[Paper Review] The Parallel Distributed Image Search Engine (ParaDISE)

Dimitrios Markonis, Roger Schaer+2 more|arXiv (Cornell University)|Jan 19, 2017

Advanced Image and Video Retrieval Techniques17 references3 citations

TL;DR

ParaDISE is an open-source, component-based image retrieval system that combines visual and text-based search using scalable, distributed processing via MapReduce. It supports flexible evaluation of visual features and retrieval pipelines, demonstrating strong performance in medical image retrieval with state-of-the-art descriptors like SIFT and BoVW outperforming more complex ones in specific contexts.

ABSTRACT

Image retrieval is a complex task that differs according to the context and the user requirements in any specific field, for example in a medical environment. Search by text is often not possible or optimal and retrieval by the visual content does not always succeed in modelling high-level concepts that a user is looking for. Modern image retrieval techniques consist of multiple steps and aim to retrieve information from large--scale datasets and not only based on global image appearance but local features and if possible in a connection between visual features and text or semantics. This paper presents the Parallel Distributed Image Search Engine (ParaDISE), an image retrieval system that combines visual search with text--based retrieval and that is available as open source and free of charge. The main design concepts of ParaDISE are flexibility, expandability, scalability and interoperability. These concepts constitute the system, able to be used both in real-world applications and as an image retrieval research platform. Apart from the architecture and the implementation of the system, two use cases are described, an application of ParaDISE in retrieval of images from the medical literature and a visual feature evaluation for medical image retrieval. Future steps include the creation of an open source community that will contribute and expand this platform based on the existing parts.

Motivation & Objective

• To address the semantic gap in content-based image retrieval by integrating visual and textual features.
• To design a flexible, extensible, and scalable image retrieval platform suitable for both research and real-world deployment.
• To support efficient large-scale image indexing and retrieval using distributed computing and modern feature representations.
• To evaluate the effectiveness of various visual descriptors in medical image retrieval contexts.
• To enable interoperability and integration into larger systems through RESTful Web services and JSON-based data exchange.

Proposed method

• ParaDISE employs a component-based, plugin-like architecture with four core components: Image Preprocessor, Feature Extractor, Indexer, and Fusor for late fusion.
• It uses the MapReduce framework for parallel indexing and retrieval, enabling scalability across large image collections.
• The system supports multiple visual descriptors (e.g., SIFT, RootSIFT, CEDD, FCTH, BoVW, VLAD) and allows parameter tuning for each component.
• It integrates text-based retrieval via late fusion using the Fusor component, enabling multi-modal query processing.
• The system uses JSON for data interchange and RESTful HTTP interfaces to ensure interoperability with diverse client applications.
• It supports binary descriptors for memory efficiency and efficient online search in large-scale datasets.

Experimental results

Research questions

• RQ1 How can a distributed, scalable image retrieval system effectively combine visual and textual features to improve retrieval accuracy?
• RQ2 Which visual feature representations perform best in medical image retrieval tasks, particularly for small-pathology detection?
• RQ3 To what extent does component modularity and extensibility enhance the system’s adaptability for research and production use?
• RQ4 Can MapReduce-based parallel indexing significantly reduce indexing time for large-scale image collections?
• RQ5 How does the integration of relevance feedback and late fusion improve user satisfaction in medical image search?

Key findings

• SIFT and BoVW outperformed more advanced descriptors like RootSIFT and VLAD in the medical image retrieval evaluation, indicating task-specific feature suitability.
• Global descriptors such as CEDD and FCTH achieved competitive performance, challenging the assumption that local features are always superior.
• The system demonstrated high user satisfaction in a real-world evaluation with radiologists, particularly regarding result trustworthiness and ease of use with relevance feedback.
• Parallel indexing using MapReduce enabled efficient processing of large-scale datasets, supporting real-time retrieval even with collections exceeding 100,000 images.
• The component-based, extensible architecture allowed for seamless integration of new features and retrieval pipelines, supporting both research experimentation and application deployment.
• The system’s open-source availability under dual licensing facilitates adoption in both academic and commercial environments.