Doctoral researcher: Anastasia Tsifouti
To date there has not been much direct research carried out on the subject of image quality of CCTV imagery. Resources come mainly from psychology for face identification, event recognition, ergonomic on comfort of reviewing CCTV footage, military applications (most of the information is restricted).
CCTV footage is used by the police for the completion of three main tasks: i) the identification of a person (ie. from facial information) ii) an action (eg. who gave the first punch), and iii) an object (ie. number plate). The police employ both subjective (visual examinations) and objective (automated systems such as face recognition) methods for completing identification tasks from CCTV imagery.
In the project, the following areas are being investigated:
- Acceptable compression and frame rate limits for human face identification, using psychophysical investigations, a variety of image quality attributes, and scene content properties.
- Performance of automated face matching algorithms with reduced video data (either compressed and/or with reduced frame rate), using a variety of image quality attributes and scene contents. Matching face algorithms provide a similarity score (eg. between 0% to 100%) between two facial images.
- Performance of detection systems (eg. detection of intruders) with reduced video data (compressed and/or frame rate reduction) using a variety of image quality attributes and scene contents. Detection systems either detect an event or not.
- Derive correlations on the effects of compression (and frame rate reduction) between subjective (human based tasks) and objective (automated based tasks) methods.
This research is intended to contribute to the production of government guidance on the use of CCTV systems. It will inspire research in other disciples to incorporate image quality aspects. For example, in the area of psychology and face identification, scene content characterisation is rarely used when conducting experiments with facial imagery.