• Frame selection

This task aims at producing a semi-automatic tool to assist the human expert in selecting the most meaningful frames. Due to the huge quantity of manufacturers operating in the CCTV marketplace, there are a broad range of system for retrieve and export images from compressed video. This task will analyze the most common formats and produce a tool that will leverage on already available retrieval techniques provided either by the developer of CCTV system or by the open-source community.

**3.6 Speech and audio processing**

*DOI: http://dx.doi.org/10.5772/intechopen.92548*

*Novel Methods for Forensic Multimedia Data Analysis: Part II*

large amount (terabytes) of data.

• Speech and audio representation

sound and audio files.

tasks.

A significant portion the data collected by law enforcement agencies are speech,

Similarity, case-based reasoning, compressive reasoning and sensing-based novel methods for speech and audio recognition using both temporal and frequency domain information will be developed. Similarity learning, case generalization and case storage and compressive learning and sensing will allow the handling of very

Development of speech and audio representation schemes using both spectral,

wavelet scattering and temporal methods such as delta modulation and zerocrossing information. This novel representation will be used in all of the above

• Case and similarity-based reasoning-based speech and audio recognition Development of 'query by example,' keyword and phrase-based retrieval schemes using conventional and structural similarity-based methods capable of part and whole similarity matching. Once the keyword and phrases are detected,

Exemplar-based speech, speaker and audio recognition. The resulting scheme will be computationally efficient, and it will work in the compressed data domain.

Handwritings constitute another important part of the collected data. The objective of this work package is to develop computer vision and pattern recognition methods to identify and recognize the large volumes of unconstrained hand-

Methods will be developed not only for writer identification which is a very important task in forensic applications but also for automatic recognition of the text found in the environment such as notes and letters written by the subject or the

Image enhancement techniques will be carefully applied to reduce the noise without deforming the original data. Alternative to character-based systems, wordbased systems will be developed to read the text 'in the wild' using methods

inspired from object detection and recognition literature. Generic image descriptors such as salient points, gradient histograms and line pairs will be considered for describing words. Efficient and effective similarity measures will be developed to match handwritten text in large volumes in a fast manner without losing any

Words in handwritten text will be described with advanced visual features used in generic object recognition. Novel descriptors based on contours, salient points

scene text in the photos taken, such as the shop labels and billboards.

analysts can manually process the proposed retrieval results.

• Compressive reasoning and recognition

**3.7 Handwriting recognition**

written text to assist the experts.

• Representation of words

and shapes will be generated.

• Similarity-based word matching

important data.

**147**

• Person and object retrieval and identification

This task will be aimed to develop tools to aid the human expert looks for individuals or objects that are useful for the investigation. Object of interests can be, for example, heads, vehicles, license plates, guns, dresses and all other objects that can link a person to the event, etc. The tool will handle the cases of videos with low quality, bad lighting condition, camera/object position and facial expressions.

The persons/objects with enough quality retrieved from the video will be compared by an automatic procedure to a set of known elements of comparison. The expert will then provide feedback on the comparison results in order to refine the search.

An important main focus of police work is the identification of people for which a decision of the public prosecutor's office or a judge to the observation or an arrest warrant was issued. Within this scope of arrangement, the use of video-supervised places and facilities should be used. At earlier not known places, the application of mobile videotechnology should be deployed.

The aim of this task is to develop methods and procedures for an automatic system for identification of one or several target people in mobile video recordings based on passport photos or other available pictures. On this occasion, the prototype-based methods should be used, which are able to work on different picture representations, like pixel, features and graphs. By means of case-based learning mechanisms, a model should be automatically learnt for procedures to the facial recognition under the described application terms. In detail, the following should be developed:


This task is aimed at devising techniques for the re-identification of people in videos captured by different cameras. In many outdoor and indoor environments, different cameras are present to monitor a given area (e.g., in a given street, you can find cameras operated by shops, banks, etc., or by the municipal police). The development of reidentification techniques allows tracking a subject that exits from the field of view of a camera and enters into the field of view of another camera placed in the neighborhood.
