General banding and codec banding artifact removal
A method and apparatus are disclosed for identifying and removing banding artifacts (i.e., false contours) resulting from insufficient bit depth caused by digital image quantization, conversion, and/or compression. This method includes: explicitly identifying texture block and flat block; de-termination of filter window sizes with the consideration of handling transitions between texture block and flat block; de-banding filtering with edge protection; and noise shaping according to de-banding filter result.
1. An apparatus for removing banding artifacts, comprising:
a block classification processing module; and
a de-banding filtering module coupled to the block classification processing module,
wherein the block classification processing module comprises:
an edge detection unit; and
a block classification unit coupled to the edge detection unit,
wherein the block classification unit is configured to differentiate flat blocks and texture blocks,
wherein the de-banding filtering unit is configured to removing banding artifacts according to the equation
{
I
′
(
x
,
y
)
=
I
0
+
FiltThresh
,
if
I
(
x
,
y
)
>
(
I
0
+
FiltThresh
)
;
I
′
(
x
,
y
)
=
I
0
-
FiltThresh
,
if
I
(
x
,
y
)
<
(
I
0
-
FiltThresh
)
;
I
′
(
x
,
y
)
=
I
(
x
,
y
)
,
else
.
I
″
(
x
,
y
)
=
∑
x
=
0
(
WinH
-
1
)
∑
y
=
0
(
WinW
-
1
)
I
′
(
x
,
y
)
WinH
×
WinW
,
and
wherein FiltThresh comprises a definable sigma in the equation that is controllable to steer the filter strength, I 0 comprises a gray value of the current pixel under processing which is located in the center of filter window, a current window size WinSize is WinW×WinH, FiltThresh further comprises the manually selected threshold for de-banding filtering, I(x,y) comprises a pray value of each pixel located within the filter window, I′(x,y) comprises an intermediate filtered output, and I″(x,y) comprises a final filtering output.
2. The apparatus of claim 1 , further comprising a noise shaping processing module coupled to the de-banding filtering module.
3. The apparatus of claim 1 , wherein the block classification processing module is configured to perform coded block location based image feature extraction and block type identification.
4. The apparatus of claim 1 , wherein the de-banding filtering module is configured to be applied to each pixel according to a filtering window.
5. The apparatus of claim 4 , wherein the de-banding filtering module is further configured to be steered by a definable sigma.
6. The apparatus of claim 2 , wherein the noise shaping processing module is configured to be conducted according to filtering results of the de-banding filtering module.
7. The apparatus of claim 1 , wherein the edge detection unit is configured to extract block-wise features of a coded block.
8. The apparatus of claim 1 , wherein the de-banding filtering module comprises:
a window size assignment unit; and
a de-banding filtering unit coupled to the window size assignment unit.
9. The apparatus of claim 1 , wherein the window size assignment unit is configured to determine a filtering window for a pixel.
10. The apparatus of claim 8 , wherein the filtering window is adaptively determined according to distance to boundary of flat block and texture block.
11. The apparatus of claim 2 , wherein the noise shaping processing module is configured to calculate a gray value difference between a de-banding output data and an original input frame, multiply the gray value difference with a spatial/temporal noise generated by a noise shaping unit as an adding-back-randomness, and adding the adding-back randomness to the de-banding output.
12. The apparatus of claim 11 , wherein the noise shaping unit is configured to generate temporal, spatial and pure flat noise.
13. A method for block classification comprising calculating a block max edge magnitude and a normalized block min edge magnitude as two axes of a 2D mask, wherein a value of the block max edge magnitude and the normalized block min edge magnitude represents the probability of banding artifacts.
14. The method of claim 13 , further comprising applying thresholdings to determine flat blocks and texture blocks.
15. A method for noise generation and noise shaping, comprising, if temporal noise is enabled, updating a seed of a noise generator temporally on each frame, otherwise, applying a fixed seed over a whole sequence.
16. The method of claim 15 , further comprising if spatial noise is enabled, adaptively adding a spatial noise on each pixel if a de-banding filter-in-out difference is not equal to zero.
17. The method of claim 16 , wherein noise strength control on luma/chroma is applied to the adaptively adding of the spatial noise.
18. The method of claim 16 , further comprising if pure flat noise is enabled, the spatial noise can be added on pure flat pixels.
19. The method of claim 18 , wherein pure flat pixels are detected by checking whether the de-banding filter-in-out difference equals to zero.
20. A method for removing banding artifacts, comprising:
one or more processors implementing on the following steps;
conducing block classification based on coded block location and block max/min edge analysis to explicitly identify texture block and flat block;
determining a de-banding window size according to surrounding block classifications and current pixel position within a current block to handle transitions between texture block and flat block;
applying a de-banding filter to the current pixel according to the determined de-banding window size; and
applying a noise shaping function according to a result of the de-banding filter to fine tune transitions between mono-tone region, de-banding region, and re-quantized output result,
wherein the de-banding filtering unit is configured to removing banding artifacts according to the equation
{
I
′
(
x
,
y
)
=
I
0
+
FiltThresh
,
if
I
(
x
,
y
)
>
(
I
0
+
FiltThresh
)
;
I
′
(
x
,
y
)
=
I
0
-
FiltThresh
,
if
I
(
x
,
y
)
<
(
I
0
-
FiltThresh
)
;
I
′
(
x
,
y
)
=
I
(
x
,
y
)
,
else
.
I
″
(
x
,
y
)
=
∑
x
=
0
(
WinH
-
1
)
∑
y
=
0
(
WinW
-
1
)
I
′
(
x
,
y
)
WinH
×
WinW
,
and
wherein FiltThresh comprises a definable sigma in the equation that is controllable to steer the filter strength, I 0 comprises a gray value of the current pixel under processing which is located in the center of filter window, a current window size WinSize is WinW×WinH, FiltThresh further comprises the manually selected threshold for de-bandinq filtering, I(x,y) comprises a gray value of each pixel located within the filter window, I′(x,y) comprises an intermediate filtered output, and I″(x,y) comprises a final filtering output.