IPC분류정보
국가/구분 |
United States(US) Patent
등록
|
국제특허분류(IPC7판) |
|
출원번호 |
US-0922328
(2004-08-20)
|
등록번호 |
US-7379626
(2008-05-27)
|
발명자
/ 주소 |
- Lachine,Vladimir
- Smith,Gregory Lionel
- Lee,Louie
|
출원인 / 주소 |
|
대리인 / 주소 |
|
인용정보 |
피인용 횟수 :
31 인용 특허 :
10 |
초록
▼
An edge adaptive system and method for image filtering. The method maps each output pixel onto input image coordinates and then prefilters and resamples the input image pixels around this point to reduce noise and adjust the scale corresponding to a particular operation. Then the edge in the input
An edge adaptive system and method for image filtering. The method maps each output pixel onto input image coordinates and then prefilters and resamples the input image pixels around this point to reduce noise and adjust the scale corresponding to a particular operation. Then the edge in the input image is detected based on local and average signal variances in the input pixels. According to the edge detection parameters, including orientation, anisotropy and variance strength, the method determines a footprint and frequency response for the interpolation of the output pixel. In a more particular implementation, the method divides the input pixel space into a finite number of directions called skews, and estimates the edge orientation with the nearest skew direction. This further facilitates pixels inclusion in the interpolation of the output pixel.
대표청구항
▼
The invention claimed is: 1. A method for expanding and enhancing input digital image data having an input coordinate space, and producing output digital image data having an output coordinate space with certain resolution and shape, by applying adaptive edge interpolation to an output pixel, said
The invention claimed is: 1. A method for expanding and enhancing input digital image data having an input coordinate space, and producing output digital image data having an output coordinate space with certain resolution and shape, by applying adaptive edge interpolation to an output pixel, said method comprising: (a) obtaining the input digital image data; (b) mapping the output pixel coordinates onto the input coordinate space obtained in (a); (c) selecting a block of M��N input pixels around the mapped point of (b) wherein the shape and size of said block is determined based on a required edge resolution; (d) applying horizontal and vertical lowpass filters to the input pixels of said block to reduce noise, and downsample to change the scale to adjust the aspect ratio, and to adapt to low angle edge detection; (e) computing local signal variance for each downsampled pixel of (d) in said block to determine signal behavior; (f) computing average signal variance, based on local signal variances determined in (e) in said block, to determine average signal behavior in the block; (g) determining orientation, anisotropy, and strength of an edge based on the average signal variance found in (f); (h) determining frequency response of resampling filter and its footprint in the input image at the mapped position determined in (b), based on orientation, anisotropy, and strength of the edge determined in (g); and (i) determining the value of the output pixel via resampling in the footprint found in (h) according to the frequency response of (h). 2. The method of claim 1, wherein the average signal variance of (f) is determined via: I. computing horizontal and vertical gradients in each input pixel of the M��N block; II. computing a local gradient squared tensor for each input pixel in said block based on the local gradients of "I"; and III. computing an average gradient squared tensor for the block based on the local gradient tensors of "II". 3. The method of claim 1, wherein the edge orientation is estimated via: (A) predefining a set of slopes in the x-y plane that have the same intersection distance from the nearest pixels in each input pixel row; (B) characterizing these slopes with different skew line, wherein a skew line is defined as having a preset slope of dx/dy, and (C) approximating the actual edge orientation with the nearest skew line, to facilitated the computation of the interpolating of the output pixel by eliminating the need for computation of the intersection of the edge orientation line with an input pixel row. 4. The method of claim 1, wherein at (h) an elliptical footprint is determined based on the edge orientation with one of its major and minor axes along the edge orientation, and wherein the method uses a single pass, two-dimensional resampling filter based on an elliptical frequency response. 5. The method of claim 3, wherein an elliptical footprint is determined with one of its major and minor axes along the skew line to facilitate the computation of pixel inclusion in the filtering, and wherein the method uses a single pass, two-dimensional resampling filter based on an elliptical frequency response. 6. The method of claim 1, wherein a parallelogram footprint is determined along the edge orientation, and wherein the method uses two-pass horizontal and vertical resampling based on a parallelogram frequency response. 7. The method of claim 3, wherein a parallelogram footprint is determined along the skew direction with the skew line parting the parallelogram in half to facilitate the computation of pixel inclusion for filtering, and wherein the method uses two-pass horizontal and vertical resampling based on a parallelogram frequency response. 8. The method of claim 1, applied to convert an interlaced video input to a progressive scan video output, wherein the method increases vertical scale and interpolates missing lines in a field to produce a progressive scan video frame. 9. The method of claim 1, applied to convert an interlaced video input to a progressive scan video output via: (I) detecting motion strength in pixels of the interlaced video fields; (II) generating the progressive scan frame by merging pixels from two adjacent interlaced fields when the detected motion strength is lower than a predetermined lower limit; (III) generating the progressive scan frame by resealing and interpolating a single interlaced field, using the method of claim 8, when the detected motion strength is higher than a predetermined upper limit; and (IV) interpolating between (II) and (III) according to the motion strength, when the motion strength lies between the lower and the upper limits. 10. The method of claim 1, wherein the method uses intermediate horizontal resampling of the output pixels from the surrounding input pixels based on shifted linear interpolation, determined by shifting the value of each pixel by a certain amount along the line connecting the values of two adjacent pixels and sampling these new values instead of the original pixel values to mitigate the high frequency loss. 11. The method of claim 1, wherein the coordinates mapping in (b) is a linear scaling. 12. The method of claim 1, wherein the coordinates mapping in (b) is warping. 13. The method of claim 1, wherein the frequency response of the filter in (i) is adaptive to the strength and anisotropy of the edge found in (g), to enhance the edge details if the edge anisotropy and strength are higher than a preset limit, and to smooth the edge details if the edge anisotropy and strength are lower than a preset limit. 14. A system for expanding and enhancing input digital image data having an input coordinate space, and producing output digital image data having an output coordinate space with certain resolution and shape, by applying adaptive edge interpolation to an output pixel, said system comprising: (a) an interface to obtain input digital image data; (b) a coordinates generator to map the output pixel coordinates onto the input coordinate space; (c) a first footprint generator, coupled to said interface and said coordinates generator, to select a block of M��N input pixels around the mapped point, wherein the shape and size of the block is determined based on the a required edge resolution; (d) a preprocessor, coupled to said first footprint generator, to apply horizontal and vertical lowpass filters to the input pixels of said block to reduce noise, and downsample to change the scale to adjust the aspect ratio, and to adapt to low angle edge detection; (e) a variance calculator, coupled to said preprocessor, to compute local signal variance for each input pixel in said block to determine signal behavior; (f) an integrator, coupled to said variance calculator, to compute average signal variance, based on local signal variances in said block, to determine average signal behavior in the block; (g) an edge estimator, coupled to said integrator, to determine orientation, anisotropy, and strength of an edge, based on the average signal variance found by said integrator; (h) a second footprint generator, coupled to said interface, said coordinates generator, and said edge estimator, to determine frequency response of a resampling filter and its footprint in the input image at the mapped position, based on orientation, anisotropy and strength of the edge determined by said edge estimator; and (i) a filter, coupled to said second footprint generator and said edge estimator, to determine the value of the output pixel via resampling in the footprint and according to the frequency response found by said second footprint generator. 15. The system of claim 14, wherein said variance calculator and said integrator are adapted to compute the average signal variance via: I. computing horizontal and vertical gradients in each input pixel of the M��N block; II. computing a local gradient squared tensor for each input pixel in said block based on the local gradients of "I". III. computing an average gradient squared tensor for the block based on the local gradient tensors of "II". 16. The system of claim 14, wherein said edge estimator is adapted to perform an edge orientation estimation via: (A) predefining a set of slopes in the x-y plane that have the same intersection distance from the nearest pixels in each input pixel row; (B) characterizing these slopes with different skew line, wherein a skew line is defined as having a preset slope of dx/dy; (C) approximating the actual edge orientation with the nearest skew line, to facilitated the computation of the interpolating of the output pixel by eliminating the need for computation of the intersection of the edge orientation line with an input pixel row. 17. The system of claim 14, wherein said second footprint generator is adapted to determine an elliptical footprint based on the edge orientation with one of its major and minor axes along the edge orientation, and wherein said filter is adapted to use a single pass, two-dimensional resampling based on an elliptical frequency response function. 18. The system of claim 16, wherein said second footprint generator is adapted to determine an elliptical footprint with one of its major and minor axes along the skew line to facilitate the computation of pixel inclusion in the filtering, and wherein said filter uses a single pass, two-dimensional resampling based on an elliptical frequency response function. 19. The system of claim 14, wherein said second footprint generator is adapted to determine a parallelogram footprint along the edge orientation, and wherein said filter is adapted to use two-pass horizontal and vertical resampling based on a parallelogram frequency response. 20. The system of claim 16, wherein said second footprint generator is adapted to determine a parallelogram footprint along the skew direction with the skew line parting the parallelogram in half to facilitate the computation of pixel inclusion for filtering, and wherein said filter is adapted to use two-pass horizontal and vertical resampling based on a parallelogram frequency response. 21. The system of claim 14, adapted to convert an interlaced video input to a progressive scan video output, wherein the system increases vertical scale and interpolates missing lines in a field to produce a progressive scan video frame. 22. The system of claim 14, adapted to convert an interlaced video input to a progressive scan video output via: (I) detecting motion strength in pixels of the interlaced video fields; (II) generating the progressive scan frame by merging pixels from two adjacent interlaced fields when the detected motion strength is lower than a predetermined lower limit; (III) generating the progressive scan frame by resealing and interpolating a single interlaced field, using the method of claim 8, when the detected motion strength is higher than a predetermined upper limit; (IV) interpolating between (II) and (III) according to the motion strength, when the motion strength lies between the lower and the upper limits. 23. The system of claim 14, wherein the system is adapted to perform intermediate horizontal resampling of the output pixels from the surrounding input pixels based on shifted linear interpolation, determined by shifting the value of each pixel by a certain amount along the line connecting the values of two adjacent pixels and sampling these new values instead of the original pixel values to mitigate the high frequency loss. 24. The system of claim 14, wherein said coordinate generator is adapted to perform coordinate mapping as a linear scaling. 25. The system of claim 14, wherein said coordinate generator is adapted to perform coordinate mapping as warping. 26. The system of claim 14, wherein said filter is adaptive to the strength and anisotropy of the edge, to enhance the edge details if the edge anisotropy and strength are higher than a preset limit, and to smooth the edge details if the edge anisotropy and strength are lower than a preset limit.
※ AI-Helper는 부적절한 답변을 할 수 있습니다.