Scene adaptive brightness/contrast enhancement

1. A method for brightness and contrast enhancement (BCE) of a digital image, the method comprising:

computing a luminance histogram of the digital image;

computing first distances from the luminance histogram to a plurality of predetermined luminance histograms;

estimating first control point values for a global tone mapping curve for the digital image from first predetermined control point values corresponding to a subset of the predetermined luminance histograms selected based on the computed first distances, wherein the predetermined luminance histograms and the first predetermined control points are predetermined by:

computing luminance histograms for each training image in a plurality of training images;

selecting a subset of the training images in the plurality of training images as a training prototype set based on distances from each luminance histogram to luminance histograms of training images in the training prototype set; and

tuning each training image in the training prototype set based on a tuning tone mapping curve to generate the first predetermined control point values,

wherein the luminance histograms of the training images in the training prototype set are the predetermined luminance histograms; and

wherein the predetermined luminance histograms, the first predetermined control points, and the second predetermined control points are predetermined by:

computing luminance histograms for each training image in a plurality of training images;

selecting a subset of the training images in the plurality of training images as a first training prototype set based on distances from each luminance histogram to luminance histograms of training images in the prototype set;

tuning each training image in the first training prototype set based on a first tuning tone mapping curve to generate the first predetermined control point values,

wherein the luminance histograms of the training images in the first training prototype set are the predetermined luminance histograms;

enhancing each training image in the first training prototype set to generate enhanced training images, the enhancing comprising:

computing third distances from the luminance histogram of a training image to the luminance histograms of all other training images in the first training prototype set;

estimating third control point values for a tone mapping curve for the training image from first predetermined control point values corresponding to a subset of the luminance histograms of the other training images, the subset selected based on the computed third distances;

interpolating the estimated third control point values to determine the tone mapping curve for the training image; and

applying the tone mapping curve to the training image to generate an enhanced training image;

dividing each enhanced training image into a plurality of image blocks to generate a second training prototype set of image blocks;

computing a mean luminance value and a mean absolute luminance deviation value for each image block in the second training prototype set; and

tuning each image block in the second training prototype set based on a second tuning tone mapping curve to generate the second predetermined control point values;

interpolating the estimated control point values to determine the global tone mapping curve.

2. The method of claim 1 , further comprising applying the global tone mapping curve to the digital image to generate an enhanced digital image.

3. The method of claim 1 , wherein the first predetermined control point values comprise a vector of control point values for each predetermined luminance histogram, and wherein estimating first control point values comprises computing weighted averages of respective control point values in the vectors of control point values corresponding to the subset of the predetermined luminance histograms.

4. The method of claim 1 , further comprising applying principal component analysis to the luminance histograms of training images in the training prototype set.

5. The method of claim 1 , wherein tuning each training image is performed by a plurality of users and the predetermined control points values are generated by computing an average of control point values selected by the plurality of users.

6. The method of claim 1 , further comprising:

dividing the digital image into a plurality of image blocks; and

enhancing pixels in the digital image to generate a final digital image, the enhancing comprising:

computing second distances from a pixel to the centers of a predetermined number of neighboring image blocks of an image block comprising the pixel;

computing an enhanced pixel value for the pixel based on the computed second distances, second predetermined control point values corresponding to the neighboring image blocks, and the global tone mapping curve; and

storing the enhanced pixel value in the final digital image.

7. The method of claim 6 , further comprising:

locating the second predetermined control point values corresponding to each image block of the neighboring image blocks using a mean luminance value and a mean absolute luminance deviation of the image block.

8. The method of claim 6 , wherein computing an enhanced pixel value comprises:

generating a local tone mapping curve for each of the neighboring image blocks based on the second predetermined control point values;

merging each local tone mapping curve with the global tone mapping curve to generate a merged tone mapping curve for each of the neighboring image blocks;

applying each merged tone mapping curve to the pixel to generate tone mapped pixel values; and

computing the enhanced pixel value as a weighted average of the tone mapped pixel values, wherein weights used in the weighted average are based on the second distances.

9. The method of claim 6 , further comprising applying principal component analysis to the luminance histograms of training images in the training prototype set.

10. The method of claim 6 , wherein turning each training image is performed by a plurality of users and the predetermined first control points values are generated by computing an average of control point values for each training image selected by the plurality of users, and wherein turning each image block is performed by the plurality of users and the predetermined second control points values are generated by computing an average of control point values for each image block selected by the plurality of users.

11. The method of claim 6 , wherein the second predetermined control point values are stored based on the mean luminance values and the mean absolute luminance deviations of the image blocks.