Pattern Recognition (PatRe)

Image search in large amounts of data

Pattern Recognition is an image search in large amounts of data. In the age of digitalisation, photography is no longer limited by expensive film material, development services and tedious archiving in photo albums. The camera is integrated in the smartphone, always ready to hand or fits in your pocket as a miniature device. Even storage space is available on huge hard disks in almost any quantity. However, this beneficial development also brings with it problems with which most users of digital cameras are familiar: the overview of the photo stock is lost, the search for more pictures with a similar motif is time-consuming and many pictures are bogged down on the hard drive as hidden treasures unnoticed in the shadows.

Approaches for automatic image retrieval have been dealt with for a long time, and search engine providers, among others, have been intensively dealing with this topic. Previous approaches primarily aim to characterise the query image as a whole. Features such as colours or colour distributions play a role in this process.

© Fraunhofer IOSB
Anfragefenster von Pat-Re - links die Bildanfrage, rechts die priorisierte Ergebnisübersicht. Die grüne Zahl steht für die gefundene Anzahl der Ähnlichkeiten der Bilder mit dem Ursprungsbild.

Image pattern analysis and pattern recognition

In contrast, researchers at the Fraunhofer IOSB in Karlsruhe have developed a software (called Pat-Re) which, through an automated analysis of the characteristic patterns present in the query image, is able to recognise similar patterns in images of a possibly large image collection. For search queries, any motifs with pattern like structures are suitable.

The first application of this software was originally developed for special applications together with consulting support by official security forces (Interpol and Federal Criminal Police Office - BKA). The development and optimisation work took place within the framework of the EU project "FastID" ("Fast Identification"). Here Pat-Re is intended to help the official end users, e.g. in coping with major incidents, by assisting the programme in victim identification by means of patterned body modifications (e.g. tattoos).
 

Application in zoological or botanical image archives

Apart from this pattern-based special application, Pat-Re is also likely to find applications in other areas, e.g. for searching zoological or botanical image archives, medical image catalogues or art media libraries.
 

Use without internet connection

Apart from the strictly pattern oriented approach of the Pat-Re programme, the big advantage is that, in contrast to other web applications, you do not have to upload the request image or the image collection to be patterned into the internet. Instead, Pat-Re operates completely autonomously on a home PC or in shielded local networks for authorised applications in the professional sector. However, the software is not suitable for facial recognition; other models and programmes are required for this.

The Fraunhofer IOSB has specialised for many years in automated image and video evaluation, classification procedures and pattern recognition. Many years of preliminary work in the field of remote sensing, aerial image evaluation, image-based security technology and the analysis of video data for object recognition form the algorithmic basis for the Pat-Re software, which can be used both privately and in professional special areas.

 

Content-Based Image Retrieval demo version

To get a personal impression of the performance of the software, the developers at Fraunhofer IOSB have provided a limited demo version (for Windows) for free download.

 

 

 

Department VID of Fraunhofer IOSB

Would you like to learn more about our topics in the area of "video evaluation systems"? Then visit the page of our VID department and get more information.

 

 

Further projects of the VID department

Would you like to get to know more projects and products in the field of "video evaluation systems"? Then visit the project page of our IAS department and get more information.