Alan Dolhasz (Birmingham City University) discusses his attempts to develop machine perception systems.
Our ability to create synthetic, yet realistic representations of the real world, such as paintings or computer graphics, is remarkable. With the continual improvement in creative digital tools we are able to blur the line between the real and synthetic even further.
Simultaneously, our ability to consciously detect minute imperfections within this imagery, which break down the illusion of realism, improves with experience. This problem of shifting realism thresholds remains paradoxical and largely underexplored.
As human expectations in this context grow, tools to assist in this problem are scarce, and computational models of perception still far from human performance. While it is possible for computers to make binary decisions regarding the realism or plausibility of imperfections and image artifacts, the problem of making them utilise similar features and methods to humans is nontrivial.
Visual realism in the context of mixed reality and synthetic combinations of objects and scenes is a complex and deeply subjective problem. Human perception of realism is affected by a range of visual properties of the scene and objects within it, from attributes of individual textures, surfaces and objects, to illumination, semantics and style of visual coding, to name a few. On top of this, individual subjective traits and experience of observers further complicate this issue.
In this talk, Alan Dolhasz discusses his work attempting to understand, quantify and leverage human perception of combinations of objects and scenes in order to develop machine perception systems that could aid us in creating more realistic synthetic scenes, as well as detect and localise imperfections.
Alan Dolhasz is a researcher and part-time PhD student at the Digital Media Technology (DMT) Lab, Birmingham City University, with a background in film, sound and visual effects. His research interests include human perception, computer vision, machine learning and mixed and augmented reality.
Prior to his research position, he lectured Sound for Visual Media and Sound Synthesis and Sequencing, as well as running a production company, focusing on filmmaking and visual effects compositing, which largely contributed to developing his research area. Alan also works closely with industry, developing application cases for the research work coming out of the DMT Lab. dmtlab.bcu.ac.uk
Dates & times
|Date||Time||Add to calendar|
|4 Oct 2017||4:00pm - 5:30pm|
If you are attending an event and need the College to help with any mobility requirements you may have, please contact the event organiser in advance to ensure we can accommodate your needs.