New methods to improve recognition performance of 3D face composite software was the second topic decided upon during the PhD after deciding to shift the focus away from DMT visual hallucinations and into a field where a similar problem existed in regards to accurate transferance of visual detail from memory, and testing was able to be completed with validation of results. The topic on a whole was much more containable and had the ability to be verifiable. One of the benefits discussed was that it would then later build into the topic on DMT visual hallucinations, but unfortunately, as it felt like this was less likely as time went on. The topic was later abandoned after building the foundations in a confirmation document.
Some unique ideas presented included:
- Using celebrities as an initial similarity departure point
- Using a novel randomised feature generator that narrowed after each selection for users
- 3D models include the ability for animation and realism to be included
- 3D models include the ability for lighting conditions matching the memory to assist in retrieval
- Holistic face recognition at all times and the abandonment of individual feature reconstruction
One of the main goals of the research was that instead of a user requiring 20-30 minutes in constructing individual feature details onto an individual, that they were instead required to spend 1-2 minutes and retrieve similar accuracy levels on any face seen for the first time. All of this research and ideas are valid, and I'm interested in discussion with any current researchers in the field on developing further these ideas.
Link to the PhD research document on 3D face composite software for police agencies is here