RU2342704C1 - Device for two-dimensional direct discrete wavelet transformation in video data compression systems - Google Patents

Abstract

FIELD: physics; computer technology.

SUBSTANCE: present invention pertains to computer technology, and particularly to direct discrete wavelet transformation in video compression systems. Device comprises a unit of filters for one-dimensional wavelet transformation on the column of the video image (on the vertical), requiring less volume of memory buffer. Proposal is given of a device for two-dimensional direct discrete wavelet transformation in video data compression systems. The device comprises a multiplexer for generating an input stream at the first and second filter units, from either the input stream of video data, or from the memory buffer unit. The first filter unit for one-dimensional wavelet transformation on the line of the video image on the horizontal is for calculating the low frequency component, and the second filter unit for one-dimensional wavelet transformation on the line of the video image on the horizontal is for calculating the high frequency component. The device has a third filter unit for one-dimensional wavelet transformation on the column of the video image on the vertical for calculating low and high frequency components of data, coming from the first filter unit and from the first and second memory buffer units. The first memory buffer unit is for storage of intermediate data necessary for calculating low and high frequency components, coming from the third filter unit. The second memory buffer unit is for storing intermediate data, necessary for calculating low and high frequency components, coming from the third filter unit. The fourth filter unit for one-dimensional wavelet transformation on the column of the video image on the vertical is for calculating low and high frequency components of data, coming from the second filter unit and from the third and fourth memory buffer units. The third memory buffer unit is for storing intermediate data necessary for calculating low and high frequency components, coming from the fourth filter unit. The fourth memory buffer unit is for storing intermediate data, necessary for calculating low and high frequency components, coming from the fourth filter unit. The fifth memory buffer unit is for storing the low frequency component after the third filter unit for wavelet transformation, is used as a memory buffer unit, from which data arrive at the multiplexer.

EFFECT: less than 4,5N memory buffer is required.

3 dwg

Description

The invention relates to computing, and in particular to the field of performing direct discrete wavelet (wavelet) transform (FDWT - Forward discrete wavelet transform) in video compression systems.

Discrete two-dimensional wavelet transform is widely used in video compression systems, in particular, its application is described in the JPEG200 standard [1]. In such systems, each color component of the input frame of size N · M, where N is the frame width and M is the height, is represented as a data array of M rows and N columns. Further, for each row, a one-dimensional transformation is performed, which is low-frequency and high-frequency data filtering. Then, on the results of the conversion, for each of the columns, similar conversions are performed.

Thus, each color component is converted into a set of 2-dimensional Sub-band signals representing the signal intensity in different frequency bands, in different spatial areas. Sub-band is understood to mean a set of coefficients obtained using the same sequence of low-pass and high-pass filtering operations horizontally and vertically. Sub-bands are denoted as follows: LL, HL, LH or HH.

The coefficients in the sub-band HL are conversion coefficients obtained by low-pass vertical filtering and high-pass horizontal filtering. Coefficients in the sub-band LH are conversion coefficients obtained by high-pass vertical filtering and low-pass horizontal filtering.

The coefficients in the sub-band LV are the conversion coefficients obtained by high-pass filtering vertically and high-pass filtering horizontally.

The coefficients in the sub-band LL are the conversion coefficients obtained by low-pass vertical filtering and low-pass horizontal filtering.

Next, the transforms described above are performed on the LL coefficients. The number of iterations for processing LL components determines the number of decomposition levels J of the input image. The number of received sub-bands 3 · J + 1.

To indicate the coefficients obtained after transformations over the LL component, use the notation: (LL) ^a L, (LL) ^a H, (LL) ^a LL, (LL) ^a LH, (LL) ^a HL, (LL) ^a HH, where index a is the level of decomposition and 0≤a <J.

An increase in the number of decomposition levels allows increasing the degree of compression (compression) of video data.

The use of FDWT for image compression is described in [1]. This reversible conversion is a reversible implementation of 5/3 filtering, where 5 and 3 are the lengths of the low-pass and high-pass filters, respectively.

Reversible filtering consists of a sequence of simple filtering operations in which the odd coefficients are updated taking into account the “weighted” sum of the odd values of the signal, rounded to an integer value, and the even coefficients are updated taking into account the “weighted” sum of the odd coefficients, rounded to an integer value.

For each one-dimensional data array, which is a row or column of data, the following calculations are performed.

First, the odd coefficients of the output signal Y are calculated for such values of n that i ₀ -1≤2n + 1 <i ₁ +1:

where i ₀ is the index of the first element in the array X, i ₁ is the index of the element following the last element of the array X.

Then, the even coefficients of the output signal Y are calculated from the even coefficients of the signal X and the odd coefficients of the signal Y for values of n such that i ₀ ≤2n <i ₁ :

The even coefficients of the signal Y are the low-frequency decimated version of the signal X, the odd are the high-frequency decimated version of the signal X.

Known technical solutions of devices [2, 3] that perform two-dimensional wavelet transform and use memory blocks with a volume of N · M to store intermediate data, where N is the frame width and M is the height. The disadvantage of these solutions is the large amount of memory required, which leads to either a significant increase in the area, power consumption and cost of the integrated circuit chip that implements video compression functions, or to the need to use external memory devices that increase the cost and volume of the equipment.

The closest technical solution of the present invention is a device for compressing video data using the wavelet transform shown in [4], including:

- a signal multiplexer for generating an input stream to the first and second filter block either from the input video stream or from the third block of the memory buffer,

- the first filter block for performing one-dimensional wavelet transform along the line of the video image (horizontal) to calculate the low-frequency component (LL) ^a L,

- the second filter block for performing one-dimensional wavelet transform along the line of the video image (horizontal) to calculate the high-frequency component (LL) ^a H,

- the first block of the memory buffer with a volume of KN (1- (1/2) ^J ) ≈5N samples, for storing intermediate data necessary for calculating the low-frequency and high-frequency components (LL) ^a LL and (LL) ^a LH when performing a wavelet transform on a column video images (vertical),

- the second block of the memory buffer with a volume of KN (1- (1/2) ^J ) ≈5N samples for storing intermediate data necessary for calculating the low-frequency and high-frequency components (LL) ^a HL and (LL) ^a HH when performing wavelet transform on the video image column (vertically),

- the third block of filters for performing one-dimensional wavelet transform on the column of the video image (vertical) to calculate the low-frequency component (LL) ^a LL,

- the fourth block of filters to perform one-dimensional wavelet transform on the column of the video image (vertical) to calculate the high-frequency component (LL) ^a LH,

- the fifth filter block for performing one-dimensional wavelet transform on the column of the video image (vertical) to calculate the low-frequency component (LL) ^a HL,

- the sixth filter block to perform one-dimensional wavelet transform on the column of the video image (vertical) to calculate the high-frequency component of the signal (LL) ^a HH,

- the third block of the memory buffer with a volume of N (1- (1/2) ^J-1 ) ≈N samples for storing the low-frequency component (LL) ^a LL after performing the wavelet transform on the video image column (vertically).

The functional diagram of the device according to the prototype [4] is illustrated in figure 1. The device contains: 1 - a signal multiplexer for generating an input stream to the first and second filter block either from the input video stream or from the third block of the memory buffer, 2 - the first filter block to perform one-dimensional wavelet transform on the video image line (horizontally) to calculate the low-frequency component (LL) ^a L, 3 - the second block of filters for performing one-dimensional wavelet transform along the line of the video image (horizontal) to calculate the high-frequency component (LL) ^a H, 4 - the first block memory frame with a volume of KN (1- (1/2) ^J ) ≈5N samples for storing intermediate data necessary for calculating the low-frequency and high-frequency components (LL) ^a LL and (LL) ^a LH when performing wavelet transforms along the video image column (vertical ), 5 - the second block of the memory buffer with a volume of KN (1- (1/2) ^J ) ≈5N samples for storing intermediate data necessary for calculating the low-frequency and high-frequency components (LL) ^a HL and (LL) ^a HH when performing the wavelet transform on the column of the video image (vertical), 6 - the third block of filters for one-dimensional wavelet transform along the column of the video image (vertical) to calculate the low-frequency component (LL) ^a LL, 7 - the fourth filter block to perform one-dimensional wavelet transform along the column of the video image (vertical) to calculate the high-frequency component (LL) ^a LH, 8 - fifth filter unit for performing one-dimensional wavelet transformation on the video column (vertical) to calculate the low-frequency component (LL) ^a HL, 9 - sixth filter unit for performing one-dimensional wavelet transformation anija video on column (vertical) to calculate high-frequency component of the signal (LL) ^a HH, 10 - third buffer memory unit volume N (1- (1/2) ^J-1) ≈N counts for storing the low frequency component (LL) ^a LL after performing the wavelet transform on the video column (vertical).

As follows from the description of the device specified in the prototype [4], the required amount of memory blocks is:

- the first block of the memory buffer KN (1- (1/2) ^J ), where K is the filter length;

- the second block of the memory buffer KN (1- (1/2) ^J ), where K is the filter length;

- the third block of the memory buffer N (1- (1/2) ^J-1 ).

Thus, the total amount of memory required to perform the conversion is 2KN (1- (1/2) ^J ) + N (1- (1/2) ^J-1 ), which tends to 2KN + N with an increase in the number of decomposition levels J.

In [1], the length of the filler K is 5 and 3 for the low-pass and high-pass filters, respectively. Therefore, to perform the conversion, memory buffers with a total volume of 2KN + N = 11N are required, at K = 5. The size of the buffer is determined by the maximum length of the filter.

The disadvantage of this solution is still the large amount of required memory buffers, amounting to about 11N, which leads to an increase in the size, power consumption and cost of the device.

The problem to which the invention is directed, is to achieve a technical result, which consists in reducing the required amount of memory buffers to 4N (1- (1/2) ^J ) + N / 2, which is less than 4.5N, which reduces the size, power consumption and the cost of the device, as a result of the fact that filter blocks are included in the device for performing one-dimensional wavelet transforms on a video image column (vertically), requiring a smaller memory buffer.

To achieve the named technical result in the device [4], containing:

- a signal multiplexer for generating an input stream to the first and second filter block either from the input video stream or from the memory buffer block, and the first filter block to perform a one-dimensional wavelet transform along the video line (horizontally) to calculate the low-frequency component (LL) ^a L,

- the second filter block for performing one-dimensional wavelet transform along the line of the video image (horizontal) to calculate the high-frequency component (LL) ^a H,

include the following blocks, which are the hallmarks of the proposed device:

- the third filter block for performing one-dimensional wavelet transforms along the vertical column of the video image to calculate the low-frequency and high-frequency components of the data coming from the first filter block and from the first and second blocks of memory buffers,

- the first block of the memory buffer for storing incoming from the third block of filters of intermediate data necessary for calculating the low-frequency and high-frequency components,

- the second block of the memory buffer for storing the intermediate data coming from the third block of filters necessary for calculating the low-frequency and high-frequency components,

- the fourth filter block for performing one-dimensional wavelet transforms along the vertical column of the video image to calculate the low-frequency and high-frequency components of the data coming from the second filter block and from the third and fourth blocks of memory buffers,

- the third block of the memory buffer for storing incoming from the fourth block of filters intermediate data necessary for calculating the low-frequency and high-frequency components,

- the fourth block of the memory buffer for storing incoming from the fourth block of filters of intermediate data necessary for calculating the low-frequency and high-frequency components,

- the fifth block of the memory buffer for storing the low-frequency component after the third block of filters has performed the wavelet transform is used as the block of the memory buffer from which the data is sent to the signal multiplexer.

Conducted patent studies have shown that in the literature there is no indication of the use of the above set of distinctive features to solve the problem of reducing the required amount of memory buffer to perform FDWT.

The functional diagram of the device according to the proposed solution is illustrated in figure 2.

The proposed device contains: 1 - a signal multiplexer for generating an input stream to the first and second filter block either from the input video stream or from the fifth block of the memory buffer, 2 - the first filter block to perform one-dimensional wavelet transform along the video image line (horizontally) to calculate the low-frequency component (LL) ^a L, 3 - a second filter unit for performing one-dimensional wavelet transformation on the video line (horizontal) to calculate high-frequency component (LL) ^a H, 11 - tr Tille filter unit for performing one-dimensional wavelet transform on a column of video (vertical) to calculate the low frequency (LL) ^a LL and the high-frequency component (LL) ^a LH, 12 - first memory buffer unit volume of N samples, to store intermediate data required to calculate the low-frequency (LL) ^a LL and the high-frequency component (LL) ^a LH when performing wavelet transforms on a video image column (vertically), 13 - the second block of the memory buffer with a volume of N samples, for storing intermediate data necessary for calculating of low-frequency (LL) ^a LL and high-frequency component (LL) ^a LH when performing wavelet transform on the video image column (vertical), 14 - the fourth filter block to perform one-dimensional wavelet transform on the video image column (vertical) to calculate low-frequency (LL) ^a HL and high-frequency component (LL) ^a HH, 15 - the third block of the memory buffer with a volume of N samples for storing intermediate data necessary for calculating the low-frequency (LL) ^a HL and high-frequency component (LL) ^a HH when performing a wavelet transform a column of the video image (vertical), 16 - the fourth block of the memory buffer with a volume of N samples for storing intermediate data needed to calculate the low-frequency (LL) ^a HL and high-frequency component (LL) ^a HH when performing wavelet transform on the video column (vertical ), 17 - the fifth block of the memory buffer with a volume of N / 2 samples for storing the low-frequency component (LL) ^a LL after performing the wavelet transform on the video column (vertical).

The proposed device operates as follows:

1. Input video data, line by line, is fed to the first input of the signal multiplexer, from the output of which data are fed to the first and second filter blocks;

2. The first and second filter blocks perform one-dimensional wavelet transform along the video line to calculate the low-frequency L and high-frequency component H for the first level of decomposition, respectively;

3. From the output of the first filter block, the data is supplied to the third filter block to perform a one-dimensional wavelet transform on the video column to calculate the low-frequency LL and high-frequency component LH;

4. From the output of the second filter block, the data is sent to the fourth filter block to perform a one-dimensional wavelet transform on the video column to calculate the low-frequency HL and high-frequency component of the LV;

5. The third and fourth filter blocks are implemented in such a way that they perform video column conversion without first accumulating five lines of input video data in the memory buffers. In this case, the filter blocks carry out the conversion simultaneously with the input data. The intermediate results of the calculations of the third filter block are stored in the first and second blocks of memory buffers, and the intermediate results of the calculations of the fourth filter blocks are stored in the third and fourth blocks of memory buffers;

6. From the output of the fourth block of filters, the conversion results (HL and LV components of the signal) are sent to the output of the device;

7. From the output of the third block of filters, the LH component goes to the output of the device, and the LL component goes to either the fifth block of the memory buffer or to the output of the device if the LL component is calculated for the last level of decomposition;

8. From the output of the fifth memory buffer, the data (LL component) is fed to the second input of the signal multiplexer, the processing of this data is carried out in a similar way.

The sequence of calculations in the third and fourth filter blocks is shown in Fig.3.

It is assumed here that at the a-th level of decomposition, data matrices of size n rows by m columns are processed, a Xi, j an element of this matrix, where i is the row number, j is the column number. LBuf, HBuf - memory buffer for the low-frequency and high-frequency component, respectively, tempL, tempH - internal buffers of the filter unit for storing temporary data, Out - the output of the filter. For the third filter block, the LBuf, HBuf buffers are implemented in the first and second blocks of the memory buffer, respectively. For the fourth filter block, the LBuf, HBuf buffers are implemented in the third and fourth memory buffer blocks, respectively.

Consider the operation of filter blocks using the example of the third filter block. The video data Xi, j is supplied sequentially, line by line, to the input of the filter unit 2. Depending on the line number, the data is processed in different ways. The first line of 3 is written to the LBuf buffer without changes 3.1. The second line 4 is processed together with the data from the LBuf buffer, and the calculation result is written to the HBuf 4.1 buffer. The third line 5, processed together with the data from the LBuf, HBuf buffers, is used to calculate the output of the Out 5.1 filter and the new values of the intermediate data written to the LBuf, HBuf 5.2 buffers. The last line of frame 6, depending on whether it is even or odd 7, is processed together with data from the LBuf, HBuf buffers, used to calculate the output of the Out filter (7.1 for even lines, 7.3 for odd lines) and new values of intermediate data written to LBuf, HBuf buffers (7.2 for even lines, 7.4 for odd lines). All other lines (8) are processed in accordance with (8.1 and 8.2) for even lines and (8.3 and 8.4) for odd lines.

As follows from the foregoing, to perform transformations, the device does not require storage of five lines of input data, for each of the filter blocks that perform column-by-video conversion, or even storage of the whole frame, but uses memory buffers to store intermediate data with two lines of memory for each of the filter blocks that perform the conversion on the column of the video image, which ensures the implementation of the claimed technical result in the present invention.

Literature

1. TITLE: JPEG 2000 Part I Final Committee Draft Version 1.0 SOURCE: ISO / IEC JTC1 / SC29 WG1, JPEG 2000 Editor Martin Boliek, Coeditors Charilaos Christopoulos, and Eric Majani.

2. Patent US 4943855.

3. Patent US 5014134.

4. Chaitali Chakrabarti, Member, and Clint Mumford “Efficient Realizations of Encoders and Decoders Based on the 2-D Discrete wavelet Transform” IEEE TRANSACTIONS ON VERY LARGE SCALE INTEGRATION (VLSI) SYSTEMS, VOL. 7, NO. 3, SEPTEMBER 1999 Fig. 5. pp. 293.

Claims (1)

A device for performing two-dimensional direct discrete wavelet transform in video compression systems, comprising a signal multiplexer for generating an input stream to the first and second filter blocks from either the video input stream or from the memory buffer block, the first filter block for performing one-dimensional wavelet transform on the video image line horizontal to calculate the low-frequency component, the second filter block to perform one-dimensional wavelet transform on the line of the video image horizontally to calculate the high-frequency component, characterized in that the device includes a third filter block for performing one-dimensional wavelet transform along the vertical video column to calculate the low-frequency and high-frequency components over the data coming from the first filter block and from the first and second blocks of memory buffers , the first block of the memory buffer for storing intermediate data coming from the third block of filters, necessary for calculating the low-frequency and high-frequency components, the second block of the memory buffer for storing intermediate data coming from the third filter block, necessary for calculating the low-frequency and high-frequency components, the fourth filter block for performing one-dimensional wavelet transform along the vertical column of the video image to calculate the low-frequency and high-frequency components over the data coming from the second filter block and from the third and fourth blocks of memory buffers, the third block of the memory buffer for storing incoming from the fourth block filters of intermediate data necessary for calculating the low-frequency and high-frequency components, the fourth block of memory buffer for storing incoming from the fourth block of filters intermediate data necessary for calculating the low-frequency and high-frequency components, and the fifth block of memory buffer for storing the low-frequency component after the third block of filters performs wavelet transform used as a block of the memory buffer, from which data is fed to the signal multiplexer.

Images