Method and apparatus for enhancing data resolution

• US 7,149,262 B1
• Filed: 07/06/2001
• Issued: 12/12/2006
• Est. Priority Date: 07/06/2000
• Status: Expired due to Fees

• Current Assignee: Trustees Of Columbia University In The City Of New York (Columbia University)
• Original Assignee: Trustees Of Columbia University In The City Of New York (Columbia University)
• Inventors: Narasimhan, Srinivasa G., Nayar, Shree K.
• Primary Examiner(s): Kumar, Pankaj
• Application Number:
  US10/312,529
• Time in Patent Office: 1,985 Days
• Field of Search: 375/341, 375/377, 375/229, 375/232, 375/316, 375/230, 375/340, 382/167, 382/162, 345/619, 345/620, 345/625, 345/643, 345/418, 396/281, 396/310, 396/439, 342/61, 342/66
• US Class Current: 375/341
• CPC Class Codes: G06T 3/4015 Image demosaicing, e.g. colour filter arrays [CFA] or Bayer patterns
• CPC Class Codes: G06T 3/4046 using neural networks

Abstract

A resolution enhancement algorithm is trained on sample images to obtain a polynomial model mapping of low resolution image data to high resolution image data. The polynomial model mapping is applied to other low resolution images to obtain corresponding higher resolution images. The mapping provides resolution enhancement which is superior to that of conventional image data interpolation techniques.

Claims

1. A method for learning a model for use in data resolution enhancement, comprising:

an image processing step of processing a plurality of high quality images, the image processing step comprising;

applying a simulated detector response function to each of the plurality of high quality images to thereby generate a set of training data,

applying a locally inhomogeneous mosaic pattern function to the training data to thereby generate a set of reduced quality data,

calculating a weighted normal matrix of the reduced quality data,

adding the weighted normal matrix to a matrix sum,

calculating a regression vector of the reduced quality data, and

adding the regression vector to a vector sum;

using the matrix sum to determine an accumulated normal matrix;

using the vector sum to determine an accumulated regression vector; and

solving a least squares equation to determine a coefficient matrix of a model for generating enhanced resolution data, the least squares equation comprising a first side and a second side, the first side comprising a product of the accumulated normal matrix and the coefficient matrix, the second side comprising the accumulated regression vector, and the coefficient matrix comprising coefficients of a polynomial function.

2. A method according to claim 1, wherein the image processing step further comprises downgrading the reduced quality data before the steps of calculating the weighted normal matrix and calculating the regression vector.

3. An apparatus for a learning model for use in data resolution enhancement, comprising:

a first processor for processing a plurality of high quality images, comprising;

a second processor for applying a simulated detector response function to each of the plurality of high quality images to thereby generate a set of training data,

a third processor for applying a locally inhomogeneous mosaic pattern function to the training data to thereby generate a set of reduced quality data,

a fourth processor for calculating a weighted normal matrix of the reduced quality data,

a fifth processor for adding the weighted normal matrix to a matrix sum,

a sixth processor for calculating a regression vector of the reduced quality data, and

a seventh processor for adding the regression vector to a vector sum;

an eighth processor for using the matrix sum to determine an accumulated normal matrix;

a ninth processor for using the vector sum to determine an accumulated regression vector; and

a tenth processor for solving a least squares equation to determine a coefficient matrix of a model for generating enhanced resolution data, the least squares equation comprising a first side and a second side, the first side comprising a product of the accumulated normal matrix and the coefficient matrix, the second side comprising the accumulated regression vector, and the coefficient matrix comprising coefficients of a polynomial function.

4. An apparatus according to claim 3, wherein the first processor further comprises an eleventh processor for downgrading the reduced quality data.

5. A computer-readable medium having a set of instructions operable to direct a processor to perform the steps of:

an image processing step of processing a plurality of high quality images, the image processing step comprising;

applying a simulated detector response function to each of the plurality of high quality images to thereby generate a set of training data,

applying a locally inhomogeneous mosaic pattern function to the training data to thereby generate a set of reduced quality data,

calculating a weighted normal matrix of the reduced quality data,

adding the weighted normal matrix to a matrix sum,

calculating a regression vector of the reduced quality data, and

adding the regression rector to a rector sum;

using the matrix sum to determine an accumulated normal matrix;

using the vector sum to determine an accumulated regression vector; and

solving a least squares equation to determine a coefficient matrix of a model for generating enhanced resolution data, the least squares equation comprising a first side and a second side, the first side comprising a product of the accumulated normal matrix and the coefficient matrix, the second side comprising the accumulated regression vector, and the coefficient matrix comprising coefficients of a polynomial function.

6. A computer-readable medium according to claim 5, wherein the image processing step further comprises downgrading the reduced quality data before the steps of calculating the weighted normal matrix and calculating the regression vector.