Metadata filtering for display mapping for high dynamic range images

1. A method for metadata filtering with a processor, the method comprising:

accessing with a processor an input video stream comprising video frames;

accessing input metadata for the input video stream, the input metadata comprising at least one of minimum (min), average (mid), or maximum (max) luminance values of the video frames in the input video stream; and

for an input frame in the input video stream:

testing whether the input frame belongs to a current scene or a new scene; and upon determining the input frame belongs to the current scene:

filtering the input metadata to generate output metadata for the input frame, wherein the filtering is based only on metadata from input frames belonging to the current scene, wherein said filtering comprises computing the output metadata with an exponential moving average filter as:

S t =L F(0) , for t = 0,

S t =α k * L F(t) +β* S t−1 ,for t> 0 ,

where S t denotes the output metadata, α k (alpha) and β (beta) denote weight factors, L F(t) denotes a function of the at least one of min, mid, or max luminance values for the input frame at time t, and t=0 denotes the start of the current scene, wherein α k is a function of a delivery frame rate such that α k decreases when the delivery frame rate increases.

2. The method of claim 1 , further comprising generating the at least one of min, mid, or max luminance values, by an image analysis processor, based on luminance values of the video frames in the input video stream.

3. The method of claim 1 , wherein β=1−α k .

4. The method of claim 1 , wherein instead of initializing the EMA filter based on the luminance values of a frame starting the current scene, the EMA filter is initialized based on the luminance values of frames in a past scene, wherein the past scene is statistically similar to the current scene.

5. The method of claim 1 , wherein determining whether the input frame belongs to the current scene comprises:

computing luminance, chromaticity, and edge-strength histograms for the video frames in the input video stream;

computing a weighted Euclidian distance between histogram values of the input frame and a prior frame; and

determining that the weighted Euclidian distance does not exceed a threshold value.

6. The method of claim 5 , further comprising determining a probability measure of whether the input frame belongs to the current scene, the method comprising computing

SceneCut

=

Δ

t

Δ

t

+

C

,

where c is a tunable constant, SceneCut denotes the probability measure, and Δ t denotes the weighted Euclidian distance between histogram values of the input frame and the prior frame.

7. The method of claim 1 , further comprising:

computing with a first EMA filter a time-forward moving average value based on the input frame and past frames;

computing with a second EMA filter a time-reverse moving average value based on the input frame and future preview frames; and

determining whether the input frame belongs to the current scene is based on a measure of a distance between the time-forward moving average value and the time-reverse forward moving average value.

8. The method of claim 7 , wherein upon detecting the input frame belongs to a new scene, initialization of the first EMA filter to compute forward moving average values for the new scene is based on the reverse moving average value for the previous scene.

9. The method of claim 1 , further comprising generating using the output metadata an output tone-mapped frame by mapping the dynamic range of the input frame to the dynamic range of a target display.

10. An apparatus comprising a processor and configured to perform the method recited in claim 1 .

11. A non-transitory computer-readable storage medium having stored thereon computer-executable instruction for executing a method with one or more processors in accordance with claim 1 .

12. A method for metadata filtering with a processor, the method comprising:

accessing with a processor an input video stream comprising video frames;

accessing input metadata for the input video stream, the input metadata comprising at least one of minimum (min), average (mid), or maximum (max) luminance values of the video frames in the input video stream; and

for an input frame in the input video stream:

testing whether the input frame belongs to a current scene or a new scene; and upon determining the input frame belongs to the current scene:

filtering the input metadata to generate output metadata for the input frame, wherein metadata filtering comprises a look-ahead sliding window comprising P frames in the current scene prior to the input frame and A frames in the current scene subsequent to the input frame in the current scene, and the output metadata is computed as

S

j

=

1

N

⁢

∑

i

=

S

E

⁢

w

⁡

(

N

,

i

-

S

)

*

L

F

⁡

(

i

)

,

j

=

0

,

1

,

…

⁢

,

F

-

1

,

where, L F(i) denotes an input metadata value for input frame i, w(N, k), for k=0, 1, . . . , N-1, denotes filter weights, and S j denotes the output metadata value for frame j, and wherein given the start of the current scene at j=0 and the end of the current scene at j=F-1, S =max(j-P,0) , E=min(F-1, j+A), and N=E-S+1.