WP5 Meeting between AGH and GHP
During the meeting a presentation was made on the “INACT” (INdect Advanced image Catalogue Tool) system, introducing police officers into principles of construction and operation services for cataloguing and searching for similar images in evidence.
During the meeting, the following modules included in the INACT have been presented:
- Indexer – with the principles of obtaining and indexing patterns from photos.
- Searcher – for searching for images based on available local resources, patterns of computer or data storage media. Further directions of development of the INACT tool has been established, as: working on the optimization of time for searching of media, and extending the functionality of the program to search for images that do not have the pattern in the database and may bear the characteristics of child pornography.
Next, the INDECT Crawler (INCR) system has been presented. INDECT Crawler is a search engine for images on the specified web pages. Development of this module (expandable through plug-ins), aims at (among others) providing the following functionality:
- search for specific images on the given websites,
- face search and recognition,
- the use of OCR (also with analysis of video sub-titles),
- speech recognition
- disclosure of the characteristics of images of child pornography
The INCR Multimedia INDECT Crawler, allows for searching for symbols that might be potentially related to various hate crimes. The symbols currently considered include swastika, pentagram, inverted cross or hammer and sickle. It should be noted that there are different juridical systems across Europe, and therefore the definition of forbidden symbols may vary depending on possible future implementation of INCR in police departments in a given European country.
The search process is not network-wide but it can only take place within Websites that have been previously, manually preselected by police operational work.
Automatic Detection of Child Pornography
In the next part of the meeting, presentation was held on work on the functionality to automatically detect child pornography on the Internet resources on the basis of body, face and age (based on the characteristics of persons in the age of 12 years for example).
Methods have been presented for the classification of age on the basis of face images, and the possibilities and limitations of this solution have been discussed. It was underlined that it is possible to determine the age of only some broad divergence which can occur up to several years. Therefore, it was suggested to utilize other ways of classifying people by age: classification on the basis of body proportions, skin properties, characteristics and development, etc.Because it is impossible to complete the classification of images using single method, pictures should be classified using several different methods, which will establish a ranking of the most suspicious images.
It has been proposed to combine two systems such as a system for language-based Internet monitoring and INACT/INCR to achieve a full analysis of the available Internet resources for graphics and text.
Also, development of additional plug-ins to the INACT software, implementing speech recognition and character recognition (OCR) with subtitles movies, has been discussed.
There was a question of installing INACT. It was proposed to use a virtual machine or a version for Windows systems to facilitate the use of software. The INACT system must operate in the field, so that it can be used during searches. Here there is also the question of optimizing the network interface. It could use some simple manual.
In the long term it is very important that INACT was extended to the movies – it is needed to index and (later) search because the police have much more to do with movies. Videos can be seen as dynamic images; however, it remains as the question of how to adequate sample moviesto not to create too large databases, support of which would be cumbersome. Probably for movies a separate database will be needed. Frames of videos can be very useful, because they contain information such as on the environment.
- It is important to extend the functionality of the possibility of indexing dynamic images (videos) and, consequently, the disclosure of electronic media in the Internet.
- The concept (model) will be created, showing what elements should be included in reports ending procedures of searching resources, in accordance with the requirements (Annex to the Memorandum, the date, time, place, etc.).
- AGH needs information about the specialized vocabulary, expressions, etc. to the report to be consistent with police documents.
- The police start distribution of software to the police departments involved in the fight against child pornography on the selected Police Headquarters in order to test and to obtain suggestions and possible observations in terms of functionality and ergonomics of the software.
- There is a proposal to use Virtual Box in order to facilitate the installation and use of software in the selected Police Headquarters.
- The planned functionalities include the configurability of age in the system searches in order to allow operation in different European legal systems as well as in connection with the planned changes to the Polish Codex framework (age greater than 18 years) as well as the classification of the shape of face, body proportions, characteristics and development, etc.
- In the longer perspective it is planned to combine functionalities of INACT and INCR, in order to optimize the search, i.e. query from the database of descriptors, the query to the Internet crawler, search, and self-contained child pornography search on the Internet. There must be a limit to the degree in subsequent links, so as it is not necessarily to search the sites that child pornography will have nothing in common.
- Software provided to the police, met with high praise, because it will increase the effectiveness of police work.