KR20060133885A - Method for upsampling a video data - Google Patents

Abstract

A method for upsampling video data is provided to generate new pixel data between pixel points while reducing blurring of upsampled image data and ringing that an image contour appears. Pixel data are added through interpolation between pixel data inside and outside a pixel data group. A part of the pixel data group to which the pixel data are added, is selected for performing first discrete cosine transform. At least one adjacent pixel data is included in the pixel data group for performing second discrete cosine transform. Data transformed by the first discrete cosine transform and the second discrete cosine transform are obtained to be separated or partially overlapped by bands for performing inverse discrete cosine transform.

Description

How to upsample video data {Method for upsampling a video data}

1 shows 6-tap FIR interpolation, which is one of the methods of upsampling,

2 illustrates a method of upsampling through zero-padding in the frequency domain,

3 illustrates a configuration block of an apparatus for upsampling image data according to an embodiment of the present invention.

4 exemplarily illustrates a process of upsampling a portion of data of one row or column of an image block by the apparatus of FIG. 3 according to an embodiment of the present invention;

5 exemplarily illustrates a process of upsampling a part of data of one row or column of an image block by the apparatus of FIG. 3 according to another embodiment of the present invention.

<Description of the symbols for the main parts of the drawings>

31: interpolation filter 32, 34: DCT section

33,35: windowing filter 36: combiner

37: IDCT Division

The present invention relates to a method of upsampling image data to add new pixel data between pixel data.

The scalable video codec (SVC) method encodes a video signal at the highest quality and decodes a partial sequence of the resulting picture sequence (a sequence of intermittently selected frames throughout the sequence). Even if it is used, it is a way to enable a low-quality video representation.

However, as described above, a picture sequence encoded in a scalable manner can be represented by a low quality image only by receiving and processing only a partial sequence. However, when the bit rate is lowered, the quality deterioration is large. In order to solve this problem, a separate auxiliary picture sequence for a low data rate, for example, a low picture sequence and a low picture sequence per second may be provided hierarchically. However, since the lower layer and the upper layer encode the same video signal source, there is redundant information (redundancy) in the encoding signals of both layers.

Therefore, in order to increase the coding rate of a specific layer encoded by the SVC scheme, a block in an arbitrary video frame of a lower layer is enlarged, and on the basis of the enlarged block, A block in the video frame is made into error data, that is, residual data.

Since the enlarged block is not transmitted to the decoder, the decoder must also enlarge and use the corresponding block of the lower layer in order to decode the encoded macro block as described above. In addition, not only for encoding the macroblock of the internal mode described above, but also for performing the inter-layer residual data prediction operation, the block of the lower layer should be enlarged.

As such, when a plurality of layers having different picture sizes (or resolutions) are provided as an encoding stream, an enlargement of an image block is required during encoding and decoding, and new pixel data is generated between pixel data in order to enlarge the image block. Upsampling is performed.

Also, in performing motion estimation for an arbitrary macro block, the target block is shifted on adjacent frames (or slices in which frames are divided) by 1/4 pel, and the pixel data of the target block is shifted. Since the current macroblock data is compared, the pixel value at the position of the non-integer pel on the frame (hereinafter referred to as the non-pixel point) is determined by the position of the integer pel (hereinafter referred to as the 'pixel point'). It is obtained from the pixel value in (). Thus, upsampling is also used at this time.

An object of the present invention is to provide a method of upsampling image data for generating new pixel data between pixel points.

Another object of the present invention is to provide a method of upsampling video data which simultaneously reduces blurring and ringing.

In order to achieve the above object, the present invention, in upsampling an input pixel data group (N pixel data), adds pixel data through interpolation between pixel data within and outside the data group, and adds pixel data. Select a part of the data group and perform a first discrete cosine transform; together with at least one adjacent pixel data in the pixel data group, a second discrete cosine transform is performed, and the converted data and the first discrete cosine The data is divided into bands or partially overlapped to perform discrete inverse transform.

In one embodiment according to the present invention, the interpolation interpolates pixel data of the pixel data group and the adjacent pixel data group to form a group of 4N pixel data, and the first discrete cosine transform is 4N. In 2 data groups, 2N pixel data spatially corresponding to the input pixel data group and one pixel data adjacent thereto are performed.

In an embodiment according to the present invention, two windowing filters are used to partially overlap each band, and the coefficient sets, {a _k } and {b _k }, of each windowing filter are respectively {a ₀ = 0, a ₁ = 0, .. a _N-1 = p _I-1 , a _N = p _I , a _{N + 1} = p _{I + 1} , .., a _2N = 1} (where p _I Is a constant that satisfies the condition 0 <p _I <p _{I + 1} <1), and (b ₀ = 1, b ₁ = 1, .., b _N-2 = q _I-1 , b _N-1 = q _I } (where q _I is a constant that satisfies the condition 1> q _I-1 > q _I > 0).

In one embodiment according to the present invention, 6-tap FIR interpolation is used for the interpolation.

In one embodiment according to the present invention, a type-1 discrete cosine transform is used for the discrete cosine transform.

1 is 6-tap FIR interpolation, which is one of the methods of upsampling. In this method, one pixel value (F '(3)) is divided into three pixel values [{F (3), F (2), F (1)}, each of its adjacent left and right (or up and down) values. {F (4), F (5), F (6)}] is multiplied by a predetermined weight (e.g., 20, -5,1 starting from adjacent pixel points), and then summed Find by dividing by.

However, the up-sampled image block by this method may have a blurring effect and a decrease in peak signal to noise ratio (PSNR).

FIG. 2 is an upsampling method with zero-padding in the frequency domain, and is an upsampling method that improves negative effects due to 6-tap FIR interpolation.

This method performs a discrete cosine transform (DCT) using one adjacent pixel value to the pixel group of pixel number N to be enlarged (S21), and converts the converted N + 1 DCT coefficients. After N zeros are padded in the high band (S22), IDCT (inverse DCT) is performed on the DCT coefficients (2N + 1) (S23). By deleting one value 201 from the pixel values (2N + 1) obtained after performing IDCT, a pixel data group having a 2N pixel size magnified 2 times is finally obtained.

However, according to this method, screen blurring does not occur, but a ringing effect occurs in which an outline of an image appears in an enlarged image block.

Hereinafter, embodiments of the present invention, which simultaneously reduce screen smearing and ringing, will be described in detail with reference to the accompanying drawings.

3 is a block diagram illustrating a configuration of an apparatus for upsampling image data according to an embodiment of the present invention.

The upsampling device of FIG. 3 includes a filter 31 for interpolating an input pixel data group 30 in a spatial domain, and a first DCT unit for DCT transforming a part of output pixel data of the filter 31 ( 32), a first windowing filter 33 for windowing the DCT coefficients by multiplying each of the converted data (hereinafter, referred to as 'DCT coefficients') by the filtering coefficients, and the pixel data group 30 of the inputted pixel data group 30. A second windowing filter for multiplying the second DCT 34 by DCT transforming the data including at least the pixel data and the DCT coefficients converted by the second DCT 34 to filter out the DCT coefficients for windowing the DCT coefficients. (35), a combiner (36) for combining the output coefficients of the first and second windowing filters (33) and the second windowing filter (35) by frequency band in the DCT domain, and IDCT for the combined DCT coefficients. It is configured to include an IDCT unit 37 to perform. The interpolation filter 31 may be the 6-tap FIR interpolation filter described above, but the present invention is not limited to the 6-tap FIR interpolation filter, and the non-pixel point data between the pixel point and the pixel point in the spatial domain is used. Any filter can be used to make it.

FIG. 4 is a diagram schematically illustrating a process of the apparatus of FIG. 3 performing a part of data of one row or column of an image block, wherein the number of pixels to be upsampled is four. However, the present invention is not limited to the number of pixels, and it is possible to upsample any number N of pixel data in the manner described below.

First, the interpolation filter 31 is configured to be left or right (or upper) with respect to N input pixel data groups to be enlarged in any one row of pixel data (this is referred to as 'low resolution data' for convenience). Or by adding the same number of pixel data adjacent to the lower side (the example shown in FIG. 4 uses the pixel data adjacent to the right side). An interpolation is performed to create 1 / 2-pel pixel data for 2N pixel points. It performs (S41). The 1DCT unit 32 takes 2N + 1 pixel data with respect to 4N pixel data generated by this interpolation (this is referred to as 'high resolution data' for convenience), and selects a DCT, for example, type-1 DCT. It performs (S42). The 2N + 1 taken by the 1DCT unit 32 is 2N corresponding to N input pixel data to be enlarged spatially in a group of interpolated 4N pixel data, and one pixel closest thereto. Data.

The first windowing filter 33 multiplies the 2N + 1 DCT coefficients obtained by the first DCT unit 32 by the following filtering coefficient {a _k } and then only N DCT coefficients belonging to the high band. {F (N + 1), ..., F (2N-2), F (2N-1), F (2N)} are obtained (S43).

{a _k } = {a ₀ = 0, a ₁ = 0, .., a _N = 0, a _{N + 1} = 1, a _{N + 2} = 1, .., a _2N-1 = 1, a _2N = 1}

Meanwhile, the second DCT unit 34 adds one pixel data adjacent to the left or right side (or the upper side or the lower side) to the N pixel data groups in the pixel data of any one row of the image block (FIG. The example of 4 uses pixel data adjacent to the right side.), DCT, for example, type-1 DCT is performed to perform N + 1 DCT coefficients {F (0), F (1), .., F ( N)} is obtained (S44). The second windowing filter 35 multiplies the N + 1 DCT coefficients by a filtering coefficient {b _k } to multiply the N + 1 DCT coefficients to add a component to the low band for combination with the DCT coefficients of the previously obtained high band component. N DCT coefficients of appropriately adjusted sizes are obtained (S45).

In the embodiment of FIG. 4 according to the present invention, the filtering coefficient {b _k } is set to {b ₀ = 1, b ₁ = 1, .., b _N = 1} for windowing.

The filtering coefficient {b _k } is for using the N + 1 DCT coefficients converted by the second DCT 34 without changing their size. In this case, the second windowing filter 35 is not used. The N + 1 DCT coefficients converted by the second DCT 34 may be applied to the combiner 36 without change.

The combiner 36 comprises N high-pass component DCT coefficients F (N + 1),..., F (2N-2), F (2N-1), obtained from the first windowing filter 33. F (2N)} and the N + 1 low-pass DCT coefficients {F (0), F (1),... Obtained from the second windowing filter 35 (or the second DCT part 34). F (N)} is arranged for each band to form 2N + 1 DCT coefficients {F (0), F (1), .., F (2N-1), F (2N)} (S46), The IDCT unit 37 performs IDCT on the 2N + 1 DCT coefficients to obtain 2N + 1 pixel data (S47), and discards the last pixel data 401. The discarded pixel data 401 is data spatially close to one pixel data added for the above discrete cosine transform S44 in both boundary data of 2N + 1 pixel data. As a result, the pixel data group {D (0), D (1), ..., D (2N-1)}, which is 2N magnified, is obtained.

In the 2N pixel data set {D (k)}, if pixel data of {D (i) | i = odd, i <2N} is taken, it becomes pixel data for 1 / 2-pel motion estimation.

In the embodiment of FIG. 4, the windowing filter is used to cut the high-frequency DCT coefficient obtained from the high resolution data and the low-frequency DCT coefficient obtained from the low resolution data so as not to overlap each other, but in another embodiment according to the present invention, The high and low DCT coefficients obtained from the high resolution data and the low resolution data may be partially overlapped.

5 is a diagram illustrating an upsampling process of image data according to an embodiment according to the present invention. In order to perform the embodiment of FIG. 5, the first windowing filter 33 of FIG. 3 has a filter coefficient {a _k } 501 set as follows.

{a _k } = {a ₀ = 0, a ₁ = 0, .. a _N-1 = p _I-1 , a _N = p _I , a _{N + 1} = p _{I + 1} , .., a _2N = 1} Here, 0 <p _I <p _{I + 1} <1, where p _I is a constant appropriately selected such that the quality of the upsampled image is partially or entirely the same or improved as the embodiment shown in FIG.

Similarly, the second windowing filter 35 also has a filter coefficient {b _k } 502 set as follows.

{b _k } = {b ₀ = 1, b ₁ = 1, .., b _N-2 = q _I-1 , b _N-1 = q _I } where 1> q _I-1 > q _I > 0 Q _I is a constant that is appropriately selected such that the quality of the upsampled image is at least partly or entirely equal to or improved from the embodiment shown in FIG. 4.

In proper selection of a _k and b _k , constants are selected that satisfy the condition of a _k + b _k = 1.

The apparatus of FIG. 3 uses the filter coefficients 501 and 502 set as described above to perform the same operations S41 to S47 as those described with respect to the embodiment of FIG. Image data of the pixel point is obtained. In addition, in the embodiment of FIG. 5, the output DCT coefficients of the first windowing filter 33 and the second windowing filter 35 together exist with values corresponding to the same position on the DCT domain (that is, Since the frequency bands have components overlapping with each other, the coupler 36 adds the values of both quantums to one DCT coefficient value and applies them to the IDCT unit 37 at a later stage.

The above-described upsampling device for video data may be implemented in a decoder mounted on a mobile communication terminal or a device for reproducing a recording medium as well as an encoder of a video signal.

It is to be understood that the present invention is not limited to the above-described exemplary preferred embodiments, but may be embodied in various ways without departing from the spirit and scope of the present invention. If you grow up, you can easily understand. If the implementation by such improvement, change, replacement or addition falls within the scope of the appended claims, the technical idea should also be regarded as belonging to the present invention.

At least one of the embodiments according to the present invention has the effect of simultaneously reducing the image blurring phenomenon and the appearance of the image contour of the upsampled image data.

Claims (9)

In the method of upsampling a group of pixel data, Adding pixel data through interpolation between pixel data in and out of the data group, selecting a part of the data group to which pixel data is added, and performing discrete cosine conversion; Including at least one adjacent pixel data in the pixel data group, and performing a discrete cosine transform, and performing the discrete cosine inversion by separating or partially overlapping the converted data and the data converted in the first step. Way. The method of claim 1, The pixel data number of the pixel data group is N, In step 1, interpolation is performed on 2N pixel data including N pixel data adjacent to the pixel data group to form a group of 4N pixel data, and 2N + 1 pixel data is taken from the data group to make a discrete cosine. To convert, In the second step, the adjacent pixel data is added to the pixel data group to perform discrete cosine transform. The converted N + 1 data and the 2N + 1 data converted in the first step are used for high-band. A method of constructing 2N pixel data by inversely transforming 2N + 1 data taking N data and then discarding one pixel data. The method of claim 1, The pixel data number of the pixel data group is N, In step 1, interpolation is performed on 2N pixel data including N pixel data adjacent to the pixel data group to form a group of 4N pixel data, and 2N + 1 pixel data is taken from the data group to make a discrete cosine. To convert, In the second step, the adjacent pixel data is added to the pixel data group, the discrete cosine transform is performed, and the data obtained by multiplying the converted N + 1 data by the first windowing filter coefficient {b _k } is performed. 2N + 1 data obtained by summing the 2N + 1 data transformed in step 1 and multiplied by the second windowing filter coefficient {a _k } for each frequency component, inversely transformed 2N + 1 data, and discarding one pixel data by discarding one pixel data. Method of constructing two pixel data. The method of claim 3, wherein The second windowing filter coefficient {a _k } is {a ₀ = 0, a ₁ = 0, .. a _N-1 = p _I-1 , a _N = p _I , a _{N + 1} = p _{I + 1} , .., a _2N = 1} and p _I is a constant satisfying the condition 0 <p _I <p _{I + 1} <1. The method of claim 3, wherein The first windowing filter coefficient {b _k } is {b ₀ = 1, b ₁ = 1, .., b _N-2 = q _I-1 , b _N-1 = q _I }, and q _I is 1 _{> q I-1> q I} > constant manner satisfying the conditions of 0. The method of claim 2 or 3, In the second step, a method of discarding one spatially adjacent data of one adjacent pixel data added for the discrete cosine transform from the data of both boundaries of the discrete cosine inversely transformed 2N + 1 pixel data. . The method of claim 2 or 3, In the step 1, the 2N pixel data spatially corresponding to the N pixel data group and one pixel data adjacent thereto are discrete cosine transformed from the group of 4N pixel data. The method of claim 1, The first step is to add pixel data between the pixel data through 6-tap FIR interpolation. The method of claim 1, Wherein the discrete cosine transform is type-1 DCT.

Images