Browse Prior Art Database

Transfer function of cameras or displays

IP.com Disclosure Number: IPCOM000131672D
Published in the IP.com Journal: Volume 5 Issue 11B (2005-12-10)
Included in the Prior Art Database: 2005-Dec-10
Document File: 3 page(s) / 94K

Publishing Venue

Siemens

Related People

Juergen Carstens: CONTACT

Abstract

For acoustic applications the transfer function of transducers (microphones, loudspeakers) is determined in order to study their characteristics (e.g. low path, high path filtering) and to eventually compensate their defaults like enhancement of sound pick-up for microphones or reproduction of sound by loudspeakers. For image and video applications the function transfer is not analyzed at present. Except for blurring and motion compensation, where a transfer function has to be estimated, an analyze is made, using a priori model or adaptive filtering. The following idea proposes a method to estimate the transfer function of cameras or displays in order to analyze their characteristics. At this, a method to determine impulse response, which is commonly used for one time varying signals, is applied for 2D image application. The new solution is based on using the characteristics of a random signal. Initial position is x(i, j), a random digital black and white image of size (M, N), and its infinite periodical extension:

This text was extracted from a PDF file.
This is the abbreviated version, containing approximately 48% of the total text.

Page 1 of 3

S

Transfer function of cameras or displays

Idea: Dr. Christophe Beaugeant, DE-Munich

For acoustic applications the transfer function of transducers (microphones, loudspeakers) is determined in order to study their characteristics (e.g. low path, high path filtering) and to eventually compensate their defaults like enhancement of sound pick-up for microphones or reproduction of sound by loudspeakers. For image and video applications the function transfer is not analyzed at present. Except for blurring and motion compensation, where a transfer function has to be estimated, an analyze is made, using a priori model or adaptive filtering.

The following idea proposes a method to estimate the transfer function of cameras or displays in order to analyze their characteristics. At this, a method to determine impulse response, which is commonly used for one time varying signals, is applied for 2D image application. The new solution is based on using the characteristics of a random signal. Initial position is x(i, j), a random digital black and white image of size (M, N), and its infinite periodical extension:


...


...


...


...

...

⎢ ⎢ ⎢ ⎢ ⎢ ⎢

...

 j i x

 
) , (

...

 j i x

 
) , (

...

 ) , ( ~

 j i x

= .


...


...


...


...

...

⎥ ⎥ ⎥ ⎥ ⎥ ⎥

...

 j i x

 
) , (

...

 j i x

 
) , (

...

.


...


...


...


...

...

For color consideration, the proposed method can be transposed by considering the luminance component of the image. The circular cross-correlation

xx -

φ has the following property (equation 1):

 
) , ( ~
) , ( j i x j i x j i

=

  ) , ( ~ *

-

 
) ,

φ( ε

   Max j i

⎩ ⎨ ⎧ = =

 
) , ( for

   N M N M N M N M N M N M j i

(

  
); ,

 2 (

  );...( ,

  
); 2 ,

 2 (

  
);...; 2 ,

(

  
); 3 ,

 2 (

  )... 3 ,

xxx

   else

with ε << Max.

A device (camera or display) conveys a transformation to the random image x(i, j). The transformation is generated by the transfer function h(i, j) so that the resulting signal is the two dimensional convolution of x(i, j) with h(i, j) (equation 2): y(i, j) = h(i, j) * x(i, j).

The random image is assumed to be perfectly printed or plotted on a perfect display. Here, the case is considered where only the recording camera modifies the image by the applied transfer function h(i, j),
i.e. there is no alteration by the printing or display where the image is plotted, no transfer function between the display and the camera, and no additive noise. The signal y(i, j) is the digital output of the camera. To analyze the display it is assumed that the random image x(i, j) is plotted on the analyzed display and recorded by a perfect camera system without any distortion.

The properties of the random signal (equation 1) leads to the following assumption (equation 3):

φ

=

φ

 ) , ( ~ *

xy

 
) , (

 j i x j i y j i

 
) , ( ~

=

xx

 
) , (

 j i h j i

 ) , ( *

 j i h

 ) , ( ~

,

where

© SIEMENS AG 2005 file: 2005J18960.doc page: 1

Page 2 of 3

S


...


...


...


...

...

...

 j i h

 
) , (

...

 j i h

 
) , (

...

 j i h

 ) , ( ~

= .

⎢ ⎢ ⎢ ⎢ ⎢ ⎢


...


.....