Degenerate Models of Additive and Multiplicative Motion Transparency

Keith Langley,
Dept. of Psychology,
University College London,
kl@psychol.ucl.ac.uk

Abstract:

We consider a gradient-based motion model that leads to the detection of both additive and multiplicative motion transparencies. The model is, however, based upon a quadratic form. It does not degenerate gracefully. To examine this degeneracy we consider the detection of two transparent 1-d signals. To this case, we present additional constraints that help to constrain the model. However, a coherent or single value motion may equally be interpreted to two local 1-d signals. In view of this, we consider how to assign a preference to one of several models that fit the image data without a residual error. Finally, we reflect upon how transparent algorithms may be realised by neural network representations and highlight some computational advantages therein.

• 1 Introduction
• 2 Multiplicative Transparent Models
• 3 An Artificial Neural Network for Transparency
• 4 Methods
• 5 Results and Discussion
• 6 Conclusions
• References
• A Degeneracy for Coherent Motion