips, including tagging and retrieval times.

As shown in the results, the tagging process is highly efficient, averaging approximately 0.20 seconds per image. Retrieval for the 100-photo database took an average of 4.06 seconds.

Key Insights and Future Work

The EPT framework effectively turns the image itself into the database, eliminating the need for external metadata management. The reversibility of the data hiding serves as a significant privacy feature, allowing users to strip personal annotations before public sharing.

However, challenges remain regarding scalability for larger libraries and performance in high-noise environments. Future developments will focus on optimizing search speeds for thousands of images and porting the framework from the current simulation to a native Android application.

Summary updated

Abstract

This letter proposes a mobile application framework named erasable photograph tagging (EPT) for photograph annotation and fast retrieval. The smartphone owner's voice is employed as tags and hidden in the host photograph without an extra feature database aided for retrieval. These digitized tags can be erased anytime with no distortion remaining in the recovered photograph.
BibTeX
@inproceedings{zhao2014erasable,
  author = {Zhenfei Zhao and Hao Luo and Hua Zhong and Bian Yang and Zhe‐Ming Lu},
  title = {{Erasable Photograph Tagging: A Mobile Application Framework Employing Owner's Voice}},
  booktitle = {IEICE},
  year = {2014},
  doi = {10.1587/transinf.e97.d.370},
}
Figure