Method and apparatus for adaptively producing noise-reduced digital image sequences

Claim 1

Method for producing a reduced-noise digital image sequence, comprising the steps of: obtaining a first chronological sequence X of digital images; and generating a second sequence S of digital images with reduced noise from said first chronological sequence X according to the following recursive steps applied pixel-by-pixel to the images in said first chronological sequence X; generating an image S(n) for a time index n according to S(n)=a(n)·S(n-1)+(1-a(n))·X(n) with S(0)=X(0), wherein n=1,2, . . . is a time index identifying the chronological position of an image in each sequence X and S, a(n) is a sequence of filter coefficients and k(n) is a sequence of variance factors, generating said filter coefficients a(n) for a time index n according to ##EQU2## wherein L(n) is a prediction error formed by spatially low-pass-filtering a recursively formed difference X(n)-S(n-1) over an environment having a size U of a pixel being observed, and wherein b(n) is a sequence of a posterior probabilities for the occurrence of motion in a pixel being observed; generating said variance factors k(n) for a time index n according to ##EQU3## with k(0)=1, and estimating said posteriori possibilities b(n) for the occurrence of motion in a pixel being observed, for a time index n according to ##EQU4## wherein p is an a priori probability for the occurrence of motion in the pixel being observed, and c 2 is the estimated noise power in the image sequence X.

Claim 3

An apparatus for producing a reduced-noise digital image sequence, comprising: means for obtaining a first chronological sequence X of digital images; and means for generating a second sequence S of digital images with reduced noise from said first chronological sequence X by acting recursively pixel-by pixel on the images in said first chronological sequence X, including: means for generating an image S(n) for a time index n according to S(n)=a(n)·S(n-1)+(1-a(n))·X(n) with S(0)=X(0), wherein n=1,2, . . . is a time index identifying the chronological position of an image in each sequence X and S, a(n) is a sequence of filter coefficients and k(n) is a sequence of variance factors, means for generating said filter coefficients a(n) for a time index n according to ##EQU5## wherein L(n) is a prediction error formed by spatially low-pass-filtering a recursively formed difference X(n)-S(n-1) over an environment having a size U of a pixel being observed, and wherein b(n) is a sequence of a posterior probabilities for the occurrence of motion in a pixel being observed; means for generating said variance factors k(n) for a time index n according to ##EQU6## with k(0)=1, and means for estimating said posteriori possibilities b(n) for the occurrence of motion in a pixel being observed, for a time index n according to ##EQU7## wherein p is an a priori probability for the occurrence of motion in the pixel being observed, and c 2 is the estimated noise power in the image sequence X.

Claim 4

An apparatus for producing a reduced-noise digital image sequence, comprising: means for obtaining a first chronological sequence X of digital images; and means for generating a second sequence S of digital images with reduced noise from said first chronological sequence X by acting recursively pixel-by pixel on the images in said first chronological sequence X, including: means for generating an image S(n) for a time index n according to S(n)=a(n)·S(n-1)+(1-a(n))·X(n) with S(0)=X(0), wherein n=1,2, . . . is a time index identifying the chronological position of an image in each sequence X and S, a(n) is a sequence of filter coefficients and k(n) is a sequence of variance factors, means for generating said filter coefficients a(n) for a time index n according to ##EQU5## wherein L(n) is a prediction error formed by spatially low-pass-filtering a recursively formed difference X(n)-S(n-1) over an environment having a size U of a pixel being observed, and wherein b(n) is a sequence of a posterior probabilities for the occurrence of motion in a pixel being observed; means for generating said variance factors k(n) for a time index n according to ##EQU6## with k(0)=1, and means for estimating said posteriori possibilities b(n) for the occurrence of motion in a pixel being observed, for a time index n according to ##EQU7## wherein p is an a priori probability for the occurrence of motion in the pixel being observed, and c 2 is the estimated noise power in the image sequence X.