A Fidelity Metric for Image Sequences
Abstract
The development of a computational model of human vision, which might also be used as an image fidelity metric, requires basic psychophysical research to characterize the mechanisms of early vision, with special emphasis on visual masking. We have performed a variety of psychophysical experiments using the test-pedestal paradigm to study visual acuity and motion discrimination performance. Performance using simple targets can often be predicted from the observers own contrast sensitivity. The mystery of how humans achieved hyper acuity performance on many spatial vision tasks has essentially been solved. However, performance on simple psychophysical tasks appears to have limited application to detection of a target in complex backgrounds, typical of video fidelity assessment tasks. Several studies indicate the visual system can use adaptive template mechanisms in complex tasks, which are not readily modeled using the fixed filter properties of current early vision models. While masking by contrast gain control mechanisms may be important in simple stimulus known exactly tasks, its roll is less significant in video quality where artifacts are to be discriminated from an unknown background pedestal. Future work should focus on the adaptive nature of visual mechanisms and their task dependent properties.
Document Details
- Document Type
- Technical Report
- Publication Date
- Mar 30, 2000
- Accession Number
- ADA379898
Entities
People
- Thom Carney