Hierarchical sharpness evaluation
Techniques are disclosed for estimating quality of images in an automated fashion. According to these techniques, a source image may be downsampled to generate at least two downsampled images at different levels of downsampling. Blurriness of the images may be estimated starting with a most-heavily downsampled image. Blocks of a given image may be evaluated for blurriness and, when a block of a given image is estimated to be blurry, the block of the image and co-located blocks of higher resolution image(s) may be designated as blurry. Thereafter, a blurriness score may be calculated for the source image from the number of blocks of the source image designated as blurry.
1. A method of estimating blurriness of a source image, captured by a camera or generated by a computer, comprising:
downsampling the source image to generate at least two downsampled images at different levels of downsampling;
parsing each downsampled images into a plurality of blocks;
starting with a most-heavily downsampled image, estimating blurriness of the downsampled images and the source image by:
estimating blurriness of each block of a current image,
when a block of a current image is estimated to be blurry:
designating the block of the current image as blurry,
designating co-located blocks of higher resolution image(s) as blurry; and
computing a blurriness score for the source image from a number of blocks of the source image designated as blurry.
2. The method of claim 1 further comprising, classifying the source image as a key image from among a plurality of other images based on the source image's blurriness score.
3. The method of claim 1 further comprising:
comparing the blurriness score to a threshold and
classifying the source image as one of blurry and not blurry based on the comparison.
4. The method of claim 1 further comprising:
repeating the method for a plurality of source images, and
estimating a least blurry source image based on respective blurriness scores of the plurality of source images.
5. The method of claim 4 , wherein the plurality of source images are members of a common video sequence.
6. The method of claim 4 , wherein the plurality of source images are selected by a user of a device that performs the method.
7. The method of claim 1 , wherein the source image is retrieved from a memory device.
8. Non-transitory computer readable medium storing program instructions that, when executed by a processing device, causes the device to:
downsample a source image, captured by a camera or generated by a computer, to generate at least two downsampled images at different levels of downsampling;
parse each downsampled image into a plurality of blocks;
starting with a most-heavily downsampled image, estimate blurriness of the downsampled images and the source image by:
estimating blurriness of each block of a current image,
when a block of a current image is estimated to be blurry:
designating the block of the current image as blurry,
designating co-located blocks of higher resolution image(s) as blurry; and
compute a blurriness score for the source image from a number of blocks of the source image designated as blurry.
9. The medium of claim 8 , further comprising, classifying the source image as a key image from among a plurality of other images based on the source image's blurriness score.
10. The medium of claim 8 , further comprising:
comparing the blurriness score to a threshold and
classifying the source image as one of blurry and not blurry based on the comparison.
11. The medium of claim 8 , further comprising:
repeating the method for a plurality of source images, and
estimating a least blurry source image based on respective blurriness scores of the plurality of source images.
12. The medium of claim 8 , wherein the source image is retrieved from the computer readable medium.
13. Apparatus, comprising:
a processor and a memory,
the processor configured by program instructions stored in the memory to operate as a downsampler, and a sharpness analyzer,
the downsampler generating at least a pair of downsampled images from a source image, captured by a camera or generated by a computer, at two downsampled resolutions,
the sharpness analyzer estimating blurriness of the downsampled images and the source image by estimating blurriness of each block of a current image and, when a block of a current image is estimated to be blurry, designating the block of the current image and co-located blocks of higher resolution image(s) as blurry, and compute a blurriness score for the source image from a number of blocks of the source image designated as blurry.
14. The apparatus of claim 13 , wherein the sharpness analyzer is a neural network.
15. The apparatus of claim 14 , wherein the neural network is trained from training images blended with blur kernels.
16. The apparatus of claim 14 , wherein the neural network is trained from training images blended with simulated sensor noise.
17. The apparatus of claim 13 , wherein the processor classifies the source image as a key image from among a plurality of other images based on the source image's blurriness score.
18. The apparatus of claim 13 , wherein the processor:
compares the blurriness score to a threshold and
classifies the source image as one of blurry and not blurry based on the comparison.
19. The apparatus of claim 13 , wherein the source image is part of a video sequence.
20. The apparatus of claim 13 , wherein the source image is selected by a user of the apparatus.
21. The apparatus of claim 13 , further comprising a camera having an output for the source image.
22. The apparatus of claim 13 , wherein the source image stored in the memory prior to processing by the processor.