Professor of Artificial Intelligence, Vice-Rector for Research and Innovation
ISO/IEC 15938 · VISUAL DESCRIPTION
MPEG-7.
Open-source C# implementations of compact visual descriptors for colour, spatial layout and edge-based image retrieval.
OVERVIEW
Describe visual content for efficient retrieval.
MPEG-7 is a multimedia content-description standard defined in ISO/IEC 15938. Its descriptors are associated with the content itself, enabling structured, efficient searching and matching across multimedia collections.
This page provides C# implementations of the Scalable Color Descriptor, Color Layout Descriptor, Dominant Color Descriptor and Edge Histogram Descriptor. The code adapts the Java implementations developed within LIRE and is distributed under the GNU General Public License.
SCD · COLOUR
Scalable Color
A scalable HSV histogram encoded through a Haar transform.
CLD · LAYOUT
Color Layout
A compact representation of the spatial distribution of colour.
DCD · REGIONS
Dominant Color
A small set of representative colours with spatial coherency.
EHD · EDGES
Edge Histogram
A spatial histogram of directional and non-directional edges.
C# · .NET · GNU GPL
Implementations and examples.
Download the compiled library, complete source code, similarity-matching class and demonstration applications.
CORE DOWNLOADS
Use the descriptors in a C# project.
Add the DLL as a reference in your project, or use the source package when you need to inspect or adapt the implementation.
Each descriptor captures a different aspect of visual content. Open an entry to review its purpose and the corresponding C# usage example.
SCD Scalable Color Descriptor +
The Scalable Color Descriptor is an HSV colour histogram encoded with a Haar transform. Its binary representation is scalable in both the number of bins and the accuracy of the bit representation across a broad range of data rates. It supports colour-based image matching, with retrieval accuracy generally increasing as more representation bits are used.
CLD Color Layout Descriptor +
The Color Layout Descriptor provides a compact representation of the spatial distribution of colour. Its small representation supports efficient image-to-image and sequence-to-sequence matching. Because it retains layout information, it can also support colour-sketch queries that are difficult to express through conventional colour histograms.
DCD Dominant Color Descriptor +
The Dominant Color Descriptor is particularly suitable for objects or regions that can be characterised by a small set of colours. Colour quantisation extracts representative colours and their relative percentages, while a spatial-coherency value supplements the description. The same approach can also describe complete images such as flags or colour trademarks.
EHD Edge Histogram Descriptor +
The Edge Histogram Descriptor represents the spatial distribution of four directional edge types and one non-directional type. It targets image-to-image matching, including example-based and sketch-based retrieval, and is especially useful for natural images with non-uniform edge distributions.
Performance can be improved by combining the descriptor with colour information. When EHD is used alone, matching can benefit from combining its local edge histograms with the semi-global and global histograms derived from them.
RELATED WORK
From global descriptors to local visual features.
The SIMPLE framework localises MPEG-7 SCD, CLD and EHD, together with CEDD, over image regions generated with SURF, SIFT and random-patch strategies. This produces a family of local feature vectors designed for content-based image retrieval.
LIRE: Lucene Image Retrieval, an Extensible Java CBIR Library
M. Lux and S. A. Chatzichristofis. ACM International Conference on Multimedia 2008, Open Source Application Competition, pp. 1085-1087, 27-31 October 2008, Vancouver, British Columbia.
