KR100269206B1 - Flexible resolution scalable binary shape coder & the method thereof - Google Patents

Abstract

PURPOSE: A scalable binary shape coder and a method for the same are provided to obtain an original image of a desired resolution by converting easily resolution between layers with an interlace scan. CONSTITUTION: A scalable binary shape coder is used for obtaining shape information of various resolution by using one bit stream. Blocks of each sample number are divided by n and m, respectively if a conversion rate of an L-layer of a lower resolution to an L+1 layer of a higher resolution is not more than 2. A sampling process(203) for only data(206) of 2(n-m) number of the L+1 layer is performed. The sampled data of the L+1 layer correspond to (n-m) number of the L layer. The remaining data(207) of 2m-n number of the L+1 layer corresponds to data(210) of 2m-n of the L layer without the sampling process.

Description

Arbitrary Resolution Multilayer Binary Coder and Its Method

The present invention relates to a multi-layer binary coder, and more particularly, to a multi-resolution binary coder using a bi-layer binary coder, an arbitrary resolution multi-layer binary code using a bi-scan line processing method capable of converting the resolution between layers in a free ratio. An encoder and a method thereof are provided.

In the related art, in consideration of the resolution of a device receiving a bit stream when transmitting a bit stream containing image information, bit streams having different formats are transmitted for the same information. By improving the transmission method according to this method, a bit stream suitable for a scalable or multi-layer in which one piece of information includes shape information of various resolutions has been transmitted.

FIG. 1 illustrates a basic conceptual diagram of a scalable binary shape coder. In general, a multilayer binary shape coder for receiving a multi-layered bit stream as described above and performing binary shape encoding on the basis of The resolution conversion ratio between the base layers 101 and 103 and the enhancement layers 102 and 104 is fixed at 1: 2. In the base layer, the video object plane (hereinafter referred to as VOP) I VOP, which is a rectangular boundary region surrounding the outermost part of the shape information, is referred to as intra coding, and P VOP (predicted VOP) as a VOP encoded at a previous time. From motion prediction encoding. Only two types of I VOP 101 and P VOP 103 exist in the base layer. The upper layer of the base layer I VOP 101 becomes the P VOP 102. The P VOP 102 reconstructs the values of the upsampling 106 of the base layer using predictive coding. When the base layer is the P VOP 103, the upper layer is the B VOP (Bidirectional VOP) 104. The B VOP 104 is reconstructed using both the interpolated 108 value of the base layer and the predicted value 107 from the previous VOP 102 value. In such a binary encoder, when there are other higher layers not shown here, the resolution conversion ratio of the base layers 101 and 103, the upper layers 102 and 104, and other higher layers (not shown) is 1: 2: 4. The resolution conversion ratio is bound to be fixed at a constant conversion ratio even if there are several more upper layers. That is, since the resolution conversion ratio between adjacent layers is fixed at 1: 2, there are many limitations in trying to realize a multi-layer or reconstructing an image having a desired resolution on one bit stream by considering multiple layers simultaneously. In addition, in the conventional method, when considering texture information of an image inside a shape, a resolution difference between texture information that can be converted to an arbitrary resolution and shape information that is not occurs occurs to reduce the additional calculation. This may be necessary or may result in degradation of image quality.

The present invention has been made in an effort to provide an arbitrary resolution multi-layer encoding method in which a multi-layer resolution conversion ratio is free in a multi-layer binary shape encoder using a bi-scan line processing method.

1 is a basic conceptual diagram of a multilayer binary coder.

2A and 2B illustrate a method of converting resolution of binary images.

3 shows an example of 4: 1 ratio resolution conversion.

4 illustrates the formation of a video object plane for resolution conversion of any ratio.

5 shows an example of mismatch of predicted rectangular boundary region (VOP) position.

6 illustrates a predicted rectangular boundary region (VOP) position correction method.

Figure 7 shows a data processing method using the correlation between every other scan line.

8 is context information according to a scan line direction for encoding.

9 is a flowchart illustrating an arbitrary resolution multi-layer binary shape coding method of the present invention.

In order to solve the above problems, an arbitrary resolution multi-layer binary shape coder that obtains shape information of various resolutions on one bitstream has a resolution conversion ratio of an L layer having a lower resolution and a higher layer L + 1 layer having a higher resolution. If it is not larger than 2, each sample block is divided into n and m in the same manner as the resolution conversion ratio of the L + 1 layer and the L layer, and only 2 (nm) data of the L + 1 layer is sampled. It corresponds to (nm) pieces of the L layer, and the remaining 2m-n pieces of data in the L + 1 layer correspond to the L layer without being sampled.

Downsampling from the L + 1 layer to the L layer includes a vertical coordinate y _L of pixel coordinates (x, y _L ) of the L layer and pixel coordinates (x, y _H ) of the L + 1 layer. For the relationship of the vertical coordinate y _H of, if y _L divided by m is less than (n-m) then y _H is 2 x (y _L % m) + nx (y _L / m) + 1 Otherwise y _H is preferably (y _L % m) + nx (y _L / m) + (n-m).

Up-sampling (upsampling) to said L layer in said L + 1 layer, y pixel coordinate y _L and the pixel coordinates (x, y _H) of said L + 1 layer of (x, y _L) of the L Layer _For the relationship of _H , if y _H divided by n is less than 2 (nm) y _L is (y _H % n) / 2 + mx (y _H / n) otherwise y _L is (y _It is preferable to set _H % n) + mx (y _H / n) + (m-n).

In order to solve the above problems, an arbitrary resolution multi-layer binary shape coder that obtains shape information of various resolutions on one bitstream has a resolution conversion ratio of an L layer having a lower resolution and a higher layer L + 1 layer having a higher resolution. If the value is greater than 2, each sample number block is divided into n and m in the same manner as the resolution conversion ratio of the L + 1 layer and the L layer, and only 2m data of the n samples of the L + 1 layer are sampled. M data samples of the L layer are obtained, and the remaining n-2m data of the L + 1 layer are ignored and do not correspond to the L layer.

Downsampling from the L + 1 layer to the L layer includes a vertical coordinate y _L of pixel coordinates (x, y _L ) of the L layer and pixel coordinates (x, y _H ) of the L + 1 layer. the vertical position in relation y _H, y _H is _{2 x (y L% m)} + nx (y L / m) + arbitrary resolution multi-layer coder binary shape as that set characteristics to 1. and otherwise It is preferable to set y _H to (y _L % m) + nx (y _L / m) + (n-m).

Up-sampling (upsampling) to said L layer in said L + 1 layer, y pixel coordinate y _L and the pixel coordinates (x, y _H) of said L + 1 layer of (x, y _L) of the L Layer _For the relationship of _H , if y _H divided by n is less than 2m then y _L is (y _H % n) / 2 + mx (y _H / n), otherwise y _L is mx (y _H / n) + m-1 is preferred.

In order to solve the above problems, in an arbitrary resolution multi-layer binary shape encoder which obtains shape information of various resolutions on one bitstream, resolution conversion of an L layer having a lower resolution and a higher layer L + 1 layer having a higher resolution is performed. If the ratio is greater than 2, and the resolution conversion ratio of the L + 1 layer and the L layer is n: m, the conversion layer is divided into several layers having a resolution difference of not less than two times for each layer. Characterized by sequentially encoding.

The downsampling between the divided layers is called N + 1 layer, lower layer is called N layer, and the sampling blocks of N + 1 and N are the same number of conversion ratios as j and i. When divided, y _L is defined as i for the relationship between the vertical coordinate y _L of the pixel coordinates (x, y _L ) of the N layer and the vertical coordinate y _H of the pixel coordinates (x, y _H ) of the L + 1 layer. Y _H is 2 x (y _L % i) + jx (y _L / i) + 1 if the remainder divided by is less than (j-i), otherwise y _H is (y _L % i) + Preferably, jx (y _L / i) + (j-i).

Up-sampling (upsampling) to said N-th layer in the N + 1 layers, y of the pixel coordinates (x, y _L) y _L and the pixel coordinates (x, y _H) in the (N + 1) -th layer of said N layer the relationship of _H, is less than the remaining value is 2 (ji) divided by y _H to j a y _L to _{(y H% j) / 2} + ix (y H / j) , otherwise, the y _L (y _H % j) + ix (y _H / j) + (i-j) is preferred.

In order to solve the above problems, in an arbitrary resolution multi-layer binary shape encoder that obtains shape information of various resolutions on one bitstream, a lower L layer having a lower resolution and a higher L + 1 layer having a higher resolution are provided. When converting the resolution of a binary image when the resolution conversion ratio is a predetermined ratio of m to n, the video object region, which is a rectangular boundary region surrounding the shape information, to eliminate the loss of reference data between the upper layer of L + 1 and the lower layer of L ( Video Object Plane (hereinafter referred to as VOP) includes: a first step of inputting binary shape information; Obtaining a rectangular area S 'surrounding the object in the binary shape with the minimum area; In order to ensure that the location of the reconstructed boundary region after the resolution conversion of the boundary of the rectangular region S matches the object boundary of the original information, the square boundary is fitted to a sampling grid divided for binary shape information sampling. Obtaining a rectangular area S that extends the boundary of the second step; And a fourth step of obtaining a rectangular boundary S ″ having an area of a unit sampling block M × N size to be encoded starting from the start point of S.

When the starting point of S '(x' _s , y ' _s ), the starting point of S (x _s , y _s ), the abscissa length of the sampling grid is d _H , the ordinate length is d _V , x _s is x and _"s in the x _'s for subtracting the remainder divided by the value _H 2d, y _s is preferred that remainders obtained by dividing the y' _s in y _'s by _V 2d.

Transverse length of the longitudinal axis of the S, and when the length of the lateral, longitudinal axis of the S 'La each H, W, minus the remaining (W% d _H) divided by the W by d _H in the d _H value to be a in the _V d obtained by adding the value obtained by subtracting the remainder (H% _V d) obtained by dividing the _V H d as each H, W in length are preferred.

In order to solve the above problems, in a multi-layer binary shape encoder that sequentially obtains shape information of various resolutions on one bitstream, resolution conversion of an L layer having a lower resolution and a higher layer L + 1 layer having a higher resolution is performed. When the ratio is a predetermined ratio of m to n, the starting point of the VOP upsampling the rectangular boundary region surrounding the object (hereinafter referred to as VOP) in the lower layer of the binary image coincides with the starting point of the VOP in the upper layer. The method may include creating a reference VOP having the same position and size as the upper layer VOP; Moving the upsampled VOP by the amount of change from the lower layer to the start point of the upsampled VOP based on the start point of the upper layer VOP; Copying contents of a range of contents of the moved VOP not exceeding the reference VOP; And allocating a remaining area of the reference VOP area where the contents of the moved VOP is not copied to '0'.

In order to solve the above problems, in a multi-layer binary shape encoder that sequentially obtains shape information of various resolutions on one bitstream, resolution conversion of an L layer having a lower resolution and a higher layer L + 1 layer having a higher resolution is performed. When a ratio of m / n is a predetermined ratio and the ratio of n / m is 2 or less, the type of scan line in the upper layer includes: a scan line to be restored by decoding in the upper layer; A reference scan line for obtaining a scanning line of a binary image of an upper layer by upsampling when the resolution changes from the lower layer; And a scan line copied as is from the lower layer.

In the structure of encoding binary image information in units of a macro alpha block (BAB), when the sample values of the upper and lower scanning lines of the pixel to be encoded are the same in the BAB, but there are exception sample data different from the pixel values to be encoded. It is preferable to perform encoding on all pixels in the BAB, and to perform encoding only on pixels when the upper and lower scanning lines are different when the exceptional sample data does not exist in the BAB.

In order to solve the above problems, in a multi-layer binary shape encoder that sequentially obtains shape information of various resolutions on one bitstream, resolution conversion of an L layer having a lower resolution and a higher layer L + 1 layer having a higher resolution is performed. When the ratio of n / m is 2 or more as a predetermined ratio whose ratio is m to n, the type of scan line in the upper layer includes: a scan line to be restored by decoding in the upper layer; A reference scan line for obtaining a scanning line of a binary image of an upper layer by upsampling when the resolution changes from the lower layer; And scan lines that are encoded as they are without reference to pixel values of lower layers in the upper layer.

In the structure of encoding binary image information in units of a macro alpha block (BAB), when the sample values of the upper and lower scanning lines of the pixel to be encoded are the same in the BAB, but there are exception sample data different from the pixel values to be encoded. It is preferable to perform encoding on all pixels in the BAB, and to perform encoding only on pixels when the upper and lower scanning lines are different when the exceptional sample data does not exist in the BAB.

In order to solve the above problems, in a multi-layer binary shape encoder that sequentially obtains shape information of various resolutions on one bitstream, resolution conversion of an L layer having a lower resolution and a higher layer L + 1 layer having a higher resolution is performed. The sample value of the upper layer is a _H [y _H ] with respect to the binary values a _L , a _H , the vertical coordinates y _L, y _H and the horizontal coordinate x of the pixel for each layer with a predetermined ratio of m to n. [x], and when the sample value in the lower layer L is a _L [y _L ] [x], for the efficiency of encoding in the structure of encoding binary image information in units of a macro alpha block (BAB) The type of BAB can be distinguished from a _H [y _H ] [x] where (y _H % n) <2 x (n-m) and ((y _H % n)% 2) is '0'. the first step to obtain the position y _H in the sample to be coded (here, "%" is an operator which takes the remainders of division, and "/" is a division result when an integer value satisfying the Only takes the operator); The upper that corresponds to the location of the sample y _H layer scan line pixels to be encoded in the above _H a location of the pixel to be coded in the _[H y] [x], obtaining a pixel value below the scanning line, as shown in Equation 10

Where y _L = (y _H % n) / 2 + mx (y _H / n) Means an exclusive OR operation.)

A second step of performing an exclusive OR (XOR); A third step of setting the value a _H [y _H ] [x] to be encoded as a transition sample value (TSD) having different sample values of the upper and lower scan line pixels when the result value of Equation 10 is '1'; If the result value of Equation 10 is '0' and the value to be encoded a _H [y _H ] [x] is not equal to the value a _L [y _L ] [x] referenced in the lower layer, the pixel value is excluded. A fourth step of ending with an exception BAB (ESD) and an ESD BAB (Exeptioanl BAB); a fifth step of repeating from the first step if (y _H , x) is the last pixel of the scan line to y _H ; Since there is no exception sample data ESD, there is a sixth step in which the corresponding BAB is a transition BAB in which no exception sample data exists.

According to another aspect of the present invention, there is provided an arbitrary resolution multi-layer binary shape encoding method comprising: inputting binary image information having a predetermined size surrounded by a rectangular boundary region (VOP); Downsampling the image information for encoding in a lower layer when the resolution ratio of the upper layer and the lower layer is n to m; Lossless encoding the down-sampled image with a predetermined shape encoder; Upsampling the encoded image for use in encoding of a higher layer and generating a reference VOP coinciding with a start point of an original VOP of the higher layer; After selecting the type of the scan line to be encoded in the upper layer, the reference scan line is obtained from the reference VOP, and whether there is exception information in the sample unit block BAB to be encoded is a pixel having a different pixel value from the same upper and lower pixel values. Distinguishing; If there is exception information in the BAB, the pixel is encoded using a binary arithmetic coder for all the pixels in every other scan line. If there is no exception information, only the pixels with exclusive ORs between adjacent scan lines is 1 Encoding with BAC.

Wherein the down-sampling are, for the relationship of said n / m is 2 or less when, y _H of the pixel coordinates (x, y _L) y _L and the pixel coordinates (x, y _H) in the upper layer of the lower layer , if y _L divided by m is less than (n-m) then y _H is 2 x (y _L % m) + nx (y _L / m) + 1 otherwise y _H is (y _L % m) + nx (y _L / m) + (n-m) is preferred.

The downsampling is, n / m is 2 or more when, in the relation of y _H of the pixel coordinate pixel coordinates (x, y _H) in the y _L and an upper layer of the (x, y _L) of the lower layer, y _H Is a 2 x (y _L % m) + nx (y _L / m) + 1 arbitrary resolution multi-layer binary shape coder, otherwise y _H is (y _L % m) + nx ( y _L / m) + (n-m) is preferred.

The up-sampling is, n / m at this time is less than 2, the relationship of y _H of the pixel coordinates (x, y _L) y _L and the pixel coordinates (x, y _H) in the upper layer of the lower layer, y _{If the} remainder of _H divided by n is less than 2 (nm) y _L is (y _H % n) / 2 + mx (y _H / n), otherwise y _L is (y _H % n) + mx ( y _H / n) + (m-n) is preferred.

The up-sampling is, n / m is 2 or more case, the relationship of y _H of the pixel coordinates (x, y _L) y _L and the pixel coordinates (x, y _H) in the upper layer of the lower layer, y _H If y is divided by n and less than 2m, then y _L is (y _H % n) / 2 + mx (y _H / n), otherwise y _L is mx (y _H / n) + m-1 This is preferred.

Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

FIG. 2A illustrates a sampling data selection method for inter-layer resolution conversion in FIG. 1, wherein a binary alpha block (hereinafter referred to as BAB) 201 in an L + 1 layer and a BAB in an L th layer are illustrated. If 202 varies in resolution by n to m in the vertical direction, n samples 205 in the vertical direction of the L + 1 layer correspond to m samples 208 in the vertical direction of the L layer. Conventional scan interleaving method is to sample the even scan line (or odd scan line) of the upper layer and to encode in the base layer and again to the odd scan line (or even scan line) of the upper layer This technique is used for restoring) and skips sampling one by one when the resolution changes. In this case n and m are limited to 2 and 1, respectively. To further extend this ratio, only 2 (nm) data 206 of the n samples of the L + 1 th layer are sampled 203 to obtain (nm) samples 209 of the m data of the L th layer. The remaining (2m-n) data 207 of the L + 1th layer are mapped to the (2m-n) data 210 of the Lth layer as it is (204). Considering the resolution conversion in the vertical direction, the binary value of the pixel for each layer, a _L, a _H, the vertical coordinate y _L, y _H, and the horizontal coordinate x d when L + 1 binary image data in the second layer a _H [y _H ] [x] is downsampled into binary image data a _L [y _L ] [x] in the L-th layer as follows. This is shown in equation (1).

a _L [y _L ] [x] = a _H [y _H ] [x]

Here, [] denotes a determinant, and the relationship between y _L and y _H may be expressed by Equation 2 according to a predetermined program language.

if ((y _L % m) <(n-m)) {

y _H = 2 x (y _L % m) + nx (y _L / m) + 1

} else {

y _H = (y _L % m) + nx (y _L / m) + (n-m)

}

This is the y value of the layer L + 1, where y _H = 2 x (y _L % m) + nx (y _L / m) + 1 if the remainder of y _L divided by m is less than n minus m. It is shown that we derive a value of y otherwise y _h = (y _L % m) + nx (y _L / m) + (n-m). Here, '%' is the remainder operator, and '/' is the operator that returns only the integer value after the decimal point is ignored. Contrary to the above logic, the binary image data a _H [y _H ] [x] is upsampled from the binary image data a _L [y _L ] [x] as shown in Equation 3 below.

a _H [y _H ] [x] = a _L [y _L ] [x]

Here, the relationship between y _H and y _L may be expressed as in Equation 4 below.

if ((y _H % n) <2 x (n-m)) {

y _L = (y _H % n) / 2 + mx (y _H / n)

} else {

y _L = (y _H % n) + mx (y _H / n) + (m-n)

}

This means that if the remainder of y _L divided by n is less than 2 x (n-m), then y in the L hierarchy is (y _H % n) / 2 + mx (y _H / n). The y value is set to (y _H % n) + mx (y _H / n) + (m-n). The horizontal resolution conversion may be performed in the same manner. The calculation of the y value as described above at the time of upsampling and downsampling can be achieved only when the ratio of n / m is greater than 1 and less than or equal to two.

2B illustrates a sampling data selection method for inter-layer resolution conversion when n / m is larger than two. Here, if the BAB 211 in the upper layer L + 1 and the BAB 212 in the lower L layer change the resolution in the vertical direction n by m, the vertical direction of the L + 1 layer The n samples 214 correspond to the m samples 215 in the vertical direction of the L layer. In this case, only 2m data 216 of the n samples of the L + 1 th layer are sampled 213 to obtain m data samples 218 of the L th layer and the remaining (n-2m) data of the L + 1 th layer. 217 is ignored. In consideration of the vertical resolution conversion, the remaining n-2m data 217 of the L + 1th layer is ignored. Considering the vertical resolution conversion, binary image data a _H [y _H ] [x] in the L + 1 layer is converted to binary image data a _L [y _L ] [x] in the _L layer. The relationship between y _L and y _H when downsampled as shown in 1 is expressed by Equation 5 below.

y _H = 2 x (y _L % m) + nx (y _L / m) + 1

In contrast, when upsampled from binary image data a _L [y _L ] [x] to binary image data a _H [y _H ] [x] as shown in Equation 3, the relationship between y _H and y _L is Can be represented as:

if ((y _H % n) <2 xm) {

y _L = (y _H % n) / 2 + mx (y _H / n)

} else {

y _L = mx (y _H / n) + m-1

}

The horizontal resolution conversion may be performed in the same manner.

As a method different from those described in FIGS. 2A and 2B, a method of changing the resolution of each step continuously may be used by dividing into several steps as shown in Equation 7 below.

here

, , , Satisfies. The change in resolution at each step uses equations (2) and (4).

3 shows that when n and m are 2 and 1, the binary image 301 of the upper layer becomes the binary image 302 of the lower layer through the sampling process 303 as shown in FIGS. 2A and 2B. . In this case, since 2m-n is '0', step 204 in FIG. 2 is omitted.

4 illustrates a method of setting a VOP consisting of a rectangular boundary region surrounding shape information as an input image when changing the resolution of a binary image. The rectangular area 401 surrounding the object in the binary image with the minimum area is obtained. In order to ensure that the position of the restored boundary area after the resolution conversion again matches the original boundary area, the area is extended to fit the sampling unit 403 to obtain a new area 404. Here, the distance d _H (d _V ) in the horizontal (vertical) direction of the sampling unit is m for the base layer and n for the upper layer if the resolution is changed from m to n from the base layer to the upper layer in the horizontal (vertical) direction. Become. This is to restore the original position through the reverse process because the position value changed during the resolution conversion does not coincide with the integer grid. The starting point of the object's boundary 401 to match the sampling unit S ′ (y ′ _s , x ′ _s ) (402) S (y _s , x _s ) Go to 405. Here, y _s and x _s may be represented by Equation 8 below.

y _s = y ′ _s- (y ′ _s % (2d _V ))

x _s = x ′ _s- (x ′ _s % (2d _H ))

The multiplication of d _H and d _V by 2 is to ensure that the four Y values in the YUV texture values match exactly one UV value in the 4: 2: 0 image format. When the horizontal and vertical lengths of the boundary after the starting point are corrected are W and H, respectively, the values dW and dH to be compensated for matching the right and lower borders to the sampling unit are calculated as in Equation 9 below.

dW = d _H- (W% d _H )

dH = d _V- (H% d _V )

here (W% d _H ) Wow (H% d _W ) Is '0', dW and dH are also '0'. Since the encoding is performed in units of macroblocks, the left and top start points 405 are fixed and the right and bottom boundaries are extended to be divided by the block size M x N units, thereby finally obtaining an input 406 of the multilayer encoder. A summary of the VOP formation process of FIG. 4 is as follows.

1) Input binary shape information

2) Extract the smallest rectangular bounds around an object

3) Matching rectangle boundaries to the sampling grid

4) Extend the size of the rectangular boundary in M x N block size units around the starting point

5 illustrates a problem caused by the position of the VOP in the process of changing the inter-layer resolution of the VOP. The VOP 501 of the input image of the upper layer may be down-sampled 503 in the vertical direction in the same manner as described with reference to FIGS. 2 and 3 to obtain the VOP 504 of the lower layer. Upsampling this back to the original resolution 505 yields a VOP 506 of the resulting image. In the lower layer, a scan line value 509, 510, 511, or 512 downsampled at the input VOP 501 of the upper layer corresponds to a value in the VOP 504 of the lower layer. Encoded by The value 502 between these scan lines is encoded using the correlation of sample values above and below the value. However, in the example of FIG. 5, the scan line values 509, 510, and 511 encoded in the lower layer become 510, 511, and 512 of the VOP 506 of the upsampled image. This results from a mismatch between the starting point 507 of the input layer VOP 501 of the upper layer and the starting point 508 of the VOP 506 restored to the upsampling of the downsampled VOP 504. This means that although the scan lines 513 to be encoded in the upper layer are not all zeros and are located at the boundary of the VOP 501, the scan lines 509 to be sampled are all zeros so that the start point of the down-sampled VOP 504 is changed. Because. This problem may cause an error in encoding.

FIG. 6 illustrates the generation of a VOP 602 to be referred to by a higher layer in order to solve a possibility of error occurrence in encoding as shown in FIG. 5. Move the upsampled VOP 601 by the start offset 603 of the start point of the upper layer VOP and the start point of the upsampled VOP 601 from the lower layer to generate the VOP 602 to be referred to in the upper layer. After that, the data is stored as long as the range 605 does not exceed the size of the referenced VOP. The other area is set to 0 (604). The reference VOP 602 matches the starting point and size of the original VOP of the upper layer.

7 (a) and 7 (b) show types of scan lines and types of pixels to be encoded in an upper layer. When the resolution change from the lower layer to the upper layer is m versus n, FIG. 7A illustrates a case where n / m is smaller than 2 and FIG. 7B illustrates a case where the resolution is larger than 2. The scan lines 701 and 711 are data lost in the downsampling process to the lower layer. The scan lines 701 and 711 should be restored after encoding using the correlation between the scan lines 702 and 712 restored from the lower layer. The scan lines 703 are copied from the upper layer to the lower layer as they are and are encoded together in the lower layer. Samples encoded in the scanning line can be divided into two types as follows.

1) Transitional sample data (TSD): The sample values of the upper and lower scanning lines of the pixel to be encoded are different (704, 714).

2) Exception sample data (ESD): The sample values of the upper and lower scan lines of the pixels to be encoded are the same, and the pixel values of these two scan lines are different from the pixels to be encoded (705, 715).

The pixels in the scanning line 713 are encoded without using a multi-stage encoding technique without reference in the lower layer. In general, in the case of shape information, when the upper and lower scanning lines are the same, the pixels to be encoded are also the same. Therefore, in the structure of encoding 16x16 binary image macroblock, that is, BAB unit, if the sample values of the upper and lower scan lines of the pixel to be encoded are the same and there are no exception sample data values different from the pixel to be encoded, Since the encoding is performed only when the sample values of the upper and lower scan lines are different (704, 714), the number of pixels to be encoded is substantially reduced, thereby improving the encoding efficiency. If at least one exception sample data value exists in the BAB (705, 715), all pixels must be encoded. However, the rate at which such cases occur is extremely low, resulting in gain in the entire multilayer coding. Depending on the existence of an exception sample data value in BAB, BAB is encoded into two types as follows.

1) Transitional BAB (TBAB): There is no exception sample data ESD in the BAB.

2) Exception BAB (EBAB): At least one exception sample data ESD is present in the BAB.

When the sample values in FIGS. 7A and 7B are a _H [y _H ] [x] and the sample values of the lower layer are a _L [y _L ] [x], the scanning line 701 encoded in BAB , 711, and the type of the pixel in the scanning line to be encoded, and the BAB of the upper layer, proceed as follows. Only the change in resolution in the vertical direction is considered here.

1) Find the position y _H of the sample to be encoded by a _H [y _H ] [x]. Where y _H is (y _H % n) <2 x (n-m) and ((y _H % n)% 2) must satisfy the condition '0'.

2) The pixel values of the upper and lower scanning lines of the pixel to be encoded are obtained at the pixel a _H [y _H ] [x] to be encoded on the upper layer scanning line corresponding to the position y _H of the sample, and is exclusively expressed as in Equation 10 below. Perform an exclusive OR (XOR).

[Equation 10]

XOR [y _H ] [x] = a _L [y _L -1] [x] a _L [y _L +1] [x]

Where y _L = (y _H % n) / 2 + mx (y _H / n) Denotes an exclusive OR operation.

3) If the value of Equation 10 is '1', the value to be encoded a _H [y _H ] [x] becomes the transition sample value TSD and proceeds to 4) below, and the value of Equation 10 is '0'. And the value to be encoded a _H [y _H ] [x] is not equal to the value a _L [y _L ] [x] referenced in the lower layer, the pixel value is called the exception sample value ESD and the BAB containing ESD is excluded. (Exeptioanl BAB) and the process ends.

4) If (y _H , x) is the last pixel of the scan line to y _H , the process proceeds to 1) above, and otherwise proceeds to 2).

5) Since there is no exception sample data ESD, set BAB as transition BAB and terminate the process.

All the pixels in the scanning line 713 are encoded without reference in the lower layer.

8A and 8B show an arrangement of context information of surrounding pixels to be used when a sample to be encoded in a BAB to be encoded is encoded by a binary arithmetic coder (BAC). Encoding using BAC is coded through the following process.

1) Using the pixel peripheral value C _k to be encoded, a 7-bit context is obtained using Equation 11 below.

2) The probability that the pixel corresponding to the context value is '0' is obtained from 128 probability tables.

3) The BAC is encoded by using the probability distribution obtained in step 2).

In fact, when coding from a lower layer to a higher layer, vertical sampling and horizontal sampling are independently encoded. FIG. 8A illustrates a case in which upsampling in the horizontal direction is performed first and upsampling in the vertical direction occurs first when encoding from the lower layer to the upper layer (803). At this time, the context information is obtained in the same array as in the case where the pixel to be encoded is the scanning line in the vertical direction (801), and the context information is obtained in the same arrangement as in the case of the scanning line in the horizontal direction (802). The context information is calculated at the resolution of the upper layer. If the number is not displayed, the context information is an unknown pixel value during decoding. In the case of FIG. 8B, the horizontal reconstruction 813 occurs after the vertical reconstruction. At this time, the context information is obtained in the same arrangement as that of the horizontal scanning line (811) and the context information is obtained in the same arrangement as the case of the horizontal scanning line (812).

9 illustrates a flowchart of an entire encoding process, in which a rectangular boundary region VOP including an input object is input as an input (901) and downsampled for encoding in a lower layer (902). Here, VOP is obtained using the method as shown in FIG. Down sampling can be done at any rate using the method shown in FIGS. 2A, 2B and 3. The down sampled image is losslessly coded with any shape encoder. The encoded image of the lower layer is upsampled (904) and used for encoding of the upper layer. At this time, the starting point of the reference VOP coincides with the starting point of the original VOP of the upper layer (905). This process is performed by the method as described in FIG. After selecting the scan lines to be encoded in the upper layer, the reference scan lines are obtained from the reference VOP, and then the distinction is made whether the BAB is an exception BAB or a transition BAB (906). It is determined whether there is exception information in the BAB (907), and if it exists, all bi-scanning lines are encoded by using the BAC. Otherwise, only pixels having exclusive OR information between adjacent scan lines are encoded by the BAC.

According to the present invention, since the resolution conversion ratio between layers is free, a reconstructed image can be obtained at a resolution desired by the user, and the sequential transmission process in the multi-layer encoding process is simplified and the compression efficiency is increased.

Claims (23)

In a multi-layer binary shape encoder for sequentially obtaining shape information of various resolutions on one bitstream, If the resolution conversion ratio of the lower L layer and the higher layer L + 1 layer having a higher resolution is not greater than 2, n blocks of the number of samples are respectively equal to the resolution conversion ratio of the L + 1 layer and the L layer. Dividing by m, and sampling only 2 (nm) data of the L + 1 layer to correspond to (nm) of the L layer, and remaining 2m-n data of the L + 1 layer in the L layer without being sampled. Matching arbitrary resolution multilayer binary shape coder. The method of claim 1, wherein downsampling from the L + 1 layer to the L layer comprises: The relationship between the vertical coordinate value y _H of the vertical coordinate y _L and the pixel coordinates (x, y _H) of said L + 1 layer of the pixel coordinates (x, y _L) of the L layer, divided by the y _L in m Y _H is 2 x (y _L % m) + nx (y _L / m) + 1 when the remainder is less than (n-m), otherwise y _H is (y _L % m) + nx ( y _L / m) + (n-m). (Where '%' is the operator that takes the remainder of the division and '/' is the operator that takes only integer values when dividing) The method of claim 1, wherein upsampling from the L + 1 layer to the L layer is performed. About the relationship of y _H y _L and the pixel coordinates (x, y _H) of said L + 1 layer of the pixel coordinates (x, y _L) of the L layer, the remaining value of 2 divided by the y _H with n (nm ) the y _L is smaller than _{(y H% n) / 2} + mx (y H / n) in, and the y _L or else _{(y H% n) + mx} (y H / n) + (m - n) An arbitrary resolution multilayer binary shape encoder characterized by the above-mentioned. (Where '%' is the operator that takes the remainder of the division and '/' is the operator that takes only integer values when dividing) In a multi-layer binary shape encoder for sequentially obtaining shape information of various resolutions on one bitstream, If the resolution conversion ratio of L layer with lower resolution and L + 1 layer with higher resolution is greater than 2, n and each block of samples are equal to the resolution conversion ratio of L + 1 layer and L layer, respectively. dividing by m, sampling only 2m of the n samples of the L + 1 layer to obtain m data samples of the L layer, and ignoring the remaining n-2m data of the L + 1 layer to correspond to the L layer. And a random resolution multilayer binary shape coder. The method of claim 4, wherein downsampling from the L + 1 layer to the L layer is In the relationship between the vertical coordinate y _L of the pixel coordinates (x, y _L ) of the L layer and the vertical coordinate y _H of the pixel coordinates (x, y _H ) of the L + 1 layer, y _H is 2 x (y _L % m) + nx (y _L / m) + 1, multi-layer binary shape coder, otherwise set y _H to (y _L % m) + nx (y _L / m) + and (n-m), an arbitrary resolution multi-layer binary shape coder. (Where '%' is the operator that takes the remainder of the division and '/' is the operator that takes only integer values when dividing) The method of claim 4, wherein upsampling from the L + 1 layer to the L layer comprises: About the relationship of y _H y _L and the pixel coordinates (x, y _H) of said L + 1 layer of the pixel coordinates (x, y _L) of the L layer, the remainders obtained by dividing the y _H to n is less than 2m Multi-layer binary shape with y _L as (y _H % n) / 2 + mx (y _H / n), otherwise y _L as mx (y _H / n) + m-1. Encoder. (Where '%' is the operator that takes the remainder of the division and '/' is the operator that takes only integer values when dividing) In a multi-layer binary shape encoder for sequentially obtaining shape information of various resolutions on one bitstream, If the resolution conversion ratio of L layer with lower resolution and L + 1 layer with higher resolution is larger than 2, the resolution layer is n: m when the resolution conversion ratio of L + 1 layer and L layer is n: m. An arbitrary resolution multi-layer binary shape encoder characterized by dividing into multiple layers having a resolution difference of not more than twice. The method of claim 7, wherein downsampling between each divided layer is When the upper layer is called an N + 1 layer and the lower layer is called an N layer, and the sampling blocks of N + 1 and N are divided by j and i, which are equal to the respective conversion ratios, the pixel coordinates (x, when the smaller than i) - the vertical coordinate of y _L) y _L, and the remainder obtained by dividing a, y _L to the relationship of the vertical coordinate y _H to i of the pixel coordinates (x, y _H) of said L + 1 layer (j Let y _H be 2 x (y _L % i) + jx (y _L / i) + 1, otherwise y _H is (y _L % i) + jx (y _L / i) + (j-i) An arbitrary resolution multi-layer binary shape encoder characterized by the above-mentioned. (Where '%' is the operator that takes the remainder of the division and '/' is the operator that takes only integer values when dividing) The method of claim 8, wherein upsampling from the N + 1 layer to the N layer is performed. The pixel coordinates of the N layer (x, y _L) of y _L and the pixel coordinates for the relationship of _{y H (x, y H)} , the remaining value of 2 divided by the y _H to j (ji of said N + 1 layer ) small, the _{y L (y H% j)} / 2 + ix (y to the _H / j), otherwise, the _{y L (y H% j)} + ix (y H / j) + (i than - j) An arbitrary resolution multi-layer binary shape encoder characterized by the above-mentioned. (Where '%' is the operator that takes the remainder of the division and '/' is the operator that takes only integer values when dividing) In a multi-layer binary shape encoder for sequentially obtaining shape information of various resolutions on one bitstream, When the resolution conversion ratio of the lower L layer with lower resolution and the upper L + 1 layer with higher resolution is a predetermined ratio of m to n, when the resolution of the binary image is converted, the upper layer between L + 1 and the lower layer L is used. In order to eliminate the loss of reference data, a method of setting a video object plane (hereinafter referred to as VOP), which is a rectangular boundary area surrounding shape information, A first step of inputting binary shape information; Obtaining a rectangular area S 'surrounding the object in the binary shape with the minimum area; In order to ensure that the location of the reconstructed boundary region after the resolution conversion of the boundary of the rectangular region S matches the object boundary of the original information, the square boundary is fitted to a sampling grid divided for binary shape information sampling. Obtaining a rectangular area S that extends the boundary of the second step; And And obtaining a rectangular boundary S ″ having an area of a unit sampling block M × N size to be encoded starting from the starting point of S. 4. The method of claim 10, When the starting point of S '(x' _s , y ' _s ), the starting point of S (x _s , y _s ), the horizontal axis length of the sampling grid is d _H , the vertical axis length is d _V , the horizontal coordinate x _s is x, and 'in _s x' _s for subtracting the remainder divided by the 2d _H value, the vertical coordinate y _s is y _'s in y' _s arbitrary resolution multi-layered binary shape encoder, it characterized in that the remainder divided by the 2d _V . 12. The length of the horizontal and vertical axes of S is according to claim 11, When the length of the lateral, longitudinal axis of the S 'La each H, W, the H in the d from _H, remove the remaining (W% d _H) divided by the W by d _H value and the d _V a d _V A random resolution multi-layer binary shape encoder, characterized in that the length obtained by subtracting the remainder (H% d _V ) is added to H and W, respectively. In a multi-layer binary shape encoder for sequentially obtaining shape information of various resolutions on one bitstream, A rectangular boundary region surrounding an object in the lower layer of the binary image (hereinafter referred to as VOP) when the resolution conversion ratio of the lower resolution L layer and the higher resolution higher layer L + 1 layer is m to n. The starting point of the upsampled VOP coincides with the starting point of the VOP in the upper layer. Creating a reference VOP having the same position and size as the upper layer VOP; Moving the upsampled VOP by the amount of change from the lower layer to the start point of the upsampled VOP based on the start point of the upper layer VOP; Copying contents of a range of contents of the moved VOP not exceeding the reference VOP; And And allocating a remaining region of the reference VOP region, in which the contents of the moved VOP is not copied, to '0'. In a multi-layer binary shape encoder for sequentially obtaining shape information of various resolutions on one bitstream, When the ratio L / 1 of lower resolution and higher layer L + 1 of higher resolution is a predetermined ratio of m to n, and the ratio of n / m is 2 or less, the type of scan line in the upper layer is A scan line to be restored by decoding in an upper layer; A reference scan line for obtaining a scanning line of a binary image of an upper layer by upsampling when the resolution changes from the lower layer; And An arbitrary resolution multi-layer binary shape encoder characterized by being divided into scan lines that are copied as is from the lower layer. The method of claim 14, In the structure of encoding binary image information in units of a macro alpha block (BAB), when the sample values of the upper and lower scanning lines of the pixel to be encoded are the same in the BAB, but there are exception sample data different from the pixel values to be encoded. And encoding all pixels in the BAB, and performing encoding only on pixels when the upper and lower scanning lines are different when the exceptional sample data does not exist in the BAB. In a multi-layer binary shape encoder for sequentially obtaining shape information of various resolutions on one bitstream, When the ratio L / 1 of lower resolution and upper layer L + 1 of higher resolution L + 1 is a predetermined ratio of m to n, and the ratio of n / m is 2 or more, the type of scan line in the upper layer is A scan line to be restored by decoding in an upper layer; A reference scan line for obtaining a scanning line of a binary image of an upper layer by upsampling when the resolution changes from the lower layer; And An arbitrary resolution multi-layer binary shape encoder characterized in that the upper layer is divided into scan lines that are encoded as they are without reference to pixel values of the lower layer. The method of claim 16, In the structure of encoding binary image information in units of a macro alpha block (BAB), when the sample values of the upper and lower scanning lines of the pixel to be encoded are the same in the BAB, but there are exception sample data different from the pixel values to be encoded. And encoding all pixels in the BAB, and performing encoding only on pixels when the upper and lower scanning lines are different when the exceptional sample data does not exist in the BAB. In a multi-layer binary shape encoder for sequentially obtaining shape information of various resolutions on one bitstream, The resolution conversion ratio of the lower L layer and the higher layer L + 1 layer having a higher resolution is m to n is a predetermined ratio, and the binary values a _L , a _H , and vertical coordinates y _{L, of} pixels for each layer. Binary image information when the sample value of the upper layer is a _H [y _H ] [x] and the sample value in the lower layer L is a _L [y _L ] [x] for y _H and the horizontal coordinate x. In the structure of encoding a in a macroblock (BAB) unit, a method of classifying a type of BAB for encoding efficiency may include: the position y of the sample to be encoded that satisfies the condition that (y _H % n) <2 x (n-m) and ((y _H % n)% 2) in _H [y _H ] [x] A first step of finding _H (where '%' is an operator that takes the remainder of the division and '/' is an operator that takes only the integer result when dividing); The upper that corresponds to the location of the sample y _H layer scan line pixels to be encoded in the above _H a location of the pixel to be coded in the _[H y] [x], obtaining a pixel value below the scanning line, as shown in Equation 10 [Equation 10] XOR [y _H ] [x] = a _L [y _L -1] [x] a _L [y _L +1] [x] Where y _L = (y _H % n) / 2 + mx (y _H / n) Means an exclusive OR operation.) A second step of performing an exclusive OR (XOR); A third step of setting the value a _H [y _H ] [x] to be encoded as a transition sample value (TSD) having different sample values of the upper and lower scan line pixels when the result value of Equation 10 is '1'; If the result value of Equation 10 is '0' and the value to be encoded a _H [y _H ] [x] is not equal to the value a _L [y _L ] [x] referenced in the lower layer, the pixel value is excluded. A fourth step of ending with an exception BAB (ESD) and an ESD BAB (Exeptioanl BAB); a fifth step of repeating from the first step if (y _H , x) is the last pixel of the scan line to y _H ; And a sixth step of terminating the process, making the BAB the transition BAB in which no exception sample data exists, since no exception sample data ESD is present. Receiving binary image information of a predetermined size surrounded by a rectangular boundary region (VOP); Downsampling the image information for encoding in a lower layer when the resolution ratio of the upper layer and the lower layer is n to m; Lossless encoding the down-sampled image with a predetermined shape encoder; Upsampling the encoded image for use in encoding of a higher layer and generating a reference VOP coinciding with a start point of an original VOP of the higher layer; After selecting the type of the scan line to be encoded in the upper layer, the reference scan line is obtained from the reference VOP, and whether there is exception information in the sample unit block BAB to be encoded is a pixel having a different pixel value from the same upper and lower pixel values. Distinguishing; If there is exception information in the BAB, the pixel is encoded using a binary arithmetic coder for all the pixels in every other scan line. If there is no exception information, only the pixels with exclusive ORs between adjacent scan lines is 1 And BAC encoding according to the BAC. The method of claim 19, wherein the downsampling comprises: The relationship of the vertical coordinate y _H of the pixel coordinates (x, y _H) of the vertical coordinate y _L and an upper layer of the pixel coordinates (x, y _L) of the n / m is 2 or less when the lower layer, y _L If y is m less than (n-m) then y _H is 2 x (y _L % m) + nx (y _L / m) + 1 otherwise y _H is (y _L % m ) + nx (y _L / m) + (n-m) multi-layer binary shape coding method characterized in that. (Where '%' is the operator that takes the remainder of the division and '/' is the operator that takes only integer values when dividing) The method of claim 20, wherein the downsampling, When n / m is 2 or more, in the relationship of y _H of the pixel coordinates (x, y _L) y _L and the pixel coordinates (x, y _H) in the upper layer of the lower layer, y _H is 2 x (y _L % m) + nx (y _L / m) + 1, a multi-layer binary shape coder, otherwise y _H (y _L % m) + nx (y _L / m) + (n-m), wherein the multi-layer binary shape coding method of arbitrary resolution. The method of claim 19, wherein the upsampling, n / m at this time is less than 2, the relationship of y _H of the pixel coordinate pixel coordinates (x, y _H) in the y _L and an upper layer of the (x, y _L) of said lower layer, dividing y _H with n If the remaining value is less than 2 (nm) then y _L is (y _H % n) / 2 + mx (y _H / n), otherwise y _L is (y _H % n) + mx (y _H / n) + (m-n), wherein the multi-layer binary shape coding method of arbitrary resolution. (Where '%' is the operator that takes the remainder of the division and '/' is the operator that takes only integer values when dividing) The method of claim 19, wherein the upsampling is When n / m is 2 or more, and the other for the relationships of y _H y _L and the pixel coordinates (x, y _H) in the upper layer of the pixel coordinates (x, y _L) of said lower layer, dividing y _H with n If y is less than 2m, y _L is (y _H % n) / 2 + mx (y _H / n), otherwise y _L is mx (y _H / n) + m-1 Resolution multi-layer binary shape coding method. (Where '%' is the operator that takes the remainder of the division and '/' is the operator that takes only integer values when dividing)