Shape Similarity and Image Retrieval Techniques

Shape similarity underpins the automated organisation and retrieval of digital images by assessing how closely object outlines resemble one another. At its core, this field seeks representations of object contours that remain stable under translation, rotation, scaling and deformations. Early work introduced global descriptors such as Fourier series and moment invariants to capture overall boundary characteristics, while later developments have embraced local and mid-level representations combining curvature, contour fragments and learned codebooks. Content-based image retrieval systems exploit these descriptors to rank library images by shape proximity, often augmented with interactive relevance feedback to refine search results. Recent trends include deep learning architectures for end-to-end shape matching, graph and diffusion-based methods to exploit manifold structures, and human-in-the-loop skeleton extraction for robust ground-truth generation. These techniques have found applications across medical diagnosis, biometric identification, remote sensing and cultural heritage archiving, offering scalable solutions to the growing demands of multimedia collections.

Research from Nature Portfolio

No recent Nature Portfolio content available.

Shape Similarity and Image Retrieval Techniques publication trend

The graph below shows the total number of articles in shape similarity and image retrieval techniques across all publications each year (not limited to Nature Index journals).

Technical terms

Content-based image retrieval (CBIR): A search paradigm that matches images by analysing visual features rather than textual metadata.

Shape descriptor: A numerical representation of an object’s contour or boundary used to quantify similarity.

Multiscale analysis: A technique that examines shape features at different levels of detail to capture both global and local characteristics.

Fourier descriptor: A global shape representation obtained by applying the Fourier transform to contour coordinates or distance functions.

Relevance feedback: An interactive process where user input on retrieved results refines subsequent search queries.

Diffusion process (in retrieval): A method that propagates similarity scores over a graph or manifold to improve ranking consistency.

References

1. Skeleton Ground Truth Extraction: Methodology, Annotation Tool and Benchmarks. International Journal of Computer Vision (2023).
2. Interactive search in image retrieval: a survey. International Journal of Multimedia Information Retrieval (2012).
3. A Fourier Descriptor of 2D Shapes Based on Multiscale Centroid Contour Distances Used in Object Recognition in Remote Sensing Images. Sensors (2019).
4. Curvature Bag of Words Model for Shape Recognition. IEEE Access (2019).
5. Efficient Rank-Based Diffusion Process with Assured Convergence. Journal of Imaging (2021).
6. Learning Contour-Based Mid-Level Representation for Shape Classification. IEEE Access (2020).