Adaptive transforms

Abstract

In one embodiment, an image is broken up into multiple regions or segments, where each segment may be of arbitrary shape, and a transform (multi-scale or otherwise) is applied on the set of segments. In another embodiment, pattern adaptive prediction is used when predicting the next finer level of the transform pyramid. The pattern adaptive prediction uses the parent grid to determine what geometry of a filter is to be used when predicting the child grid. At the boundaries of the domain, the pattern adaptive prediction can coupled with the domain adaptive prediction technique.

1 - 3 . (canceled) 
     
     
         4 . A transform coder device for processing a multi-dimensional signal, the transform coder being configured to perform at least the steps of:
 obtaining the multi-dimensional signal comprising image data for a digital image frame;   applying an inverse transform to the multi-dimensional signal, the inverse transform interpolating parent data on a coarser scale to predict child data on a finer scale;   using an interpolation filter with data-dependent filter coefficients in performing said interpolating; and   adapting the interpolation filter to a pattern in the parent data.   
     
     
         5 - 42 . (canceled) 
     
     
         43 . The transform coder device of  claim 4 , wherein the interpolation filter comprises an array of filter coefficients. 
     
     
         44 . The transform coder device of  claim 43 , wherein the array of filter coefficients comprises a 4×4 array. 
     
     
         45 . The transform coder device of  claim 43 , wherein each filter coefficient comprises an initial fixed coefficient that is scaled by an inverse of a gradient value. 
     
     
         46 . The transform coder device of  claim 45 , wherein the gradient value is computed from a center to a value in the parent data. 
     
     
         47 . The transform coder device of  claim 4 , wherein the pattern comprises a trough of low points, and wherein the gradient values in a direction along the trough are small, while the gradient values in a direction perpendicular to the trough are high. 
     
     
         48 . The transform coder device of  claim 4 , wherein the pattern comprises a slanted surface, and said interpolating is along constant contour lines of the slanted surface. 
     
     
         49 . The transform coder device of  claim 4 , wherein the multi-dimensional signal comprises a two-dimensional image frame. 
     
     
         50 . An electronically-implemented method of transforming image data, the method comprising:
 obtaining the multi-dimensional signal comprising image data for a digital image frame;   applying an inverse transform to the multi-dimensional signal, the inverse transform interpolating parent data on a coarser scale to predict child data on a finer scale;   using an interpolation filter with data-dependent filter coefficients in performing said interpolating; and   adapting the interpolation filter to a pattern in the parent data.   
     
     
         51 . The electronically-implemented method of  claim 50 , wherein the interpolation filter comprises an array of filter coefficients. 
     
     
         52 . The electronically-implemented method of  claim 51 , wherein the array of filter coefficients comprises a 4×4 array. 
     
     
         53 . The electronically-implemented method of  claim 51 , wherein each filter coefficient comprises an initial fixed coefficient that is scaled by an inverse of a gradient value. 
     
     
         54 . The electronically-implemented method of  claim 53 , wherein the gradient value is computed from a center to a value in the parent data. 
     
     
         55 . The electronically-implemented method of  claim 50 , wherein the pattern comprises a trough of low points, and wherein the gradient values in a direction along the trough are small, while the gradient values in a direction perpendicular to the trough are high. 
     
     
         56 . The electronically-implemented method of  claim 50 , wherein the pattern comprises a slanted surface, and said interpolating is along constant contour lines of the slanted surface. 
     
     
         57 . The electronically-implemented method of  claim 50 , wherein the multi-dimensional signal comprises a two-dimensional image frame. 
     
     
         58 . A computer-readable non-transitory tangible medium comprising:
 computer-readable code configured to obtain the multi-dimensional signal comprising image data for a digital image frame;   computer-readable code configured to apply an inverse transform to the multi-dimensional signal, the inverse transform interpolating parent data on a coarser scale to predict child data on a finer scale;   computer-readable code configured to use an interpolation filter with data-dependent filter coefficients in performing said interpolating; and   computer-readable code configured to adapt the interpolation filter to a pattern in the parent data.   
     
     
         59 . The computer-readable non-transitory tangible medium of  claim 58 , wherein the interpolation filter comprises an array of filter coefficients. 
     
     
         60 . The computer-readable non-transitory tangible medium of  claim 59 , wherein the array of filter coefficients comprises a 4×4 array. 
     
     
         61 . The computer-readable non-transitory tangible medium of  claim 59 , wherein each filter coefficient comprises an initial fixed coefficient that is scaled by an inverse of a gradient value.