2DKD: a toolkit for content-based local image search

Area of Science:

• Computer Vision
• Image Analysis
• Bioinformatics

Background:

• Direct comparison of 2D images is computationally intensive due to transformations like translation, rotation, and scaling.
• Identifying local image regions is crucial in fields such as digital pathology and cryo-electron microscopy (cryo-EM).
• There is a need for invariant descriptors to efficiently retrieve local image patches or subimages.

Purpose of the Study:

• To present a software package for efficient local subimage search in 2D images.
• To enable fast and accurate retrieval of local patterns and small particles within large image databases.
• To provide a solution for computationally cumbersome local image comparison tasks.

Main Methods:

• Development of the Two-Dimensional Krawtchouk Descriptors (2DKD) software package.
• Utilizing a small number of invariant descriptors per image for local retrieval.
• Implementing efficient querying and comparison of local image patterns.

Main Results:

• The 2DKD package facilitates efficient local subimage search.
• The toolkit uses a minimal set of invariant descriptors for effective image retrieval.
• Demonstrated utility in searching local patterns and small particles across diverse test cases.

Conclusions:

• The 2DKD toolkit significantly reduces computational complexity and runtime for local image comparison.
• It achieves efficient subimage search with minimal memory usage.
• The software is beneficial for developers and users in image analysis applications.