JPH0974562A - Image signal encoding method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To determine a scanning order excellent in encoding efficiency for a quantization coefficient in an encoding method for spatially dividing the inside of a block and performing an arbitrary shaped conversion. SOLUTION: An image is blocked and the blocked image data 11 is divided into two areas A and B. For area A and B data 12 and 13, an arbitrary shape DGT is performed according to the shapes of the data The coefficient obtained by performing the arbitrary shape DCT is quantized and quantization coefficients 14 and 15 are obtained. An index matching is performed for the index showing the row and columns of the qunatization coefficients 14 and 15 and the shapes of the areas, and a scanning is performed for the area where the coefficient exists in order from the lowest order coefficient. Based on these scanning information 16 and 17, the area A and the area B are integrated and the conversion coefficient for every area is scanned. Whether a 0 level is continuous or not according to the integrated scanning order 18 is investigated and a 0-run is detected. As a result, an entropy encoding is performed according to the level outputted according to the scanning order and the event of the 0-run.

Description

Detailed Description of the Invention

[0001]

TECHNICAL FIELD The present invention relates to an encoding method,
In particular, an image signal in which transform or filter processing is performed in units of overlapping blocks, boundaries such as edges are detected in these blocks, and the filter processing or transform is spatially divided and encoded based on the position of this boundary. The present invention relates to an encoding method.

[0002]

2. Description of the Related Art In a conventional image coding method, an image is divided into blocks of a certain size, and each block is divided into two-dimensional Discrete Cosine Transform (DC).
T) is transformed, the transform coefficient is quantized, and the quantization number is entropy coded. Or DC
A two-dimensional filter bank may be used instead of T. For these, one transform is used in block units or in overlapping block units, and the quantized coefficients are scanned in a zigzag manner in the block.

[0003]

When a block is spatially divided, an arbitrary shape is transformed, and the transform coefficient is quantized, the coefficient is generated by a plurality of transforms of different orders. Zigzag scanning cannot be applied as it is.

It is an object of the present invention to provide an image signal which determines a scan order which is coding efficient for quantization coefficients so that entropy codes can be used for mixed events of quantization level and zero level length. It is to provide an encoding method.

[0005]

According to the image signal coding method of the present invention, an arbitrary shape is transformed for each region spatially divided by a boundary, the transform coefficient is quantized, and a low-order basis is dealt with. From the conversion coefficient, the primary combination of the row number and the column number at the position where the conversion coefficient exists is given, a scanning direction change rule is given, and the conversion coefficient is scanned according to the order of the value of the primary combination and the scanning direction change rule. Then, entropy coding is performed according to the quantization level number and the continuous length value of level 0.

In another image signal coding method of the present invention, an arbitrary shape is transformed for each region spatially divided by a boundary, the transform coefficient is quantized, and the transform coefficient corresponding to a base of a low order is sequentially arranged. The primary combination of the row number and the column number at the position where the conversion coefficient exists is given, the scan change rule is given, and the order of the values of the primary combination and the scan order according to the scan direction change rule are used to perform the conversion. Order multiple areas,
Scan across the first region from the first region with respect to the first transform coefficient in the scan order, then scan across the multiple regions from the first region with respect to the second transform coefficient in the scan order. Then, scanning is performed over a plurality of regions up to the last transform coefficient in the scanning order, and entropy coding is performed according to the quantization level number and the value of the continuous length of level 0.

[0007]

BEST MODE FOR CARRYING OUT THE INVENTION In the present invention, an image is divided into blocks, and each block is spatially divided to perform conversion of an arbitrary shape. A sharp change of image data in a block is detected by using a technique such as an edge filter or texture analysis, and filtering processing and conversion are spatially divided based on the detected boundary position.

An example of boundary detection and an example of spatial division are shown in FIG. This is a case where a boundary having different luminance levels exists in the block of 8 × 8 pixels from the upper right to the lower left. At this time, the block shape can be divided into A and shape B. Usually, a block-based DCT (for example, 8 × 8 point DCT) is used to convert the inter-frame prediction error. However, if the shapes A and B can be separated, each of them is transformed using the arbitrary shape DCT. That is, instead of using the transform in which the basis vector of the DCT extends over the entire block, two types of transforms, one having a base matching the shape A and the other having a base matching the shape B are used.

For example, the shape A has the number of pixels per row and column as shown in Table 1.

[0010]

[Table 1] At this time, the data included in the shape A is the N-point DCT (N: any integer) corresponding to the number of pixels included in the row and the column.
Is converted using.

For the two-dimensional arbitrary shape DCT, first, a one-dimensional N-point DCT is used for a vector in the row direction. The image signal is f (x, y) (x: horizontal pixel number, y:
(Vertical pixel number), the result of performing DCT in the row direction is G (u, y)

[0012]

[Equation 1] It is. Next, the obtained G (u, y) is subjected to a one-dimensional N-point DCT in the column direction to obtain a DCT coefficient F (u,
v) is obtained.

[0013]

[Equation 2] The vectors in the row and column directions are shown in FIG. 2 by conversion
Nine coefficients are obtained. Arbitrary shape DCT for shape A
Then, the conversion coefficient F _A (i, j) is as shown in Table 2.

[0014]

[Table 2] On the other hand, the shape B is also converted using the N-point DCT (N: arbitrary integer) based on the lengths of the rows and columns, and 35 coefficients are obtained. When the arbitrary shape DCT is applied to the shape B, the conversion coefficient F _B (i, j) is as shown in Table 3. In the arbitrary shape DCT, the vectors having the largest number of pixels are rearranged in both column and row conversions.

[0015]

[Table 3] According to claim 1 of the present invention, the conversion coefficients F _A (u, v) and F _B
The numbers indicating the scanning order of (u, v) (meaning that the smaller value is scanned first) are respectively O _A (k) (k =
0, ····, 28), O B (l) (l = 0, ···
., 34) is equivalent to setting as follows.

[0016]

(Equation 3) It is.

The condition on the right side corresponds to the rule of changing the direction for zigzag scanning. In this formula u1 <u2
(U1: odd number) and u1> u2 (u1: even number),
Conversely, u1 <u2 (u1: even number) and u1> u2 (u1: odd number) can be set.

The same applies to O _B (k). The scanning order of the entire block, begins scanning O _B (k) after completing the scanning of O _A (k).

NTT Technology Journal 1994, Vol.
6, No. 10 “Trend of Image Coding Technology (1)” p. Four
In FIG. 8 of 3, zigzag scanning is performed with 8 × 8 blocks,
An example is described in which the number of consecutive level numbers zero (length of zero run) and the subsequent non-zero level number (non-zero coefficient) are grouped and one code is assigned to one group. The present invention is intended to realize this zigzag scanning with an arbitrary shape. It is basically zigzag, but in order to include skipping and zigzag tilting when the conversion coefficient does not exist, the primary combination of the row number and the column number is considered.

In general, it is also possible to use an inclined zigzag scan depending on the order of the magnitude of the value of au + bv and the rule of changing the scan direction. For example, 2u +
When ordering by the value of v and giving the direction change as a rule, F (0,0) <F (0,1) <F (0,2) <F
(1,0) <F (1,1) <F (0,3) <F (1,
2) <F (2,0) <..., and the scan order is tilted vertically. On the contrary, when u + 2v is used, the scan is tilted horizontally.

[0021] Claim 2 of the present invention has two applications, that is, when processing is performed by closing in a block and when processing is performed in a more enlarged area in the image (including scanning of all areas in the image). There is an example. To close the block,

[0022]

(Equation 4) And When the scanning progresses and the conversion coefficient F _A (u, v) or F _B (u, v) disappears, only the area where the conversion coefficient exists is scanned.

When scanning across a larger area in the image, F ₁ (u, v), F ₂ (u, v) ,.
.., F _m (u, v), when the areas are numbered, the order in the corresponding area is O ₁ (k), O ₂
(K), ..., O _m (k),

[0024]

(Equation 5) Defined by As the scanning proceeds, the coefficient F _k (u,
When v) does not exist, the process of skipping to the next area is performed.

Here, the case where the block is divided into image data of an arbitrary shape and the N-point DCT is used has been described. However, it is also possible to divide an image using "overlapping blocks" that allow overlapping with adjacent blocks. in this case,
The overlapping block is divided into arbitrary shapes, and an N-point subband filter bank is used instead of the DCT.

According to the present invention, when image data in a spatially divided block is transformed using a DCT of an arbitrary shape, scanning is performed in order from a low-order coefficient having a statistically large variance in each area. Therefore, for a coding method for designing an entropy code for a mixed event of a quantized level value and a continuous length of level 0, the continuous length of level 0 becomes longer and the coding efficiency is improved. This is because the higher order coefficient of the conversion coefficient has a statistically small level,
This is because the level often becomes 0 as the quantization width increases.

Also, when using a method of sequentially scanning the low-order coefficients having a large dispersion across the area,
It is expected that the continuous length of level 0 becomes longer and the coding efficiency is improved.

[0028]

Next, embodiments of the present invention will be described with reference to the drawings.

FIG. 1 is a flowchart of an image coding method (corresponding to claim 1) of an embodiment of the present invention.

First, the image is divided into blocks (step 1), and the block image data 11 having a predetermined size is divided into regions (step 2). However, it is necessary to obtain boundary information by some means in advance. When there is one boundary in the block, the blocked image data 11 is divided into two areas A and B. Now these are areas A
Data 12 and area B data 13. An arbitrary shape DCT is applied to these data 12 and 13 according to the shape (steps 3 and 4). Arbitrary shape D
The coefficient obtained by performing CT is quantized, and the quantized coefficient 1
4, 15 are obtained. The indexes representing the rows and columns of the quantized coefficients 14 and 15 are index-matched with the shape of the area (steps 5 and 6), and the area where the transform coefficient exists is scanned sequentially from the lowest coefficient. This scanning order information 16,
On the basis of 17, information on areas A and B is integrated and scanned for each area (step 7). Depending on the integrated scanning order 18, whether 0 levels continue or not, a 0 run is detected (step 8). As a result, the entropy coding is performed according to the level 19 and the event 19 of 0 run according to the scanning order (step 9).

FIG. 2 is a flowchart of an image coding method (corresponding to claim 2) of another embodiment of the present invention.

The image data 31 read in the image input (step 21) is divided into blocks of a predetermined size (steps 22 ₁ and 22 ₂ ). Here, for the sake of simplicity, the flowchart will be described taking two blocks as an example. However, usually the entire image is divided into blocks. Therefore, the image is divided into a large number of blocks.
The block image data 32 ₁ and 32 ₂ are divided into regions (steps 23 ₁ and 23 ₂ ). This is premised on that the boundary information is obtained in advance by some means (for example, edge detection). If there is one boundary in the block, the block image data 32 ₁ , 32 ₂
Is divided into two areas. Blocked image data 32
_For area ₁ , data 33 _{1 for} area A and data 33 _{2 for} area B are used, and for blocked image data 32 ₂ , data 33 _{3 for} area C and data 33 _{4 for} area D are used.
And

[0033] For these data 33 _to 333 _4,
An arbitrary shape DCT is applied according to the shape (steps 24 ₁ , 24 ₂ , 24 ₃ , 24 ₄ ). Arbitrary shape DCT
The coefficient obtained by performing
₁ , 34 ₂ , 34 ₃ , 34 ₄ are obtained. Quantization coefficient 3
4 _1, 34 _2, 34 _3, 34 ₄ of the shape of the index and the region representing the rows and columns are indexed verification (Step 25
₁ , 25 ₂ , 25 ₃ , 25 ₄ ) are sequentially scanned from the lowest coefficient in the area where the coefficient exists. Based on the scanning order information 35 ₁ , 35 ₂ , 35 ₃ , 35 ₄ , information on the areas A, B, C, and D is integrated, and low-order coefficients are crossed across the individual areas. Scanned (step 2
6).

Whether or not the 0 level continues according to the integrated scanning order 36 is detected (step 27). As a result, entropy coding is performed according to the level 37 and the 0 run event 37 output according to the scan order (step 28).

[0035]

As described above, according to the present invention, the image data in the spatially divided block is converted into D of an arbitrary shape.
When the transform is performed using CT, the transform coefficients are scanned in order from the low-order coefficient having a statistically large variance in each region, so that a mixed event having a continuous length of the quantization level value and the level 0 is obtained. On the other hand, when designing an entropy code, the continuous length of level 0 is statistically lengthened, and since coding can be performed with a smaller number of bits, the coding efficiency is improved.

[Brief description of drawings]

FIG. 1 is a flowchart of an image signal coding method (corresponding to claim 1) of an embodiment of the present invention.

FIG. 2 is a flowchart of an image signal coding method (corresponding to claim 2) of another embodiment of the present invention.

FIG. 3 is a diagram illustrating an example of detecting a boundary of an image signal and an example of spatial division.

FIG. 4 is a diagram showing conversion vectors in a row direction and a column direction.

[Explanation of symbols]

1-9, 21-28 Step 11 Blocked image data 12 Area A data 13 Area B data 14,15 Quantization coefficient 16,17 Scan order information 18 Integrated scan order 19 Level and 0 run event 31 Image Data 32 ₁ and 32 ₂ Blocked image data 33 ₁ Data of area A 33 ₂ Data of area B 33 ₃ Data of area C 33 ₄ Data of area D 34 _{1 to} 34 ₄ Quantization coefficient 35 _{1 to} 35 ₄ Scan order information 36 Integrated Scan Order 37 Level and 0 Run Events

Claims (2)

[Claims] 1. Transforms and filters are performed in blocks or overlapping blocks, boundaries such as edges are detected in these blocks, and the filters and transforms are spatially divided based on the positions of these boundaries. In the image signal coding method to be coded, transforms of arbitrary shape are performed for each of the regions spatially divided at the boundary, the transform coefficients are quantized, and transform coefficients corresponding to low-order bases exist in order. A linear combination of the row number and the column number of the position is taken, a scan direction change rule is given, the transform coefficient is scanned according to the order of the value of the primary combination and the scan direction change rule, and the quantization level number and level 0 An image signal coding method, characterized in that entropy coding is performed according to the value of the continuous length of. 2. Transform or filter processing is performed in units of blocks or overlapping blocks, boundaries such as edges are detected in these blocks, and the filter processing or transform is spatially divided based on the position of this boundary. In the image signal coding method to be coded, transforms of arbitrary shape are performed for each of the regions spatially divided at the boundary, the transform coefficients are quantized, and transform coefficients corresponding to low-order bases exist in order. The scanning direction conversion rule is given by taking the primary combination of the row number and the column number of the position, and the scanning order is performed by the order of the values of the primary combination and the scanning direction changing rule, and the plurality of converted regions are ordered. Scanning for a first transform coefficient in the scan order from a first region over a plurality of regions, and then a first region for the second transform coefficient in the scan order. To a plurality of regions, and up to the last transform coefficient in the scanning order, a plurality of regions are scanned, and entropy coding is performed according to the quantization level number and the continuous length value of level 0. An image signal encoding method characterized by performing.