Reference pixel reduction for intra LM prediction

1. A video codec comprising:

a processor configured to:

generate a prediction block for a chroma block, wherein the prediction block comprises a predicted chroma sample, wherein the predicted chroma sample is based on:

a filtered reconstructed luma sample located in a corresponding reconstructed luma block;

a plurality of downsampled filtered reconstructed luma samples located in positions neighboring the corresponding reconstructed luma block, the downsampled filtered reconstructed luma samples comprise every Nth filtered reconstructed luma sample of a top neighboring vector of reconstructed luma samples and a left neighboring vector of reconstructed luma samples, where N is an integer greater than one; and

a plurality of downsampled chroma samples located in positions neighboring the chroma block, the downsampled chroma samples comprise every Nth chroma sample of a top neighboring vector of chroma samples and a left neighboring vector of chroma samples;

generate a second reconstructed luma block corresponding to a second chroma block; and

generate a second prediction block for the second chroma block based on the corresponding second reconstructed luma block, wherein the second prediction block is further based on a subset of a plurality of filtered reconstructed luma samples located in positions neighboring the second reconstructed luma block when the second chroma block is of dimension 16×16 or 8×8, wherein the subset is obtained by downsampling, and otherwise wherein the second prediction block is further based on the plurality of filtered reconstructed luma samples located in positions neighboring the second reconstructed luma block.

2. A video codec comprising:

a processor configured to:

generate a prediction block for a chroma block, wherein the prediction block comprises a predicted chroma sample, wherein the predicted chroma sample is based on:

a filtered reconstructed luma sample located in a corresponding reconstructed luma block;

a plurality of downsampled filtered reconstructed luma samples located in positions neighboring the corresponding reconstructed luma block, wherein the downsampled filtered reconstructed luma samples comprise a sliding average of a top neighboring vector of reconstructed luma samples and a left neighboring vector of reconstructed luma samples; and

a plurality of downsampled chroma samples located in positions neighboring the chroma block, the downsampled chroma samples comprise a sliding average of a top neighboring vector of chroma samples and a left neighboring vector of chroma samples;

wherein the sliding average of the top neighboring vector of reconstructed luma samples comprises

1

N

⁢

∑

i

=

0

N

-

1

⁢

p

Y

′

⁡

[

N

*

x

+

i

,

-

1

]

for x=0, . . . , nS/N−1, where nS is a dimension of the chroma block, N is a downsampling factor, and p Y ′[x, −1] for x=0, . . . , nS−1 is the left neighboring vector, and wherein the sliding average of the left neighboring vector of reconstructed luma samples comprise

1

N

⁢

∑

i

=

0

N

-

1

⁢

p

Y

′

⁡

[

-

1

,

N

*

y

+

i

]

for y=0, . . . , nS/N−1, where p Y ′[−1, y] for y=0, . . . , nS−1 is the top neighboring vector.

3. A method used in video coding to generate predicted chroma samples for a chroma block, comprising:

downsampling, performed by a video coder at least partially implemented in hardware, filtered reconstructed luma samples located in positions neighboring a reconstructed luma block to generate a plurality of downsampled filtered reconstructed luma samples, the downsampled filtered reconstructed luma samples comprise every Nth filtered reconstructed luma sample of a top neighboring vector of reconstructed luma samples and a left neighboring vector of reconstructed luma samples, where N is an integer greater than one;

downsampling, performed by the video coder, reconstructed chroma samples located in positions neighboring a chroma block to generate a plurality of downsampled reconstructed chroma samples, the downsampled chroma samples comprise every Nth chroma sample of a top neighboring vector of chroma samples and a left neighboring vector of chroma samples; and

generating, performed by the video coder, a prediction block for the chroma block, wherein the prediction block comprises a predicted chroma sample, wherein the predicted chroma sample is based on:

a filtered reconstructed luma sample located in the reconstructed luma block;

the plurality of downsampled filtered reconstructed luma samples; and

the plurality of downsampled reconstructed chroma samples; and

generating a second reconstructed luma block corresponding to a second chroma block; generating a second prediction block for the second chroma block based on the corresponding second reconstructed luma block, wherein the second prediction block is further based on a subset of a plurality of filtered reconstructed luma samples located in positions neighboring the second reconstructed luma block when the second chroma block is of dimension 16×16 or 8×8, wherein the subset is obtained by downsampling, and otherwise wherein the second prediction block is further based on the plurality of filtered reconstructed luma samples located in positions neighboring the second reconstructed luma block.

4. The method of claim 3 , wherein the downsampled filtered reconstructed luma samples comprise an average of a top neighboring vector of reconstructed luma samples and a left neighboring vector of reconstructed luma samples.

5. The method of claim 4 , wherein the downsampled chroma samples comprise a sliding average of a top neighboring vector of chroma samples and a left neighboring vector of chroma samples.

6. The method of claim 5 , wherein the sliding average of the top neighboring vector of reconstructed luma samples comprise

1

N

⁢

∑

i

=

0

N

-

1

⁢

p

Y

′

⁡

[

N

*

x

+

i

,

-

1

]

for x=0, . . . , nS/N−1, where nS is a dimension of the chroma block, N is a downsampling factor, and p Y ′[x, −1] for x=0, . . . , nS−1 is the left neighboring vector, and wherein the sliding average of the left neighboring vector of reconstructed luma samples comprises

1

N

⁢

∑

i

=

0

N

-

1

⁢

p

Y

′

⁡

[

-

1

,

N

*

y

+

i

]

for y=0, . . . , nSN−1, where p Y ′[−1, y] for y=0, . . . , nS−1 is the top neighboring vector.

7. A video codec comprising:

a processor configured to:

generate a filtered reconstructed luma block corresponding to a chroma block;

downsample a plurality of filtered reconstructed luma samples and a plurality of reconstructed chroma samples to generate downsampled filtered reconstructed luma samples and downsampled reconstructed chroma samples, respectively, when the chroma block is larger than a threshold block size, the plurality of filtered reconstructed luma samples are located in neighboring blocks of the filtered reconstructed luma block, and the plurality of reconstructed chroma samples are located in neighboring blocks of the chroma block;

generate a prediction block for the chroma block, the prediction block comprises a plurality of predicted chroma samples, each of the plurality of predicted chroma samples is based on:

a corresponding reconstructed luma sample located in a corresponding reconstructed luma block;

the downsampled filtered reconstructed luma samples and the downsampled reconstructed chroma samples when the chroma block is larger than the threshold block size; and

the plurality of filtered reconstructed luma samples and the plurality of reconstructed chroma samples when the chroma block is equal to the threshold block size,

wherein the threshold block size is 4×4, and the downsampling of is by a factor of two when the chroma block is 8×8 and the downsampling is by a factor of four when the chroma block is 16×16.

8. The video codec of claim 7 , wherein the downsampling of the filtered reconstructed luma samples comprises performing a sliding average of a top neighboring vector of reconstructed luma samples and a left neighboring vector of reconstructed luma samples.

9. The video codec of claim 8 , wherein the downsampling of the reconstructed chroma samples comprises performing a sliding average of a top neighboring vector of chroma samples and a left neighboring vector of chroma samples.

10. The video codec of claim 7 , wherein the downsampling of the filtered reconstructed luma samples comprises selecting every Nth filtered reconstructed luma sample of a top neighboring vector of reconstructed luma samples and a left neighboring vector of reconstructed luma samples, where N is an integer greater than one.

11. The video codec of claim 10 , wherein the downsampling of chroma samples comprises selecting every Nth chroma sample of a top neighboring vector of chroma samples and a left neighboring vector of chroma samples.