Analysis, synthesis and control of data signals with temporal textures using a linear dynamic system

• US 6,864,897 B2
• Filed: 04/12/2002
• Issued: 03/08/2005
• Est. Priority Date: 04/12/2002
• Status: Expired due to Fees

• Current Assignee: Mitsubishi Electric Research Laboratories, Inc. (Mitsubishi Electric Corporation)
• Original Assignee: Mitsubishi Electric Corporation
• Inventors: Brand, Matthew E.
• Primary Examiner(s): Bella, Matthew C.
• Assistant Examiner(s): BRAUTIGAM, ALYSA N
• Application Number:
  US10/121,785
• Publication Number:
  US 20030193509A1
• Time in Patent Office: 1,061 Days
• Field of Search: 345/582, 701/29, 702/17, 702/196
• US Class Current: 345/582
• CPC Class Codes: G06T 11/001 Texturing; Colouring; Generation of texture or colour inpainting G06T5/77

Abstract

A method generates a synthetic textured data signal by first acquiring a time-invariant input textured data signal. The input textured data signal is sampled to construct an observation matrix. The observation matrix is eigen-coding and factoring to identify a linear dynamic system modeling the input textured data signal. Then, the linear dynamic system can be run forward from an initial state using a quadratic regulator and a random noise signal to generate the synthetic textured data signal.

Claims

1. A method for generating a synthetic textured data signal, comprising:

acquiring a time-invariant input textured data signal including a sequence of frames;

sampling the input textured data signal to construct an observation matrix;

eigen-coding and factoring the observation matrix to identify a linear dynamic system modeling the input textured data signal; and

running the linear dynamic system forward from an initial state using a quadratic regulator and a random noise signal to generate the synthetic textured data signal, further comprising;

selecting a set of keyframes from the sequence, each keyframe having an associated state at a particular time; and

running the linear dynamic system through the keyframes to pass through predetermined states at predetermined times.

2. The method of claim 1 wherein the input textured data signal includes a sequence of frames, and the initial state is estimated from a last frame of the sequence.

3. The method of claim 1 wherein the input textured data signal includes a sequence of frames, and the initial state is estimated from a random frame of the sequence.

4. The method of claim 1 further comprising:

estimating the random noise signal from an estimated noise distributions of the input textured data signal.

5. The method of claim 1 wherein the quadratic regulator is linear.

6. The method of claim 1 wherein the quadratic regulator is non-linear.

7. The method of claim 6 wherein the non-linear quadratic regulator is modeled by a probability density function.

8. The method of claim 7 wherein the probability density function is a Gaussian mixture model.

9. The method of claim 1 wherein the input textured data signal is an acoustic signal.

10. The method of claim 1 wherein the input textured data signal is a video sequence of a natural scene.

11. The method of claim 1 wherein the input textured data signal is a video sequence, and further comprising:

acquiring motion parameters from the input video sequence;

estimating and extracting translations and rotations from the motion parameters to produce residual articulatory motion;

eigen-coding the residual articularly motion to produce a sequence of morph weights; and

concatenating the morph weights with the rotation parameters and the translation to produce the observation matrix.