초록 ▼

A method for enhancing at least one image within a series of images is provided. The method includes: selecting the series of images; upscaling each image within the series of images; selecting a reference image among the series of images; performing image registration to align series of images with

A method for enhancing at least one image within a series of images is provided. The method includes: selecting the series of images; upscaling each image within the series of images; selecting a reference image among the series of images; performing image registration to align series of images with the reference image; evaluating the series of aligned images for a subset of pixel locations that exhibit high cross-frame variation; performing learning processing to substantially reduce noise and exclude motion biases at the subset of pixel locations; performing pixel fusion from the series of aligned and processed images to produce the super-resolution reference image. A computer program product and an imaging system are disclosed.

대표청구항 ▼

1. A method for providing a super-resolution image within a series of images, the method comprising: selecting the series of images;upscaling each image within the series of images;selecting a reference image among the series of images;performing image registration to align the series of images with

1. A method for providing a super-resolution image within a series of images, the method comprising: selecting the series of images;upscaling each image within the series of images;selecting a reference image among the series of images;performing image registration to align the series of images with the reference image;evaluating the series of aligned images for a subset of pixel locations that exhibit high cross-frame variation;performing learning processing to substantially reduce noise and exclude motion biases at the identified subset of pixel locations;performing pixel fusion for the series of aligned and processed images to produce the super-resolution image. 2. The method as in claim 1, wherein the series of images comprises two or more sequential images. 3. The method as in claim 1, wherein image registration comprises correlating at least some of the pixels of the reference image with at least some of the pixels of another image. 4. The method as in claim 1, wherein image registration comprises comparing at least one property of the selected pixels from the reference image with at least one property of the selected pixels from the another image. 5. The method as in claim 4, wherein the at least one property of the selected pixels includes a scale-invariant feature (SIFT) transform. 6. The method as in claim 1, wherein performing learning processing comprises assigning a plurality of keypoints to each image in the series of images. 7. The method as in claim 1, wherein performing image registration comprises processing with the SIFT flow algorithm. 8. The method as in claim 1, wherein evaluating the series of aligned images for a subset of pixel locations that exhibit high cross-frame variation comprises determining a deviation of pixels aligned at each location, and comparing the result to a threshold value. 9. The method as in claim 8, wherein the deviation comprises a mean square distance to the median of one or more pixel channels. 10. The method as in claim 9, wherein the deviation comprises a mean square distance to median for each of the local descriptors. 11. The method as in claim 1, wherein the learning processing comprises performing unsupervised K-means processing. 12. The method as in claim 1, wherein performing image registration comprises processing an energy function for a scale-invariant feature transform (SIFT) that comprises: E⁡(w)=∑p⁢min⁡(s1⁡(p)-s2⁡(p+w⁡(p))1,t)+∑p⁢η⁢(u⁡(p)+v⁡(p))+∑(p,q)∈ɛ⁢min⁡(α⁢u⁡(p)-u⁡(q),d)+min⁡(α⁢v⁡(p)-v⁡(q),d), wherein p(x,y) represents the grid coordinate for the series of images; andw(p)=(u(p),v(p)) represents the flow vector at p; and, wherein u(p) and v(p) are integers and there are L possible states for u(p) and v(p), respectively; assuming that s1 and s2 are two dense SIFT images to be matched and set c contains all of the spatial neighborhoods in a four-neighbor system. 13. The method as in claim 1, wherein performing pixel fusion comprises at least one of median filtering and mean filtering across frames of images within the series of images. 14. A computer program product stored on non-transitory machine readable media, the computer program product comprising machine executable instructions for implementing a neural network by executing a method for producing a super-resolution reference image within a series of images, the method comprising: selecting the series of images;upscaling each image within the series of images;selecting a reference image among the series of images;performing image registration to align the series of images with the reference image;evaluating the series of aligned images for a subset of pixel locations that exhibit high cross-frame variation;performing learning processing to substantially reduce noise and exclude motion biases at the identified subset of pixel locations;performing pixel fusion from the series of aligned and processed images to produce the super-resolution reference image. 15. The computer program product as in claim 14, wherein the selecting comprises receiving the series of images through a communications network. 16. The computer program product as in claim 14, wherein executing the method is performed on data that is one of stored, streamed, and a live feed. 17. The computer program product as in claim 14, further configured to provide output to an intelligent system. 18. An imaging system, comprising: a device configured to provide a series of images;non-transitory machine readable media comprising machine executable instructions for producing a super-resolution reference image, the instructions stored thereon; anda processor configured to read the machine executable instructions and perform the image enhancement by:selecting the series of images;upscaling each image within the series of images;selecting a reference image among the series of images;performing image registration to align the series of images with the reference image;evaluating the series of aligned images for a subset of pixel locations that exhibit high cross-frame variation;performing learning processing to substantially reduce noise and exclude motion biases at the identified subset of pixel locations;performing pixel fusion from the series of aligned and processed images to produce the super-resolution reference image. 19. The imaging system as in claim 18, wherein the device configured to provide the series of images comprises one of a smart phone, a camera, a tablet computer, a scanner, and a specialized device. 20. The method as in claim 18, wherein output of the imaging system is configured for at least one of: digital zoom in, image stabilization, video surveillance, medical image analysis, and image recognition.