Prof. Jesus Malo from the University of Valencia will give a talk titled
"V1 non-linearities perform non-linear ICA on natural images"
Time and place: Tue 22nd June at 14:15 in C122
Abstract:
A number of relations may be established among the phenomenology
of human vision, image statistics, and practical image processing tasks:
(1) Image statistics may explain the non-linear behavior and the
adaptation abilities of biological visual sensors. (2) The experimental
non-linear behavior of visual mechanisms has appealing statistical
properties. (3) The experimental non-linear behavior of visual sensors
is quite useful in image processing. (4) Statistical models of images
are the key in many image processing applications.
In the first part of the talk I will review our latest contributions in
the above directions:
* Non-linearities and adaptation abilities of human color mechanisms can
be obtained from natural colors under different illuminations using our
Principal Curves Unfolding (PCU) for manifold description [1].
* Psychophysically tuned V1 model factorizes the PDF of natural images [2].
* Psychophysically tuned V1 model gives rise to better measures of image
distortion than SSIM, VIF and VSNR [3].
* Image PDFs obtained using our Rotation-Based Iterative Gaussianization
(RBIG) provide promising results in a number of image processing
applications such as texture synthesis, image classification, image
denoising and multi-information estimation [4].
In the second part of the talk I will focus on the details of the
psychophysically tuned Divisive Normalization and its statistical
properties.
References:
[1] V. Laparra, G.Camps and J. Malo,
Principal Curves Unfolding with Local Metric
Submitted to NIPS 2010
[2] J. Malo and V. Laparra,
Psychophysically tuned V1 model factorizes the PDF of natural
images
In press. Neural Computation 2010
http://www.uv.es/vista/vistavalencia/papers/aneural10.html
[3] Divisive Normalization Image Quality Metric Revisited
J. Opt. Soc. Am. A 27(4): 852-864 (2010)
http://www.uv.es/vista/vistavalencia/papers/ajosa10.html
[4] V. Laparra, G. Camps, J. Malo.
Iterative Gaussianization: from ICA to Random Rotations.
Submitted to IEEE Tr. Neur. Networks. 2010