초록 ▼

Input data comprising a video signal is processed using a combination of a known image processing method to deblur, or sharpen, the image and convolution with Pixon짰 kernels for smoothing. The smoothing process utilizes a plurality of different size Pixon짰 kernels which operate in parallel so that t

Input data comprising a video signal is processed using a combination of a known image processing method to deblur, or sharpen, the image and convolution with Pixon짰 kernels for smoothing. The smoothing process utilizes a plurality of different size Pixon짰 kernels which operate in parallel so that the input data are convolved with each different Pixon짰 kernel simultaneously. The smoothed image is convolved with the point response function (PRF) to form data models that are compared against the input data, then the broadest Pixon짰 kernel that fits the input data within a predetermined criterion are selected to form a Pixon짰 map. The data are smoothed and assembled according to the Pixon짰 map, then are deconvolved and output to a video display or other appropriate device, providing a clearer image with less noise.

대표청구항 ▼

We claim: 1. A method for high speed enhancement of a video image comprising: inputting a signal containing input data corresponding to a frame of the video image, the input data comprising a plurality of data points on a grid; convolving the input data with each kernel of a plurality of different

We claim: 1. A method for high speed enhancement of a video image comprising: inputting a signal containing input data corresponding to a frame of the video image, the input data comprising a plurality of data points on a grid; convolving the input data with each kernel of a plurality of different sized kernels to generate a plurality of convolved data sets; determining a change in data between the input data and each convolved data set of the plurality; comparing the change to a fit criterion; selecting the broadest kernel of the plurality of different sized kernels that satisfies the fit criterion for each portion of the plurality of data points; identifying indices within the grid for the selected kernels corresponding to each portion of data points and assembling a kernel map using the indices; assembling the convolved data according to the kernel map to produce smoothed data; deblurring the smoothed data for form a reconstructed image; and outputting the reconstructed image to a video display device. 2. The method of claim 1, wherein the input signal comprises a luminance component of a standard NTSC video stream and the reconstructed image is an enhanced luminance signal. 3. The method of claim 2, wherein, prior to outputting the image to a video display device, the enhanced luminance signal is merged with a chrominance component of the standard NTSC video stream. 4. The method of claim 3, wherein the chrominance component is synchronized with the enhanced luminance signal. 5. The method of claim 2, wherein the NTSC video stream is de-interlaced. 6. The method of claim 5, wherein the image is interlaced prior to output to a video display device. 7. The method of claim 1, wherein the step of deblurring comprises convolving the smoothed image with an approximate inverse point response function. 8. The method of claim 7, wherein the approximate inverse point response function is obtained by minimizing an optimization function of the form: where δi0 is a residual point response function, Hi-j is a point response function, fj is the approximate inverse point response function, λ is an adjustable parameter that sets a tradeoff between resolution gain and noise amplification, and μ is a Lagrange multiplier. 9. The method of claim 1, wherein the change is determined according to the relationship fj =|{tilde over (H)}{circle around (x)}Dj |, where {tilde over (H)} is a broadened point response function (PRF), Dj is the convolved data that was convolved by kernel j, and {circle around (x)} is the convolution operator. 10. The method of claim 9, wherein the fit criterion is f j≦α쨌σ, where α is a variable kernel map smoothness control and σ is the standard deviation of noise in the input image. 11. The method of claim 1, wherein the step of convolving the input data is performed in parallel by all kernels of the plurality of kernels. 12. A method for image reconstruction comprising: (a) receiving input data corresponding to at least one input image, the input data comprising a plurality of data points on a grid; (b) smoothing the at least one input image according to a kernel map comprising selected kernels from a plurality of different kernels, wherein the selected kernels are selected by: (i) convolving the input data with each kernel of the plurality of kernels, wherein each kernel has a different size for encompassing different portions of the plurality of data points and generating convolved data corresponding to each kernel; (ii) for each kernel, determining a change from the input data to the convolved data; (iii) comparing the change to a fit criterion; (iv) selecting the broadest kernel of the plurality of kernels that satisfies the fit criterion for a portion of the plurality of data points; (v) identifying indices within the grid for the selected kernels corresponding to the portion of data points and assembling the kernel map using the indices; (c) before or after the step of smoothing, deblurring the image to form a deblurred image, wherein, if deblurring occurs after the step of smoothing, the smoothed image is deblurred; and if deblurring occurs before the step of smoothing, the deblurred image is smoothed; and (d) outputting a reconstructed image comprising the deblurred and smoothed image to a display device. 13. The method of claim 12, wherein the plurality of kernels comprises 5 to 7 different kernel sizes. 14. The method of claim 12, wherein the step of deblurring comprises convolving the smoothed image with an approximate inverse point response function (PRF). 15. The method of claim 14, wherein the approximate inverse point response function (PRF) is a minimized optimization function which is minimized by minimizing the difference between a deconvolved PRF and a residual PRF, minimizing noise amplification, and normalizing the optimized small kernel deconvolver. 16. The method of claim 12, wherein the step of deblurring comprises convolving the input image with an approximate inverse point response function (PRF). 17. The method of claim 12, wherein the at least one input image comprising at least one video frame. 18. The method of claim 17, wherein the at least one input image comprises a plurality of image frames, and further comprising coadding the data from each image frame to produce the input data. 19. The method of claim 17, wherein the input data comprises a luminance component of a video image frame of a standard NTSC video stream and further comprising, before the step of receiving, dividing the video image frame into the luminance component and a chrominance component. 20. The method of claim 19, wherein the reconstructed image is an enhanced luminance signal. 21. The method of claim 17, wherein the display device is a video display device and further comprising, prior to outputting the image to the video display device, merging the enhanced luminance signal with the chrominance component. 22. The method of claim 21, further comprising synchronizing the chrominance component with the enhanced luminance signal prior to the step of merging. 23. The method of claim 12, wherein steps (b)(i) and (b)(ii) are performed in parallel by all kernels of the plurality of different kernels. 24. A system for high speed enhancement of a video image comprising: an input for inputting a signal containing input data corresponding to a frame of the video image, the input data comprising a plurality of data points on a grid; a processor for convolving the input data with each kernel of a plurality of different sized kernels, assembling a kernel map by selecting the broadest kernel of the plurality of different sized kernels that satisfies a fit criterion for each portion of the plurality of data points, and smoothing the input data in accordance with the kernel map to generate smoothed data; an image processing function within the processor for deblurring the smoothed data for form a reconstructed image; and a video display device for receiving and displaying the reconstructed image. 25. The system of claim 24, wherein the input signal comprises a luminance component of a standard NTSC video stream and the reconstructed image is an enhanced luminance signal. 26. The system of claim 25, further comprising a switch for combining the enhanced luminance signal with a chrominance component of the standard NTSC video stream. 27. The system of claim 26, further comprising a delay element for synchronizing the chrominance component with the enhanced luminance signal. 28. The system of claim 26, further comprising a de-interlacer for de-interlacing the NTSC video stream prior to processing the luminance component. 29. The system of claim 28, further comprising an interlacer for interlacing the combined enhanced luminance signal and the chrominance component prior to output to the video display device. 30. The system of claim 24, wherein the image processing function comprises an approximate inverse point response function. 31. The system of claim 30, wherein the approximate inverse point response function is obtained by minimizing an optimization function of the form: where δi0 is a residual point response function, Hi-j is a point response function, fj is the approximate inverse point response function, λ is an adjustable parameter that sets a tradeoff between resolution gain and noise amplification, and μ is a Lagrange multiplier. 32. The system of claim 24, further comprising a user interface for inputting selections for standard deviation and smoothness of the kernels. 33. The system of claim 24, wherein the processor convolves the input data in parallel, simultaneously using all kernels of the plurality of different kernels.