Low-resolution video coding content extraction

Abstract

Low complexity method embodiments directly decode low-resolution frames from compressed high-resolution videos that were encoded using predictive coding techniques like the H.264 video coding standard. The smaller the decoding resolution, the higher will be the computation and power savings of using the method. Low-frequency coefficients of 2D transformed predictions are added to the low-frequency coefficients of the transformed residual error. Low-frequency coefficients of the reconstructed data are then inverse transformed taking a smaller size transform. Further savings are obtained by reconstructing only those reference pixels that will be needed for accurate decoding of further Intra blocks.

What is claimed is: 
     
         1 - 7 . (canceled) 
     
     
         8 . A system comprising:
 a down-sampler operable to produce a first low-resolution frame from a selected high-resolution compressed frame, the first low-resolution frame corresponding to a first DC coefficient and a plurality of frequency coefficients; and   a frame reconstruction unit operable to produce a second low-resolution frame according to the plurality of frequency coefficients of the first low-resolution frame, the second low-resolution frame having a second DC coefficient, wherein the frame reconstruction unit is operable to produce a third low-resolution frame according to a residual error between the first low-resolution frame and the second low-resolution frame, a third DC coefficient for the third low-resolution frame being a weighted average of the first DC coefficient and the second DC coefficient.   
     
     
         9 . The system of  claim 8 , wherein the selected high-resolution compressed frame is a bi-predictive frame (B-frame) of a group of pictures (GOP) and is located immediately before an instantaneous decoding refresh (IDR frame). 
     
     
         10 . The system of  claim 8 , wherein the selected high-resolution compressed frame is a predictive frame (P-frame) of a group of pictures (GOP) and is located immediately before an instantaneous decoding refresh (IDR frame). 
     
     
         11 . The system of  claim 8 , wherein the frame reconstruction unit is operable to add low frequency coefficients of a 2D transformed prediction to low frequency coefficients of the residual error to produce low frequency coefficients of the third low-resolution frame. 
     
     
         12 . The system of  claim 8 , wherein the frame reconstruction unit is operable to produce only those reference pixels required for an accurate decoding of Intra blocks. 
     
     
         13 . A system comprising:
 a down-sampler operable to produce a first low-resolution frame from a selected high-resolution compressed frame, wherein the selected high-resolution compressed frame is an intra frame (I-frame) of a group of pictures (GOP); and   a frame reconstruction unit operable to:
 generate frequency coefficients of a prediction, wherein the number of frequency coefficients of the prediction is no more than one fourth of the frequency coefficients in the I-frame of the GOP; 
 generate frequency coefficients of a residual error; 
 reconstruct those pixels of a particular block that will be used as reference pixels for decoding future blocks; 
 take the inverse transform of a sum of the frequency coefficients of the prediction and the frequency coefficients of the residual error; and 
 generate a reference DC coefficient for a next frame as a weighted average of four reference DC coefficients. 
   
     
     
         14 . The system of  claim 13 , wherein the down-sampler is operable to sub-sample I-frames in a predictive coding video bitstream by:
 using an integer portion of a motion vector to point to a reference block selected in a reference frame;   generating a first set of frequency coefficients from the selected reference block, wherein the number of frequency coefficients in the first set is no more than one fourth of the frequency coefficients in the selected reference block;   generating a second set of frequency coefficients for a residual error;   adding the first set of frequency coefficients to the second set of frequency coefficients; and   taking an inverse transform of the sum of the frequency coefficients.   
     
     
         15 . The system of  claim 13 , wherein the low-resolution frame is used as a reference frame. 
     
     
         16 . A method for extracting a low-resolution video, comprising:
 inputting a video comprising a plurality of high-resolution frames, at least one of the plurality of high-resolution frames being a reference frame, the plurality of high-resolution frames being encoded with a predictive coding technique;   down-sampling the reference frame to produce a first low-resolution frame with a first DC coefficient;   generating a plurality of frequency coefficients corresponding to the first low-resolution frame;   generating a second low-resolution frame according to the plurality of frequency coefficients and the first low-resolution frame, the second low-resolution frame having a second DC coefficient;   generating a residual error between the first low-resolution frame and the second low-resolution frame;   generating a third low-resolution frame according to the plurality of frequency coefficients and the residual error; and   determining a third DC coefficient for the third low-resolution frame according to a weighted average of at least the first DC coefficient and the second DC coefficient.   
     
     
         17 . The method of  claim 16 , wherein the plurality of high-resolution frames comprises a compressed predictive frame that immediately precedes an Instantaneous Decoding Refresh (IDR) frame. 
     
     
         18 . The method of  claim 16 , the method comprising:
 adding a low frequency coefficient of a two dimensional transformed prediction to the residual error; and   taking an inverse transform.   
     
     
         19 . The method of  claim 16 , the method comprising:
 reconstructing a plurality of reference pixels required for an decoding an Intra block.   
     
     
         20 . The method of  claim 16 , the method comprising:
 extracting a low resolution Intra frame from a high resolution compressed Intra frame by:   generating frequency coefficients of a prediction needed, wherein the number of frequency coefficients are a fraction of the number of frequency coefficients in the high resolution compressed Intra frame being sampled;   generating corresponding frequency coefficients of the residual error;   reconstructing pixels of a particular block that will be used as reference pixels for decoding future blocks;   adding the frequency coefficients of the prediction needed to corresponding frequency coefficients of the residual error;   taking the inverse transform of the frequency coefficients;   normalizing a resulting down-sampled image; and   if a down-sampled frame generated is to be used as a reference frame, then any reference DC coefficients for a next frame are generated by a weighted average of four reference DC coefficients, so the DC coefficient of the reference block of the next frame are within a region covered by the four reference DC coefficients;   wherein using down-sampled Intra frames as reference frames for Intra frame reconstruction introduces drift errors which are accepted.   
     
     
         21 . The method of  claim 16 , the method comprising sub-sampling an Inter block in a predictive coding video bitstream by:
 using an integer portion of a motion vector to point to a reference block selected in a reference frame;   disregarding the fractional part of the motion vector;   generating frequency coefficients needed from a selected reference block, wherein, the number of frequency coefficients is no than one fourth of the frequency coefficients in an Intra frame;   dropping a fractional part in sub-sampling;   generating corresponding frequency coefficients of a residual error;   adding frequency coefficients of a prediction needed to corresponding frequency coefficients of a reconstructed residual error;   taking an inverse transform of one or more frequency coefficients; and   normalizing a resulting down-sampled block to form a down-sampled image.   
     
     
         22 . The method of  claim 16 , the method comprising:
 generating a reference DC coefficient for a next frame using a weighted average of four reference DC coefficients, wherein the next frame is to be used as a reference frame.