DNA Encoding-Based Nucleotide Pattern and Deep Features for Instance and Class-Based Image Retrieval.

Area of Science:

• Bioinformatics
• Computer Vision
• Data Storage

Background:

• DNA encoding offers efficient data storage by representing information as nucleotide sequences (A, C, T, G).
• Traditional methods face challenges in storage efficiency and retrieval accuracy for large image datasets.

Purpose of the Study:

• To enhance content-based image retrieval (CBIR) accuracy by leveraging DNA encoding principles.
• To develop a novel DNA encoding scheme for compact and salient image information representation.

Main Methods:

• Implemented a most significant bit-plane-based DNA encoding scheme to generate DNA planes from images.
• Utilized encoded DNA planes for nucleotide pattern-based feature extraction and image retrieval.
• Applied deep learning architectures (ResNet-50, VGG-16/19, Inception V3) for classification-based retrieval using translated and amplified DNA planes.

Main Results:

• The proposed DNA encoding scheme effectively captures salient visual information in a compact format.
• Nucleotide pattern-based and deep learning-based retrieval demonstrated improved performance.
• Evaluated on diverse datasets (corals, objects, medical images) with 28,200 images across 134 classes.

Conclusions:

• The DNA encoding approach significantly improves image retrieval accuracy compared to existing state-of-the-art methods.
• This method offers a promising direction for efficient image storage and retrieval in CBIR systems.