Method for improving the quality of a digital image

• US 20030218679A1
• Filed: 12/18/2002
• Published: 11/27/2003
• Est. Priority Date: 12/24/2001
• Status: Active Grant

• Current Assignee: STMicroelectronics SRL (STMicroelectronics NV)
• Original Assignee: STMicroelectronics SRL (STMicroelectronics NV)
• Inventors: Castorina, Alfio, Battiato, Sebastiano, Mancuso, Massimo
• Granted Patent: US 7,227,991 B2
• Time in Patent Office: Days
• US Class Current: 348/272
• CPC Class Codes: G06T 2207/10024 Color image
• 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

A method for improving the quality of a digital image acquired with a non-optimal exposure. In one embodiment, the method, which is applied directly to the image in CFA (Color Filter Array) format, identifies the regions of the image that are most important from the perceptive or contextual point of view by means of simple statistical measures. The image is then transformed by inverting the response function in such a manner as to optimize the quality of these regions.

Claims

1. A method for improving the quality of a digital image, the digital image comprising a matrix P of digital values P(x,y), or pixels, each of which is associated on the basis of the position (x,y) with a chromatic component of the red, green or blue type, each of said digital values forming part of a range of possible digital values comprised between a maximum and a minimum number, said method comprising the following operations:

establishing a response function f that expresses a relationship between the digital values P(x,y) of the image P and a measure q(x,y) of the quantity of light that produced said digital values,

extracting from the image P the matrix G of the pixels associated with the green chromatic component obtained from P by discarding the pixels associated with the red and the blue component,

subdividing P into a plurality of adjacent and not overlapping regions,

selecting the visually important regions of G,

establishing a digital value Mean_v substantially at the center of the range of the possible digital values,

modifying the image P to obtain a corrected image P^T by performing a translation of the digital values of P by means of the inverting the response function f in such a manner as to make the mean of the digital values of G forming part of the important regions coincide with Mean^v,

characterized in that the important regions of G are selected by performing the following sequence of operations;

establishing a digital value Mean_v substantially at the center of the range of the possible digital values,

obtaining from the plane G a plane G^T by performing a translation of the digital values of G by inverting the response function f in such a manner as to make the mean of the pixels of G substantially coincide with Mean,

subdividing the plane G^T into regions corresponding to the regions of G,

assigning to every region of G an index of visual importance V calculated on the basis of at least one contrast measure D and at least one focus measure F of the pixels of the corresponding region G^T,

establishing a threshold value V_t, and

selecting the regions of G whose index of visual importance V is greater than V_t.

2. A method in accordance with claim 1, wherein the digital image P(x,y) is a CFA image in Bayer format.

3. A method in accordance with claim 1 wherein the planes G and G^T are subdivided into square matrices of equal size.

4. A method in accordance with claim 1 wherein the contrast measure D for a given region is the deviation of the histogram of the distribution of the digital values of G^T forming part of said region.

5. A method in accordance with claim 1 wherein the focus measure F for a given region is obtained by filtering the digital values of G^T forming part of said region with a digital high-pass filter and determining the mean of said filtered digital values.

6. A method in accordance with claim 5, wherein the digital high-pass filtering is obtained by performing the convolution of the region with a 3×

3 Laplace filter.

7. A method in accordance with claim 5 wherein the digital values that exceed a certain threshold after filtering are excluded from the mean.

8. A method in accordance with claim 1 wherein the measures F and D are normalized in the range [0;

1] and the index V is calculated as;

V=α

×

D+(1−

α

)×

F

where α

is any real number between 0 and 1.

9. A method in accordance with claim 1 wherein the index V is calculated by including an additional coefficient for the regions that occupy central positions of the digital image.

10. A camera for generating a digital image of a scene, the camera comprising:

a sensor array operable to acquire pixels of the image having a plurality of regions; and

a processor circuit operable to, calculate at least one respective characteristic value for each of the regions of the image, and to identify at least one of the regions as visually important based on the corresponding at least one characteristic value.

11. The camera of claim 10 wherein the sensor array is operable to acquire the image in a color-filter-array format.

12. The camera of claim 10 wherein the processor circuit is operable to calculate a respective contrast value and a respective focus value for each region of the image.

13. The camera of claim 10 wherein the processor circuit is operable to exclude predetermined pixels from the calculation of the at least one respective characteristic value for the region.

14. The camera of claim 10 wherein the processor circuit is operable to calculate a respective contrast value for each region of the image by:

calculating a histogram for the region;

calculating a mean of the histogram; and

calculating the contrast value equal to the deviation from the mean of the histogram.

15. The camera of claim 10 wherein the processor circuit is operable to calculate a respective focus value for each region of the image by:

high-pass filtering the pixels within the region; and

calculating the focus value equal to the mean of the filtered pixels.

16. The camera of claim 15 wherein the processor circuit is operable to high-pass filter the pixels by convoluting the region with a 3×

3 LaPlace filter.

17. The camera of claim 10 wherein the sensor array comprises filter elements arranged in a Bayer pattern.

18. The camera of claim 10 wherein the sensor array comprises a charge-coupled-device sensor.

19. The system of claim 10 wherein the sensor array comprises a CMOS-type sensor array.

20. A method for improving the quality of a digital image, the method comprising:

acquiring a digital image having a plurality of pixels;

determining a subset of visually important pixels from the pixels of the acquired image; and

modifying the plurality of pixels based on a characteristic of the visually important pixels.

21. The method of claim 20 wherein acquiring the digital image comprises acquiring a color filter array of the image.

22. The method of claim 20 wherein determining the visually important pixels comprises:

determining a contrast value and a focus value for each of a plurality of regions of the image;

identifying at least one visually important region based on the contrast and focus values of the at least one region; and

identifying pixels located within the at least one visually important region as the visually important pixels.

23. The method of claim 20 wherein modifying the plurality of pixels comprises:

calculating the mean of the visually important pixels;

determining a difference between the mean of the visually important pixels and a desired mean; and

shifting each of the plurality of pixels by the difference

24. A method, comprising:

extracting the green pixels from a color-format-array image having multiple regions to generate a green image plane having corresponding regions;

shifting the pixels of the green image plane such that the green pixels have a desired mean;

for each corresponding region of the green image plane,

calculating a histogram,

calculating the mean of the histogram,

calculating a contrast value D equal to the deviation from the mean of the histogram;

high-pass filtering the pixels in the region, and

calculating a focus value F equal to the mean of the filtered pixels;

normalizing the contrast values for the corresponding regions of the green plane;

normalizing the focus values for the corresponding regions of the green plane;

calculating a respective visual-importance value V=α

D+(1−

α

)F for each corresponding region of the green plane;

comparing each value V to a threshold to identify at least one visually important region of the image;

calculating the amount of shift needed to cause the green pixels of the at least one visually important region to have a desired mean; and

shifting the pixels of the image by the calculated amount.