KR0153984B1 - Method and apparatus for encoding video signal using classified vector quantization - Google Patents

Abstract

The present invention relates to an image compression method and apparatus for improving the compression coding efficiency of an image by using classification vector quantization. To this end, the present invention provides a cosine function for a differential image signal based on motion estimation and prediction. Converts DCT coefficients of 8 × 8 size in frequency domain, quantizes the DCT transform coefficients through nonlinear operation, and divides the quantized 8 × 8 DCT coefficient blocks into a plurality of regions having a predetermined size. Next, classification vector quantization is performed by matching the pattern of each region partitioned with each reference pattern in each codebook using a plurality of codebooks corresponding to the plurality of partitioned regions of the image encoder. Encoding efficiency of compression coding by restraining bit generation amount of image data as much as possible by encoding DCT transform coefficient The.

Description

Image Compression Method and Apparatus Using Classified Vector Quantization

1 is a block diagram of an image compression apparatus using classification vector quantization according to a preferred embodiment of the present invention.

2 is a diagram illustrating an example in which a DCT block of an 8x8 area is partitioned into 2x2 areas for matching with patterns in each codebook according to the present invention.

3 is a schematic block diagram of a conventional image encoder.

4 shows DCT coefficients before quantization and DCT coefficients after quantization according to a conventional method as an example.

FIG. 5 is a diagram for explaining a process of variable length encoding a quantized coefficient through zigzag scanning based on an encoding table.

* Explanation of symbols for main parts of the drawings

110: block cutting portion 121 to 136: pattern generating portion

121-2 to 136-2: Codebook 121-4 to 136-4: Pattern matching part

121-6 to 136-3: Output selector

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of compression coding video signals, and more particularly, to an image compression method and apparatus for compressing and coding video signals using classified vector quantization.

As is well known in the art, the transmission of discrete video signals can maintain better image quality than analog signals. When a video signal composed of a series of image frames is represented in digital form, a considerable amount of data must be transmitted, especially for high-definition television (HDTV). However, since the usable frequency range of the conventional transmission channel is limited, in order to transmit a large amount of digital data, it is necessary to compress the data to be transmitted and reduce its transmission amount. Due to the characteristics of these signals, they are encoded through different encoding techniques. In addition, it can be said that the compression method of the video signal, which generates a greater amount of digital data than the audio signal, is particularly important in such an encoding method.

Therefore, when transmitting a video signal, the transmitting side compresses and encodes the video signal using the spatial and temporal correlation of the video signal and then transmits the compressed and encoded video signal to the receiving side through the transmission channel.

On the other hand, as the various compression methods mainly used for encoding the image signal, the hybrid coding method combining the stochastic coding method and the temporal and spatial compression method is known to be the most efficient. Here, the present invention is a hybrid combining the temporal and spatial compression method. Related to the coding scheme.

Most of hybrid coding schemes, which are one of the efficient coding schemes described above, use motion compensation DPCM (differential pulse code modulation), two-dimensional discrete cosine transform (DCT), quantization of DCT coefficients, variable length coding (VLC), and the like. Here, the motion compensation DPCM determines a motion of the object between the current frame and the previous frame, and predicts the current frame according to the motion of the object to generate a differential signal representing the difference between the current frame and the prediction value. Such a method is, for example, Staffan Ericssn's Fixed and Adaptive Predictors for Hybrid Predictive / Transform Coding, IEEE Transactions on Communication, COM-33, NO.12 (1985, December), or A motion compensated Interframe from Ninomiy and Ohtsuka. Coding Scheme for Television Pictures, IEEE Transactions on Communication, COM-30, NO.1 (January, 1982).

More specifically, the motion compensation DPCM predicts the current frame from the previous frame according to the motion of the object estimated between the current frame and the previous frame. Here, the estimated motion may be represented by a two-dimensional motion vector representing the displacement between the previous frame and the current frame.

In general, there are various approaches to estimating the pixel displacement of an object, and these are generally classified into two types, one of which is a block-based motion estimation method and the other is a pixel-based motion estimation method.

On the other hand, in the block-based motion estimation of the motion estimation method, the optimal matching block is determined by comparing the blocks of the current frame with the blocks of the previous frame, and thereafter, the inter-frame displacements of the entire block with respect to the current frame to be transmitted. The vector (how much the block moved between frames) is estimated.

On the other hand, in the case of the pixel-by-pixel motion estimation method, since the displacement can be obtained for each pixel, the pixel value can be estimated more accurately, and the scale change (for example, the zooming (movement perpendicular to the image plane) is performed. zooming) also has the advantage of easy handling. However, in such a pixel-by-pixel motion estimation method, since a motion vector is determined for each pixel, it is virtually impossible to transmit all motion vectors to a receiver. Therefore, in the pixel-by-pixel motion estimation method, a motion vector for a selected set of pixels (i.e., feature points) is transmitted to a receiver in consideration of the above points, and the feature points may represent adjacent pixels. The motion vectors for the non-feature points at the receiver as the present pixels may be reconstructed from the motion vectors for the feature points.

Therefore, when transmitting a video signal, the transmitting side encodes a video signal that is coded through a transmission channel after encoding by considering the spatial and temporal correlation of the video signal in block units or pixel units through the above-described encoding technique. Will be sent to.

More specifically, the encoding system on the transmitting side removes spatial redundancy of the video signal by using transform coding such as discrete cosine transform (DCT), and further uses temporal encoding of the video signal by using differential coding through motion estimation and prediction. By eliminating redundant redundancy, the video signal can be efficiently compressed.

As described above, an example of a typical encoder using a hybrid encoding technique that compresses and encodes a video signal using temporal and spatial correlation for compression transmission to a receiver is one of a type as shown in FIG. As shown in the figure, a typical encoder includes a subtractor 10, a DCT unit 20, a quantizer 30, an inverse quantizer 40, an IDCT unit 50, an adder 60, and a frame memory 70 ), A motion estimation unit 80, a motion compensation unit 90, and a variable length coding unit (hereinafter, abbreviated as VLC) 100.

First, the subtractor 10 subtracts the predicted previous frame signal from the motion compensator 90 constituting the local decoder in the encoder for the motion compensation differential encoding from the current frame signal input from the input side. That is, the difference signal representing the difference pixel value is encoded by a series of quantized transform coefficients through the DCT unit 20 and the quantization unit 30. Then, the quantized transform coefficients are provided to the VLC unit 100 and the dequantizer 40 at the same time.

In more detail, as can be seen from FIG. 4, the DCT unit 20 uses the cosine function to select a video signal (pixel data) in the time domain for the differential signal based on the input motion estimation and prediction. Convert to DCT conversion coefficients in the frequency domain of 8 units. In addition, as illustrated in FIG. 4, the quantization unit 30 quantizes the DCT transform coefficient from the DCT unit 20 to a finite number of values through nonlinear operations. The difference signal between the frame and the predicted frame is quantized. In such quantization, the quantization unit 30 adjusts the quantization step size by the quantization parameter QP according to the fullness of data from the output side buffer, which is not shown in FIG.

Accordingly, the VLC unit 100 zigzags the differentially coded image data (quantized DCT transform coefficient) quantized through the quantization unit 30 as described above, for example, as shown in FIG. Encode them into runs and coefficients, such as by scanning. More specifically, the VLC unit 100 is variable according to the frequency of occurrence of each code by using one table, as shown in FIG. 5 (b) as an example. The code with less frequent occurrences of the code is encoded by a long code and then provided to an output buffer (not shown) for transmission to the receiver. The reason for allocating different lengths to all codes through the VLC unit 100 is to increase the coding efficiency by substantially reducing the average value of the code lengths, and the present invention substantially improves the efficiency of such variable length coding. It relates to the improvement of variable length coding schemes.

On the other hand, the inverse quantization unit 40 and the IDCT unit 50 which are employed as a typical encoder as a means for performing the motion prediction differential coding as described above and form a local decoder are the DCT unit 10 and the quantization unit 30. The compressed video signal (quantized DCT transform coefficient) is restored to the original signal before being encoded for motion estimation and compensation, and then provided to the adder 60. After that, the adder 60 is provided with an IDCT unit. The restored current frame signal (differential signal) provided from 50 and the predicted previous frame signal provided from the motion compensator 90 are added to the frame memory 70 and then restored to the frame memory 70. The current frame signal, that is, the frame immediately before the current frame signal input for current encoding.

Therefore, through this process, the previous frame signal stored in the frame memory 70 is continuously updated with the image data immediately before the input image data currently encoded.

Next, the motion estimation unit 80 estimates the motion of the current input frame to be encoded in 16 × 16 units within a predetermined search range from the previous frame stored in the frame memory 70, that is, the current frame and the most. A block of a similar previous frame is determined and provided to the motion compensator 90, which is based on the output information (search information) from the motion estimator 80 from the frame memory 70. The corresponding block of the previous frame is read and provided to the subtractor 10 and the adder 60, respectively. In addition, although not shown in FIG. 3, the motion vector determined by the motion estimation unit 80 is encoded through a predetermined encoding process for transmission to a receiving side decoder, and then is transmitted to an output buffer (not shown).

Accordingly, the subtractor 10 obtains the difference signal (differential pixel value) by subtracting the current frame signal from the input side with the predicted previous frame signal provided from the moving compensator 90, and the difference thus obtained. Since the signal is provided to the DCT section 20 of the next stage, DCT and quantization for the difference signal as described above are performed.

On the other hand, according to the conventional method of the typical encoder for performing the compression encoding of the video signal through the above-described process, in one frame or a system during variable length coding of the quantized DCT coefficient, Figure 5 (b) As shown in Fig. 2, Zigzag scanning uses only one variable-length codebook to allocate short codes when the run is small and long codes when the run is long, which is statistically significant. Occurrences and long runs are due to their low incidence, which also depends only on statistical values. Where run is the number of zeros before the nonzero coefficient.

However, in practice, the length of these runs varies substantially depending on a number of factors, for example, the type of the picture, the encoding form of the image, or the coefficient at any position within one DCT block. In the related art, since this point is not considered at all, there is a problem in that the amount of bit generation is unnecessarily large and the coding efficiency is lowered.

Accordingly, an object of the present invention is to provide an image compression method and apparatus capable of improving compression encoding efficiency of an image using classification vector quantization.

According to the present invention according to a consistent point to achieve the above object, a differential image signal based on motion estimation and prediction is converted into a DCT coefficient of size 8 × 8 in the frequency domain using a cosine function, and the DCT is performed through a nonlinear operation. A video compression method of an image encoder in which the quantized DCT coefficient of variation is quantized according to the frequency of occurrence of each code by quantizing a transform coefficient. After dividing into a plurality of areas having a predetermined size, the pattern of each area partitioned using a plurality of codebooks corresponding to the plurality of partitioned areas each of the image coder and each reference pattern in the codebook Compression coding of the quantized DCT transform coefficients is performed by performing classification vector quantization while matching. It provides an image coding method.

On the other hand, in the image compression method of the present invention as described above, 8 × 8 DCT coefficient blocks are substantially divided into 16 2 × 2 regions, and codebooks are provided with a number corresponding to 16 divided regions. . In addition, in the method of the present invention, the matching between the input image pattern and each reference pattern is performed by a mean square error calculation, and a reference pattern having a minimum providing error for the input image pattern is selected as the final output.

According to another aspect of the present invention, a differential image signal based on motion estimation and prediction is converted into a DCT coefficient of size 8 × 8 in a frequency domain using a cosine function, and the DCT is performed through a nonlinear operation. A video compression apparatus of an image encoder which quantizes a transform coefficient and compresses and encodes the quantized DCT transform coefficients according to an occurrence frequency of each code by using an empty codebook. Block cutting means for dividing into a plurality of areas having a predetermined size; A plurality of patterns having a number corresponding to the plurality of regions divided into the predetermined size and having respective codebooks, and performing classification vector quantization by matching patterns of the divided regions with reference patterns in the respective codebooks. Matching means; And a plurality of selection means for selecting one of the respective reference patterns drawn from the respective codebooks and providing an index thereof to the output side corresponding to each matching result from each of the plurality of pattern matching means. An image compression device using classification vector quantization is provided.

Other objects and various advantages of the present invention will become more apparent from the preferred embodiments of the present invention described below with reference to the accompanying drawings by those skilled in the art.

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

1 is a block diagram of an image compression apparatus using classification vector quantization according to an embodiment of the present invention. As shown in the figure, the image compression apparatus of the present invention substantially corresponds to the VLC unit 100 shown in FIG. 4, and includes a block cutting unit 110 and a plurality of patterns connected in parallel to the block cutting unit. The generators 121 to 136 are included. In addition, each pattern generation section is composed of codebooks 121-2 to 136-2, pattern matching sections 121-4 to 136-4, and output selection sections 121-6 to 136-6.

On the other hand, the classification vector quantization to be used in the present invention has the codebook and the best match, that is, the most similar to the data of each block divided into blocks of a predetermined size, for example, 2 × 2 area among the patterns in the codebook Instead of selecting and outputting the encoded bit value for each pixel of the corresponding block, an encoding of the pattern is output, that is, encoding using vector quantization for each block divided into a predetermined size. Of course, according to the present invention, the encoded image data may be restored to the original signal through a decoder on the receiving side having the same codebook as the encoder on the transmitting side.

In the image compression apparatus according to the exemplary embodiment of the present invention using the classification vector quantization as described above, the block cutting unit 10 is divided into quantizations of 8 × 8 blocks input from the quantization unit 30 of FIG. The DCT coefficient block is cut into 16 2 x 2 areas as shown in FIG.

Meanwhile, each of the plurality of pattern generators 121 to 36, which are connected in parallel to the output of the block cutting unit 110, has codebooks 121-2 to 136-2, and each pattern matching unit is described above. The pattern (coefficient) of the input image cut into the 2 × 2 size input from the block cutting unit 110 and the reference pattern (coefficient) drawn from each codebook are replaced with each other to distinguish the best matching pattern from each other.

More specifically, the pattern matching unit 12 determines a reference pattern that is optimal for the pattern of the input image by using a calculation method such as a mean square error (MSE), which is well known in the art, and determines the angle within each pattern generator. The output selection section provides an index for an optimal reference pattern that best matches the input pattern to the buffer on the output side, not shown in FIG. 3, so that the compression-encoded image data is transmitted to the receiving side.

Therefore, as described above, by dividing an 8 × 8 DCT block into 16 2 × 2 blocks and performing vector quantization using each codebook, patterns frequently generated in each divided area are generated. As a codebook is formed, the compression efficiency of an image is improved, that is, the bit generation amount can be reduced by applying vector quantization according to the characteristics of each divided region (2 × 2).

As an example, in the quantized DCT block shown in FIG. 2 (a), the patterns in the corresponding codebook in region # 1 will be mainly as shown in FIG. 2 (b) as an example. However, the patterns in the codebook in area # 16 will mainly be in the form as shown in FIG. As a result, according to the present invention, a bit generation amount can be reduced by classifying the pattern of the image data of each divided region as described above and replacing the divided image data with the index according to the pattern.

In other words, in processing a single DCT coefficient block in the encoder on the transmitting side, in the region # 1, the A ₅₁ pattern is most similar to the input image pattern, that is, the best match, and in the region # 16, the A ₅₀ pattern is the corresponding input. If it is determined that the image pattern is most similar, that is, the best match, the encoder provides the index 51 to the output buffer through the output selector 121-6 when it is the region # 1 of the quantized DCT coefficient block, and the region # 16. In this case, an index 50 is provided to the output buffer through the output selector 136-6.

Therefore, in the decoder on the receiving side having a plurality of codebooks substantially the same as the encoder on the transmitting side as described above, the A ₅₁ pattern and the A ₅₀ pattern corresponding to the corresponding indexes 51 and 50 received through the channel are stored in the corresponding codebook. By finding and replacing, it is possible to recover the original signal before compression coding.

As described above, the image compression method according to the present invention partitions an 8x8 DCT coefficient block into regions of a predetermined size (2x2) through the above-described process, and then classifies vector quantization according to each partitioned region. By providing only the index according to the receiver on the receiving side, the coding efficiency of the image data can be improved by suppressing the generation of the compression-coded image data amount (bit generation amount) as much as possible while substantially overcoming the above-described problems of the prior art. have.

Claims (6)

A differential image signal based on motion estimation and prediction is transformed into a DCT coefficient of size 8 × 8 in the frequency domain using a cosine function, quantized the DCT transform coefficient through nonlinear operation, and then coded using each codebook. In an image compression method of an image encoder which compresses and encodes the quantized DCT transform coefficients according to a frequency of occurrence of the quantized DCT coefficient blocks, the quantized DCT coefficient blocks of 8x8 size are divided into a plurality of regions having a predetermined size. The quantized DCT is performed by performing classification vector quantization while matching a pattern of each region partitioned with each reference pattern in each codebook using a plurality of codebooks corresponding to the plurality of partitioned regions of the image encoder. An image compression method comprising compressing and transforming a transform coefficient. The image compression method according to claim 1, wherein the quantized 8x8 DCT coefficient block is partitioned into 2x2 regions, and the plurality of codebooks has a number corresponding to the partitioned regions. The method of claim 1 or 2, wherein the matching between each input image pattern and each reference pattern is performed by a mean square error calculation, and a reference pattern having a minimum providing error is selected for the input image pattern. Characterized in that the video compression method. A differential image signal based on motion estimation and prediction is transformed into a DCT coefficient of size 8 × 8 in the frequency domain using a cosine function, quantized the DCT transform coefficient through nonlinear operation, and then coded using each codebook. An image encoder of an image encoder which compresses and encodes the quantized DCT transform coefficients according to a frequency of occurrence of the block, wherein the block divides the 8 × 8 quantized DCT coefficient block into a plurality of regions having a predetermined size. Cutting means; A plurality of patterns having a number corresponding to the plurality of regions divided into the predetermined size and having respective codebooks, and performing classification vector quantization by matching patterns of the divided regions with reference patterns in the respective codebooks. Matching means; And a plurality of selection means for selecting one of the respective reference patterns drawn from the respective codebooks and providing an index thereof to the output side corresponding to each matching result from each of the plurality of pattern matching means. Image compression apparatus using classification vector quantization. 5. The apparatus of claim 4, wherein the block cutting means divides the 8x8 quantized DCT coefficient block into 16 2x2 regions. The method of claim 4 or 5, wherein each of the plurality of pattern matching means performs matching of each of the input image patterns and each of the reference patterns through a mean square error calculation, and each of the plurality of selection means comprises a corresponding input image. And a reference pattern having a minimum square error between a pattern and each reference pattern.