Reparing corrupted data in a frame of an image sequence

Claim 1

A method of repairing a damaged pixel, having coordinates x and y, in a current frame f of an image sequence having a plurality of frames, the method comprising: determining a corrected value for the damaged pixel by an autoregression process using an autoregression model in the form of a linear weighted combination of the values of pixels in a zone containing the damaged pixel and located in the current frame and in substantially the same zone of a motion compensated frame preceding and/or succeeding the current frame in the image sequence; and replacing the value I (x,y,f) of the damaged pixel by said corrected value.

Claim 2

A method of repairing a damaged pixel, having coordinates x and y, in a current frame f of an image sequence having a plurality of frames, the method comprising: determining a corrected value for the damaged pixel by an autoregression process using an autoregression model in the form of a linear weighted combination of the values of pixels in a zone containing the damaged pixel and located in the current frame and in substantially the same zone of a motion compensated frame preceding and/or succeeding the current frame in the image sequence; and replacing the value I (x,y,f) of the damaged pixel by said corrected value.

Claim 3

A method of repairing a damaged pixel in a current frame f of an image sequence having a plurality of frames, the method comprising the steps of: (a) applying a motion compensation process to at least one of a preceding and a succeeding frame; (b) selecting a zone in said current frame containing said damaged pixel and a plurality of surrounding pixels; (c) selecting, using motion compensation data from step (a), substantially the same zone of the at least one of the preceding and succeeding frame, thus to define a three-dimensional support region of pixels; (d) applying an autoregression process, using an autoregression model in the form of a linear weighted combination of the values of the pixels in the selected region, to determine a corrected value for said damaged pixel; and (e) replacing the value I (x,y,f) of said damaged pixel with said corrected value.

Claim 4

A method of repairing a damaged pixel in a current frame f of an image sequence having a plurality of frames, the method comprising the steps of: (a) applying a motion compensation process to at least one of a preceding and a succeeding frame; (b) selecting a zone in said current frame containing said damaged pixel and a plurality of surrounding pixels; (c) selecting, using motion compensation data from step (a), substantially the same zone of the at least one of the preceding and succeeding frame, thus to define a three-dimensional support region of pixels; (d) applying an autoregression process, using an autoregression model in the form of a linear weighted combination of the values of the pixels in the selected region, to determine a corrected value for said damaged pixel; and (e) replacing the value I (x,y,f) of said damaged pixel with said corrected value.

Claim 15

A method of detecting a damaged pixel in a current frame f of an image sequence having a plurality of frames, the method comprising evaluating the magnitude of a difference ε(x,y,f) between the actual intensity value I(x,y,f) of a current frame test pixel and a predicted intensity value I(x,y,f) for the test pixel by an autoregression process using an autoregression model involving the intensity values of pixels in a zone containing the test pixel and located in both the current frame and in substantially the same zone of a motion compensated frame preceding and succeeding the current frame, comparing the difference to a threshold error value ε t and flagging the test pixel as damaged when said difference exceeds said threshold error value.

Claim 16

A method of detecting a damaged pixel in a current frame f of an image sequence having a plurality of frames, the method comprising evaluating the magnitude of a difference ε(x,y,f) between the actual intensity value I(x,y,f) of a current frame test pixel and a predicted intensity value I(x,y,f) for the test pixel by an autoregression process using an autoregression model involving the intensity values of pixels in a zone containing the test pixel and located in both the current frame and in substantially the same zone of a motion compensated frame preceding and succeeding the current frame, comparing the difference to a threshold error value ε t and flagging the test pixel as damaged when said difference exceeds said threshold error value.

Claim 19

A method as claimed in claim, 14 wherein there are two of said support regions and one of said support regions comprises only the corresponding pixel in the directly preceding frame and the other said support region comprising only the Corresponding pixel in the directly succeeding frame, so that M=1, and comparing the results obtained by using the two regions.

Claim 27

A method of estimating image displacement between two frames of an image sequence, wherein each frame comprises pixels defining image intensity values and of predetermined resolution, the method comprising the steps of: (a) generating for each frame L0 at least one sub-level Ln of reduced resolution, thus to provide n+1 levels L0 to LN of first and second frames of successively reducing resolution; (b) take the level Ln of lowest resolution; (c) divide each of the first and second frames so taken into a plurality of blocks; (d) determine displacement vectors for the blocks of the first of the two frames relative to the blocks of the second of the two frames by: (d)(i) calculating the mean absolute error MAE 0 between the block in the first of the frames and the block in the corresponding location in the second of the frames; (d)(ii) calculating the mean absolute error MAE m between the first block in the first frame and a plurality of blocks of a search region of m blocks in the second frame and which region contains the second block and each of the m blocks having an offset from the second block in the second frame; (d)(iii) identifying which of said plurality of searched blocks of the search region in the second frame gives rise to the minimum MAE m (MAE min ); and (d)(iv) when r*MAE min ≦MAE 0 , where r is a constant greater than 1, assigning to the first block, in the said first frame, a displacement vector corresponding to said offset of the identified block, else assigning a zero displacement vector; (e) when the current level is less than L0, take the frames at the next high resolution level and repeating steps (c) and (d), using the displacement vectors already obtained to define said corresponding location in step d(i).

Claim 28

A method of estimating image displacement between two frames of an image sequence, wherein each frame comprises pixels defining image intensity values and of predetermined resolution, the method comprising the steps of: (a) generating for each frame L0 at least one sub-level Ln of reduced resolution, thus to provide n+1 levels L0 to LN of first and second frames of successively reducing resolution; (b) take the level Ln of lowest resolution; (c) divide each of the first and second frames so taken into a plurality of blocks; (d) determine displacement vectors for the blocks of the first of the two frames relative to the blocks of the second of the two frames by: (d)(i) calculating the mean absolute error MAE 0 between the block in the first of the frames and the block in the corresponding location in the second of the frames; (d)(ii) calculating the mean absolute error MAE m between the first block in the first frame and a plurality of blocks of a search region of m blocks in the second frame and which region contains the second block and each of the m blocks having an offset from the second block in the second frame; (d)(iii) identifying which of said plurality of searched blocks of the search region in the second frame gives rise to the minimum MAE m (MAE min ); and (d)(iv) when r*MAE min ≦MAE 0 , where r is a constant greater than 1, assigning to the first block, in the said first frame, a displacement vector corresponding to said offset of the identified block, else assigning a zero displacement vector; (e) when the current level is less than L0, take the frames at the next high resolution level and repeating steps (c) and (d), using the displacement vectors already obtained to define said corresponding location in step d(i).

Claim 31

A method of restoring a damaged current frame of an image sequence having a plurality of frames, the method comprising: (a) dividing current frame f into a plurality of blocks containing substantially equal numbers of pixels; (b) matching said blocks with corresponding blocks in at least one of a preceding frame and a succeeding frame using a motion estimation process to define a displacement vector for each block; (c) selecting a data volume containing the current frame and a preceding and a succeeding frame; (d) using a 3-D-autoregression process on said volume and utilizing the displacement vectors to compensate for motion between frames to generate a prediction intensity error for substantially each pixel in the current frame f; (e) identifying damaged pixels in the current frame by comparing the predicted errors with a threshold error; and (f) using a 3D-autoregression process involving the intensity values of pixels surrounding identified pixels to obtain corrected values for the identified pixels.

Claim 32

A method of restoring a damaged current frame of an image sequence having a plurality of frames, the method comprising: (a) dividing current frame f into a plurality of blocks containing substantially equal numbers of pixels; (b) matching said blocks with corresponding blocks in at least one of a preceding frame and a succeeding frame using a motion estimation process to define a displacement vector for each block; (c) selecting a data volume containing the current frame and a preceding and a succeeding frame; (d) using a 3-D-autoregression process on said volume and utilizing the displacement vectors to compensate for motion between frames to generate a prediction intensity error for substantially each pixel in the current frame f; (e) identifying damaged pixels in the current frame by comparing the predicted errors with a threshold error; and (f) using a 3D-autoregression process involving the intensity values of pixels surrounding identified pixels to obtain corrected values for the identified pixels.