Super-resolution image reconstruction using high-frequency band extraction
Generating a super-resolved reconstructed image includes acquiring a multitude of monochromatic images of a scene and extracting high-frequency band luma components from the acquired images. A high-resolution luma image is generated using the high-frequency components and motion data for the acquired images. The high-resolution luma image is combined with an up-sampled color image, generated from the acquired images, to generate a super-resolved reconstructed color image of the scene.
1. A method of generating a super-resolved reconstructed image, the method comprising:
acquiring a plurality of monochromatic images of a scene;
extracting high-frequency band luma components from the acquired images;
generating a high-resolution luma image using the high-frequency band luma components and motion data for the acquired images; and
replacing luma data in an up-sampled color image, generated from the acquired monochromatic images, with high-resolution luma data in the high-resolution luma image to obtain a super-resolved reconstructed color image.
2. The method of claim 1 wherein extracting the high-frequency band luma components includes:
convolving raw monochromatic image data of the plurality of monochromatic images of the scene using a Gaussian filter to obtain a Gaussian filtered image; and
subtracting the raw monochromatic image data from the Gaussian filtered image.
3. The method of claim 2 including constructing an array of raw monochromatic images, including the raw monochromatic image data, based on the acquired images.
4. The method of claim 2 wherein extracting the high-frequency band luma components further includes eliminating polarity differences after subtracting the raw monochromatic image data from the Gaussian filtered image.
5. The method of claim 4 wherein eliminating polarity differences includes calculating an absolute value of a difference previously obtained by subtracting the raw monochromatic image data from the Gaussian filtered image.
6. The method of claim 4 wherein extracting the high-frequency band luma components further includes, after eliminating the polarity differences, equalizing magnitudes of intensity of the high-frequency band luma components.
7. The method of claim 4 wherein extracting the high-frequency band luma components further includes applying a sigmoid function after eliminating polarity differences.
8. The method of claim 2 including:
executing a motion estimation protocol based on the high-frequency band luma components and based on common channels in the raw monochromatic image data to obtain further motion data; and
executing a super-resolution protocol to obtain the high-resolution luma image based on the high frequency luma components and the further motion data in a linear transform.
9. The method of claim 1 wherein the extracted high-frequency band luma components correspond to at least one of edge locations or texture features.
10. The method of claim 1 including:
performing color image fusion based on raw monochromatic image data of the plurality of monochromatic images to obtain a low-resolution chroma image; and
up-sampling the low-resolution chroma image to obtain the up-sampled color image.
11. The method of claim 1 including visually presenting the super-resolved reconstructed color image on a display device.
12. The method of claim 1 including using the super-resolved reconstructed color image to engage a safety feature of a machine.