Distortion-free image contrast enhancement

• US 6,633,684 B1
• Filed: 07/07/2000
• Issued: 10/14/2003
• Est. Priority Date: 07/07/2000
• Status: Expired due to Term

• Current Assignee: Athentech Imaging, Inc. (EyeQ Imaging, Inc.)
• Original Assignee: Athentech Technologies Corp.
• Inventors: James, Brian G.
• Primary Examiner(s): Mehta, Bhavesh M.
• Assistant Examiner(s): Choobin, Barry
• Application Number:
  US09/611,773
• Time in Patent Office: 1,194 Days
• Field of Search: 382/274, 382/132, 382/131, 382/168, 382/172, 382/298, 382/270, 250/205, 345/20, 345/63, 345/77, 345/596-599, 345/690-697, 348/251, 348/254, 358/461
• US Class Current: 382/274
• CPC Class Codes: G06T 2207/10116 X-ray image
• CPC Class Codes: G06T 2207/30004 Biomedical image processing
• CPC Class Codes: G06T 5/40 using histogram techniques
• CPC Class Codes: G06T 5/94 based on local image properties, e.g. for local contrast enhancement

Abstract

An image, such as an X-ray, is enhanced through local enhancement of the contrast of the image'"'"'s point intensities. A first, low frequency upper curve is fitted to the local maximums and a second independent, low frequency lower curve is fit to the local minimums, forming a fairway with the raw image data residing therebetween. A local range, between the fairway local maximum intensity and fairway local minimum intensity, is extracted for each point. Each point is scaled by the ratio between the fairway'"'"'s local range and the dynamic range for the image so as to maximize its variation in intensity between it and its neighboring points. Preferably an iterative moving average technique is used to establish the fairway curves. In a preferred embodiment, outlier points scaled outside the fairway are temporarily stored at higher precision than the dynamic range. A histogram of the fairway corrected data is formed, having a range greater than the dynamic range and encompassing substantially all the outlier points. Only the most deviant of the outliers are trimmed in this histogram correction and the resulting range limits for the entire image are scaled to the dynamic range.

Claims

1. A method for maximizing contrast between neighboring points of a data population selected from a digital image, each image point being defined by a position and an intensity, the image having an intensity dynamic range, the method comprising the steps of:

(a) determining a first low frequency trending function which is curve fit to the data population'"'"'s maximum intensities;

(b) determining a second low frequency trending function, independent from the first trending function, and which is curve fit to the data population'"'"'s minimum intensities;

(c) establishing a maximum and minimum fairway for the data population bounded by the first and second trending functions; and

for each point in the data population,

(d) extracting a range of the local minimum and maximum intensity from the fairway;

(e) determining a local scaling factor as the ratio between the dynamic range for the image and the extracted local range; and

f) scaling the point by the local scaling factor.

2. The method of claim 1 wherein the point is scaled using the relationship of $\frac{\text{(Image }}{}$

 

Point-Local 

 

Minimum)(Local 

 

Maximum-Local 

 

Minimum)

Dynamic 

 

Range.

3. The method of claim 1 wherein one or more outlier points for the data population exist, which have intensities outside the fairway, further comprising the steps of:

(a) scaling each outlier point by the local scaling factor with the result that the scaled intensity is outside the image'"'"'s dynamic range;

(b) storing the outlier point'"'"'s scaled intensity at a precision greater than the dynamic range to prevent loss of intensity information;

(c) forming an intensity histogram of the entire data population, the histogram establishing a predetermined lower intensity and a predetermined upper intensity for forming a range which is greater than the dynamic range and which encompasses the intensities of substantially all the outliers;

(d) trimming the intensity histogram of points which have an intensity which is below the predetermined lower intensity, and above the predetermined upper intensity, for establishing a trimmed data population having a trimmed range between minimum and maximum trimmed intensities; and

(e) scaling each point of the trimmed data population by the ratio of the trimmed range to the image'"'"'s dynamic range.

4. The method of claim 2 wherein the data population is adjusted to the greater intensities suited to human vision, further comprising the steps of:

(a) scaling each point of the trimmed data population to an output range less than the image'"'"'s dynamic range; and

(b) offsetting each scaled point by an incremental intensity value which is less than the difference between the dynamic range and the output range so that the scaled points reside in a higher range of intensities which is still within the dynamic range.

5. The method of claim 1 wherein determination of each of the first and second trending functions comprises the steps of:

(a) determining a first function representing the entire data population;

(b) iteratively determining a successive upper function for a residual subset of points of the data population which have intensities greater than the greater of the first function or a previous successive upper function, converging upwardly until fewer than a predetermined number of points are greater than the successive upper function, the converged successive upper function forming the first trending function; and

(c) iteratively determining a successive lower function for a residual subset of points of the data population which have intensities lower than the lower of the first function or a previous successive lower function, until fewer than a predetermined residual number of points are lower than the successive lower function, the converged successive lower function forming the second trending function.

6. The method of claim 5 wherein the convergence of one or both of the upper and lower functions is improved by:

(a) determining the differences between the residual subset of points and each successive function in the iteration;

(b) amplifying the differences and adding it to a subset of points to form an exaggerated subset of points; and

(c) applying the iterative determination of successive functions to the exaggerated subset of points.

7. The method of claim 6 wherein the differences are amplified by incrementing a counter each iteration and multiplying the differences by the counter.

8. The method of claim 5 wherein one or more of the functions applied to determine each of the first and second trending functions is a moving average.

9. The method of claim 8 wherein the function applied for each of the first, successive upper and successive lower curves is a moving average.

10. The method of claim 8 wherein the moving average has a filter box of predetermined size of a column dimension and a row dimension and which is optimized in two dimensions for the image by:

(a) extending the image at the edges of the image by a number of points complementary to the filter box column and row dimensions;

(b) determining a first sum of the intensities of a one dimensional subset of points, for the size of the filter box, and about each image point along a first column or row dimension;

(c) storing the subset sums, indexed to each point in the image;

(d) determining a second sum of the subset sums, for the size of the filter box, and about each image point along a row or column dimension; and

(e) normalizing the second sum by dividing by the number of points in the filter box.

11. The method of claim 10 wherein the image has specified dimensions and the filter box dimensions are set to a percentage of the image dimensions for adjusting the frequency of one or both of the first and second trending functions.

12. The method of claim 11 wherein the filter box dimensions are between about 5 and 20% of the image dimensions.

13. The method of claim 12 wherein the image is an X-ray of a chest and the filter box dimensions are about 5%.

14. The method of claim 10 wherein the local range of the fairway image is constrained to be no narrower than a percentage of the image dynamic range for suppressing noise.

15. The method of claim 14 wherein the minimum range of the fairway image is constrained between about 3 and 12%.

16. The method of claim 15 wherein the minimum range of the fairway image is constrained to about 6%.

17. The method of claim 10 wherein the image is extended at the image edges by mirroring of the image data at the edges.

18. The method of claim 10 wherein

(a) the image has specified dimensions and the filter box dimensions are set to a percentage of the image dimensions for adjusting the frequency of one or both of the first and second trending functions;

(b) the local range of the fairway image is constrained to be no narrower than a percentage of the image dynamic range for suppressing noise; and

(c) the frequency and noise suppression percentages are adjustable by an diagnostician.

19. A system for maximizing contrast between neighboring points of a data population selected from a digital image, each image point being defined by a position and an intensity, the image having an intensity dynamic range, the system comprising:

(a) means for determining a first low frequency trending function which is curve fit to the data population'"'"'s maximum intensities;

(b) means for determining a second low frequency trending function, independent from the first trending function, and which is curve fit to the data population'"'"'s minimum intensities; and

(c) means for establishing a maximum and minimum fairway for the data population bounded by the first and second trending functions so that for each point in the data population, a range of the local minimum and maximum intensity from the fairway can be extracted, a local scaling factor can be determined as the ratio between the dynamic range for the image and the extracted local range, and each point can be scaled by local scaling factor.