Image contrast enhancing method

• US 5,717,791 A
• Filed: 11/09/1995
• Issued: 02/10/1998
• Est. Priority Date: 11/10/1994
• Status: Expired due to Term

• Current Assignee: AGFA Healthcare NV (Agfa-Gevaert N.V.)
• Original Assignee: Agfa-Gevaert N.V.
• Inventors: Labaere, Francis, Vuylsteke, Pieter
• Primary Examiner(s): Rogers, Scott A.
• Application Number:
  US08/554,918
• Time in Patent Office: 824 Days
• Field of Search: 382/274, 382/131, 382/132, 382/128, 382/263, 382/266, 382/254, 382/299, 378/4, 378/21, 378/901, 395/128
• US Class Current: 382/274
• CPC Class Codes: G06T 2207/20016 Hierarchical, coarse-to-fine, multiscale or multiresolution image processing; Pyramid transform
• CPC Class Codes: G06T 2207/20064 Wavelet transform [DWT]
• CPC Class Codes: G06T 5/75 Unsharp masking
• CPC Class Codes: G06T 5/90 Dynamic range modification of images or parts thereof

Abstract

A contrast enhancing processing method is disclosed that is based on a transformation of an original image into a multiresolution edge representation based on a wavelet transform. The multiresolution edge representation is modified and subsequently subjected to a reconstruction procedure. Different embodiments are disclosed that are based on different forms of the multiresolution edge representation.

Claims

1. A method of enhancing the contrast of an original image by processing said image in a digital computer, said processing comprising the steps oftransforming the original image into a multi-resolution edge representation comprising wavelet maxima at multiple resolution levels and a residual image,

each wavelet maximum being associated with an edge in the original image, and having a value that is proportional to the maximal edge slope at a specific resolution level and the residual image being an approximation of the original image at a low resolution level,

for each wavelet maximum at the highest of said resolution levels determining a wavelet maxima curve comprising wavelet maxima at subsequent lower resolution levels that correspond to the same edge in the original image,modifying the wavelet maxima of each wavelet maxima curve by multiplying said wavelet maxima with a factor `af` that depends on the average value of the values of the wavelet maxima of said wavelet maxima curve,

said dependence being such that if a wavelet maxima curve with larger average wavelet maximum yields a factor af₁, and a wavelet maxima curve with smaller average wavelet maximum yields a factor af₂, then af₁ ≦

af₂

4) computing a processed image by applying an inverse procedure to the residual image and the modified wavelet maxima, the inverse procedure being such that if it is applied to the residual image and the unmodified wavelet maxima, then the original image or a close approximation thereof will result.

2. A method according to claim 1 wherein said multiresolution edge representation is obtained bydecomposing the original image into a weighted sum of wavelet basis functions at multiple resolution levels and in multiple directions and a residual image so as to yield wavelet coefficients and a residual image,

each wavelet coefficient being proportional to the edge slope in a specific direction determined at a specific resolution level, and the residual image being an approximation of the original image at a low resolution level,

for each resolution level determining wavelet maxima as the locations in the multiresolution edge representation where the modulus of the corresponding wavelet coefficients is locally maximal.

3. A method according to claim 2 wherein said decomposition is performed by:

1) applying one-dimensional high-pass filters independently to the rows and the columns of the original image and applying a two-dimensional low-pass filter to the original image, the former filters yielding horizontal and vertical wavelet coefficients respectively at the highest resolution level, and the latter filter yielding an approximation of the original image at the highest of said resolution levels,

2) identifying the above operations as the first step of an iteration loop, and performing additional iterations using an approximation image resulting from the previous iteration instead of the original image, where the pass-through frequency of the high-pass filters and the cut-off frequency of the low-pass filter are reduced at each iteration, yielding wavelet coefficients and an approximation of the original image at subsequent lower resolution levels and wherein said residual image is the result of the low-pass filter after the last iteration.

4. A method according to claim 3 wherein said reduction of pass-through frequencies at each iteration is performed by a factor of two.

5. A method according to claim 4 wherein the filter coefficients used in the first iteration are (-2, 2) for the high-pass filters, and ((1/64, 3/64, 3/64, 1/64), (3/64, 9/64, 9/64, 3/64), (3/64, 9/64, 9/64, 3/64), (1/64, 3/64, 3/64, 1/64)) for the low-pass filter and all coefficients are multiplied by a correction factor, and wherein the filters in the subsequent iterations are obtained by using the above values multiplied by a correction factor, and by inserting 0-valued coefficients every other tap position horizontally, vertically, or both, respectively at each iteration,. said correction factor being equal to 0.66 for the first iteration, 0.89, 0.97, 0.99 for the second through the fourth iteration, and 1 for the subsequent iterations.

6. A method according to claim 2 wherein said wavelet maxima curves comprise connected wavelet maxima, each wavelet maxima curve being determined byiterating the following procedure, starting at the second highest resolution level i=1, down to the lowest resolution level i=L:

i) for each wavelet maximum at the current resolution level i, centering a rectangular neighbourhood box at the corresponding location at the higher resolution level i-1, oriented either horizontally or vertically according to the principal direction of the wavelet coefficients

ii) in each such neighbourhood box selecting the largest wavelet maximum for which the vector angle of the corresponding wavelet coefficients is equal up to some tolerance to the vector angle of said wavelet maximum at the current resolution level i

iii) if such largest wavelet maximum at the higher resolution level i-1 has been selected, appending said wavelet maximum at the current resolution level i to a list of candidates for connection associated with said selected wavelet maximum

iv) after having performed steps i) through iii) for all wavelet maxima at the current resolution level i,

establishing a connection between each wavelet maximum at the higher resolution level i-1 and the largest wavelet maximum at the current resolution level i that has been appended to said list of candidates associated with said wavelet maximum at the higher resolution level i-1,

2) for each wavelet maximum at the highest resolution level, i=0, associating a wavelet maxima curve, and including said wavelet maximum at the highest resolution level and all its subsequently connected wavelet maxima at lower resolution levels in said wavelet maxima curve.

7. A method according to claim 1 wherein said factor is equal to af=(x/m)^p-1, wherein x is the average of the wavelet maxima of said wavelet maxima curve, and m specifies the range of applicable abscissa values:

0≦

x≦

m, and p is a parameter in the range 0<

p<

1.

8. A method according to claim 1 wherein said factor is equal to:

space="preserve" listing-type="equation">af=(m/x)*(c/m).sup.q *(x/c).sup.p, if 0≦

x<

c,

space="preserve" listing-type="equation">af=(x/m).sup.q-1 if c≦

x≦

m,

wherein x is the average of the wavelet maxima of said wavelet maxima curve, wherein m specifies the range of applicable abscissa values;

0≦

x≦

m, and the cross-over parameter c is in the range 0<

c<

m, and wherein p and q are parameters such that 0<

q<

1 and p≦

q.

17. A method according to claim 1 wherein said original image obtained by scanning a photostimulable phosphor screen that had been exposed to X-radiation by means of stimulating radiation, detecting light emitted upon stimulation and converting said light into an electric representation.

9. A method of enhancing the contrast of an original image by processing said image in a digital computer, said processing comprising the steps oftransforming the original image into a multiresolution edge representation, comprising wavelet coefficients at multiple resolution levels and in multiple directions and a residual image, each wavelet coefficient being associated with an edge pixel in the original image, and having a value that is proportional to the edge slope in a specific direction determined at a specific resolution level and the residual image being an approximation of the original image at a low resolution level,for each resolution level determining wavelet maxima as the locations in the multiresolution edge representation where the modulus of the corresponding wavelet coefficients is locally maximal,for each wavelet maximum at the highest of said resolution levels determining an initial wavelet maxima cluster, said initial wavelet maxima cluster comprising wavelet maxima at subsequent lower resolution levels, that correspond to the same edge in the original image,extending said initial wavelet maxima clusters at each resolution level by assigning each location of said multiresolution edge representation to an initial wavelet maxima cluster,modifying the wavelet coefficients of said multiresolution edge representation by multiplication with a factor `af` which depends on the average of the wavelet maxima of the corresponding wavelet maxima cluster,

wherein the dependence is such that if a wavelet maxima cluster with larger average wavelet maximum yields a factor af₁, and a wavelet maxima cluster with smaller average wavelet maximum yields a factor af₂, then af₁ ≦

af₂,

6) computing the processed image by applying an inverse transform to the residual image and the modified wavelet coefficients, the inverse transform being such that if it is applied to the residual image and the unmodified wavelet coefficients, then the original image or a close approximation thereof will result.

10. A method according to claim 9 wherein said transformation is performed byapplying one-dimensional high-pass filters independently to the rows and the columns of the original image and a two-dimensional low-pass filter to the original image, the former filters yielding horizontal and vertical wavelet coefficients respectively at the highest resolution level, and the latter filter yielding an approximation of the original image at the highest resolution level,identifying the above operations as the first step of an iteration loop, and performing additional iterations using the approximation image resulting from the previous iteration instead of the original image, where the pass-through frequency of the high-pass filters and the cut-off frequency of the low-pass filter are reduced at each iteration, yielding wavelet coefficients and an approximation of the original image at subsequent lower resolution levels,

and wherein said residual image is the result of the low-pass filter after the last iteration.

11. A method according to claim 10 wherein said reduction of pass-through frequency and cut-off frequency at each iteration is performed by a factor of two.

12. A method according to claim 11 wherein the filter coefficients used in the first iteration are (-2, 2) for the high-pass filters, and ((1/64, 3/64, 3/64, 1/64), (3/64, 9/64, 9/64, 3/64), (3/64, 9/64, 9/64, 3/64), (1/64, 3/64, 3/64, 1/64)) for the low-pass filter, all coefficients are multiplied by a correction factor, and wherein the filters in the subsequent iterations are obtained by using the above values multiplied by a correction factor, and by inserting 0-valued coefficients every other tap position horizontally, vertically, or both, respectively at each iteration, said correction factor being equal to 0.66 for the first iteration, 0.89, 0.97, 0.99 for the second through the fourth iteration, and 1 for the subsequent iterations.

13. A method according to claim 9 wherein said wavelet maxima clusters initially comprise connected wavelet maxima, each initial wavelet maxima cluster being determined byiterating the following procedure, starting at the second highest resolution level i=l, down to the lowest resolution level i=L:

i) for each wavelet maximum at the current resolution level i, centering a rectangular neighbourhood box at the corresponding location at the higher resolution level i-1, oriented either horizontally or vertically according to the principal direction of the wavelet coefficients,

ii) in each such neighbourhood box selecting the largest wavelet maximum for which the vector angle of the corresponding wavelet coefficients is equal up to some tolerance to the vector angle of said wavelet maximum at the current resolution level i,

iii) if such largest wavelet maximum at the higher resolution level i-1 has been selected, appending said wavelet maximum at the current resolution level i to a list of candidates for connection associated with said selected wavelet maximum,

iv) after having performed steps i) through iii) for all wavelet maxima at the current resolution level i, establishing a connection between each wavelet maximum at the higher resolution level i-1 and the largest wavelet maximum at the current resolution level i that has been appended to said list of candidates associated with said wavelet maximum at the higher resolution level i-1,

2) for each wavelet maximum at the highest resolution level, i=0, associating an initial wavelet maxima cluster, and including said wavelet maximum at the highest resolution level and all its subsequently connected wavelet maxima at lower resolution levels in said initial wavelet maxima cluster.

14. A method according to claim 9 wherein said initial wavelet maxima clusters are extended byat each resolution level for each initial wavelet maxima cluster determining at each location of said multiresolution edge representation a field strength generated by said initial wavelet maxima cluster according to E=L_wave /(r-1), wherein L_wave is the number of wavelet maxima in said initial wavelet maxima cluster,

and wherein r is the distance from said location to the location of the corresponding wavelet maximum at said resolution level included in said initial wavelet maxima cluster,

at each resolution level assigning each location of said multiresolution edge representation to that initial wavelet maxima cluster which generates the largest field strength at said resolution level and location.

15. A method according to claim 9 wherein said factor is equal to af=(x/m)^p-1, wherein x is the average of the wavelet maxima of said initial wavelet maxima cluster, wherein m specifies the range of applicable abscissa values:

0≦

x≦

m, and wherein p is a parameter in the range 0<

p<

1.

16. A method according to claim 9 wherein said factor is equal to:

space="preserve" listing-type="equation">af=(m/x)*(c/m).sup.q *(x/c).sup.p, if 0≦

x<

c,

space="preserve" listing-type="equation">af=(x/m).sup.q-1 if c≦

x≦

m,

wherein x is the average of the wavelet maxima of said initial wavelet maxima cluster, wherein m specifies the range of applicable abscissa values;

0≦

x≦

m, and the cross-over parameter c is in the range 0<

c<

m, and p and q are parameters such that 0<

q<

1 and p<

q.