CN118250475A - Two-dimensional image compression method, decompression method and decompression system - Google Patents

Abstract

The application provides a two-dimensional image compression method, a decompression method and a decompression system, wherein the system comprises an acquisition module, a decoding module, a storage module, a quantization module and an encoding module; the acquisition module acquires a target image; decomposing a target image of the uncompressed image into a matrix of pixels; the storage module executes DCT conversion of different conversion modes according to the bit width and the size of the DCT conversion matrix to obtain a frequency domain signal matrix; the decoding module performs decoding on a target image of the compressed image to obtain a decoding matrix; the storage unit executes IDCT transformation of different transformation modes according to the bit width and the size of the IDCT transformation matrix to form a time domain signal matrix. Based on the decompression system, the operation efficiency in DCT/IDCT in the image compression and decompression process is accelerated, a large number of multiplication operations are realized by using a form of bitwise shift addition, and DCT/IDCT calculation of matrixes with different sizes and bit widths is realized by switching different modes, so that the image compression and decompression efficiency is improved.

Description

Two-dimensional image compression method, decompression method and decompression system

Technical Field

The present application relates to the field of image processing, and in particular to a two-dimensional image compression method, a decompression method and a decompression system.

Background Art

In order to conveniently and efficiently transmit and store digital information such as images and videos, it is necessary to compress images and other data to reduce the amount of data stored and transmitted. Image compression methods include lossless compression algorithms, lossy compression algorithms, feature extraction algorithms, image compression coding algorithms, deep learning algorithms, etc.

Image compression coding algorithms include Discrete Cosine Transform (DCT) and Inverse Discrete Cosine Transform (IDCT). DCT and IDCT have good energy concentration characteristics. After DCT transformation, the data is divided into DC component and AC component, which paves the way for further compression.

DCT and IDCT involve a large number of matrix multiplication operations, which means that a large number of multipliers are required. The multipliers have high precision, but the hardware complexity of the multipliers is several times that of the adders. In addition, the hardware requires storage units to store the corresponding transformation matrix data, and the area cannot be effectively controlled. At the same time, when performing multiplication operations, the transformation matrix data needs to be read from the storage unit, which will cause a large delay in on-chip calculations, resulting in low image compression and decompression efficiency.

Summary of the invention

The present application provides a two-dimensional image compression method, a decompression method and a decompression system to solve the problem of low image compression and decompression efficiency.

In a first aspect, the present application provides a two-dimensional image compression method, comprising:

Get the target image;

If the target image is an uncompressed image, decomposing the target image into a pixel matrix;

Performing DCT transformations of different transformation modes according to the bit width and size of the DCT transformation matrix to obtain a frequency domain signal matrix;

Performing data quantization on the frequency domain signal matrix to obtain a quantization matrix;

The quantization matrix is encoded to form a compressed image format, indicating that the compression of the target image is completed.

In some feasible embodiments, the size includes a first size and a second size, and the bit width includes a first bit width and a second bit width;

The method of performing DCT transformations of different transformation modes according to the bit width and size of the DCT conversion matrix to obtain a frequency domain signal matrix includes:

Obtaining a transposed matrix of the pixel matrix;

Predefined DCT transformation matrix;

Acquire the transform mode according to the size and bit width of the DCT transform matrix;

The size of the DCT conversion matrix is a first size, and the bit width of the DCT conversion matrix is a first bit width, obtaining a first DCT transformation mode;

The size of the DCT conversion matrix is a first size, and the bit width of the DCT conversion matrix is a second bit width, obtaining a second DCT transformation mode;

The size of the DCT conversion matrix is the second size, and the bit width of the DCT conversion matrix is the first bit width, obtaining a third DCT transformation mode;

The size of the DCT conversion matrix is the second size, and the bit width of the DCT conversion matrix is the second bit width, and a fourth DCT transformation mode is obtained.

In some feasible embodiments, performing DCT transformations of different transformation modes according to the bit width and size of the DCT conversion matrix to obtain a frequency domain signal matrix includes:

When the transform mode is the first DCT transform mode, the DCT conversion matrix is input into the storage unit by column;

According to the element position of the DCT conversion matrix, each column of the transposed matrix is rearranged according to the position of the storage and calculation unit to obtain a first rearranged matrix;

Performing a first matrix multiplication operation on the first rearrangement matrix and the conversion matrix to obtain a first result matrix;

A first matrix multiplication operation is performed on the first result matrix and the conversion matrix to obtain a frequency domain signal matrix.

In some feasible embodiments, performing a first matrix multiplication operation on the first rearrangement matrix and the conversion matrix to obtain a first result matrix includes:

Input each column element of the first rearrangement matrix from bit 0, and multiply it with each column element of the DCT conversion matrix to obtain multiple columns of first elements, where the first elements are 8-bit numbers;

Perform vertical addition on the first element to obtain a second element, where the second element is an 11-bit number;

Performing a rounding operation on the second element to obtain the first element;

The first element is input into a shift adder to obtain a first result matrix.

In some feasible embodiments, performing a first matrix multiplication operation on the first rearrangement matrix and the conversion matrix to obtain a first result matrix includes:

When the transform mode is the second DCT transform mode, dividing the element positions of the DCT conversion matrix into a first part of positions and a second part of positions;

The first partial position and the second partial position are written with a weight of the second bit width;

Input each column element of the first rearranged matrix from position 0, and multiply it with each column element of the first partial position to generate a third element, wherein the third element is a 4-bit number;

Connecting the two third elements end to end to obtain the first element;

Performing vertical addition on the first element to obtain a second element;

Performing a rounding operation on the second element to obtain the first element;

The first element is input into a shift adder to obtain a first result matrix.

In some feasible embodiments, performing a first matrix multiplication operation on the first rearrangement matrix and the conversion matrix to obtain a first result matrix includes:

When the transform mode is the third DCT transform mode, dividing the transposed matrix of the first bit width into matrices of the first size and storing them in the storage unit;

Splitting the DCT transformation matrix into matrices of a first size;

Performing vertical addition on the first elements in every two of the storage and calculation units to obtain a second element, and adding two of the second elements to obtain two fourth elements, where the fourth elements are 12-bit numbers;

Performing a rounding operation on the fourth element to obtain the first element;

The first element is input into a shift adder to obtain a first result matrix.

The first aspect provides a two-dimensional image compression method, which implements multiplication operations in the form of bit-by-bit shift addition, and based on a storage and calculation unit, accelerates matrix operations, reduces the delay of hardware calculations, can realize real-time calculations, and can realize multi-mode DCT calculations to improve the efficiency of image compression.

In a second aspect, the present application provides a two-dimensional image decompression method, comprising:

Get the target image;

If the target image is a compressed image obtained by the method of the first aspect, decoding is performed on the target image to obtain a decoding matrix;

IDCT transformations of different transformation modes are performed according to the bit width and size of the IDCT transformation matrix to form a time domain signal matrix, which represents the decompression of the target image.

In some feasible embodiments, the size includes a first size and a second size, and the bit width includes a first bit width and a second bit width;

The IDCT transformation of different transformation modes is performed according to the bit width and size of the IDCT transformation matrix to form a time domain signal matrix, including:

Obtaining a transposed matrix of the pixel matrix;

Predefined IDCT transformation matrix;

Obtaining the transformation mode according to the size and bit width of the transposed matrix;

The size of the IDCT conversion matrix is a first size, and the bit width of the IDCT conversion matrix is a first bit width, and a first IDCT conversion mode is obtained;

The size of the IDCT conversion matrix is the first size, and the bit width of the IDCT conversion matrix is the second bit width, obtaining a second IDCT conversion mode;

The size of the IDCT conversion matrix is the second size, and the bit width of the IDCT conversion matrix is the first bit width and the second bit width, and a third IDCT transformation mode is obtained.

In some feasible embodiments, performing IDCT transformations of different transformation modes according to the bit width and size of the IDCT transformation matrix to form a time domain signal matrix includes:

When the transformation mode is the third IDCT transformation mode, the IDCT transformation matrix is input into the storage unit by column;

According to the element positions of the IDCT conversion matrix, each column of the transposed matrix is rearranged according to the position of the storage and calculation unit to obtain a second rearranged matrix;

dividing the second rearranged matrix of a second size into matrices of a first size;

Performing vertical addition on every two first elements in the storage and calculation units to obtain a second element, and adding two of the second elements to obtain two fourth elements, where the fourth elements are 12-bit numbers;

Performing a rounding operation on the fourth element to obtain the first element;

Inputting the elements of the first elements into a shift adder to obtain a second result matrix;

A first matrix multiplication operation is performed on the second result matrix and the IDCT conversion matrix to obtain a time domain signal matrix.

The second aspect provides a two-dimensional image decompression method, which implements multiplication operations through bit-by-bit shift addition, and based on the storage and calculation unit, accelerates matrix operations, reduces the delay of hardware calculations, can realize real-time calculations, and can realize multi-mode IDCT calculations to improve the efficiency of image decompression.

In a third aspect, the present application provides a two-dimensional image decompression system, including: an acquisition module, a decoding module, a storage module, a quantization module and an encoding module;

The acquisition module is used to acquire the target image;

If the target image is an uncompressed image, decomposing the target image into a pixel matrix;

The storage and calculation module is used to perform DCT transformations of different transformation modes according to the bit width and size of the DCT transformation matrix to obtain a frequency domain signal matrix;

The quantization module is used to perform data quantization on the frequency domain signal matrix to obtain a quantization matrix;

The encoding module is used to perform encoding on the quantization matrix to form a compressed image format, indicating that the compression of the target image is completed;

If the target image is a compressed image, the decoding module is used to decode the target image to obtain a decoding matrix;

The storage and calculation unit is also used to perform IDCT transformations of different transformation modes according to the bit width and size of the IDCT transformation matrix to form a time domain signal matrix to represent the decompression of the target image.

It can be seen from the above technical scheme that the present application provides a two-dimensional image compression method, a decompression method and a decompression system, and the system includes an acquisition module, a decoding module, a storage and calculation module, a quantization module and an encoding module; the acquisition module is used to acquire a target image; if the target image is an uncompressed image, the target image is decomposed into a pixel matrix; the storage and calculation module is used to perform DCT transformations of different transformation modes according to the bit width and size of the DCT transformation matrix to obtain a frequency domain signal matrix; the quantization module is used to perform data quantization on the frequency domain signal matrix to obtain a quantization matrix; the encoding module is used to encode the quantization matrix to form a compressed image format, indicating that the compression of the target image is completed; if the target image is a compressed image, the decoding module is used to decode the target image to obtain a decoding matrix; the storage and calculation unit is also used to perform IDCT transformations of different transformation modes according to the bit width and size of the IDCT transformation matrix to form a time domain signal matrix, indicating that the decompression of the target image is completed. Hardware technology based on in-memory computing accelerates the efficiency of matrix operations in DCT/IDCT during two-dimensional image compression and decompression, and can achieve end-to-end real-time operations; based on hardware reconfigurable technology, the same memory and computing unit can perform DCT/IDCT calculations, matrix quantization, encoding, and decoding of matrices of different sizes and bit widths according to configuration signals; the multiplier is replaced by bit-by-bit shift addition to achieve a large number of multiplication operations, and the calculation bit width can be adjusted at the same time to solve the problem of low image compression and decompression efficiency.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the technical solution of the present application, the drawings required for use in the embodiments are briefly introduced below. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

FIG1 is a schematic flow chart of a two-dimensional image compression method according to the present embodiment;

FIG2 is a schematic diagram of pixel matrix segmentation shown in this embodiment;

FIG3 is a schematic diagram of color block division shown in this embodiment;

FIG4 is a schematic diagram of a transposed matrix of a pixel matrix shown in this embodiment;

FIG5 is a schematic diagram of a DCT transform mode shown in this embodiment;

FIG6 is a schematic diagram of a process of generating a frequency domain signal matrix according to this embodiment;

FIG7 is a schematic diagram of storing a conversion matrix into a storage unit according to the present embodiment;

FIG8 is a schematic diagram of the sorting method of the transposed matrix in the DCT transformation shown in this embodiment;

FIG9 is a schematic diagram of obtaining a matrix B according to this embodiment;

FIG10 is a schematic diagram of a third DCT transform mode shown in this embodiment;

FIG11 is a schematic diagram of conversion matrix partitioning shown in this embodiment;

FIG12 is a schematic diagram of transposed matrix partitioning shown in this embodiment;

FIG13 is a schematic diagram of scanning coding shown in this embodiment;

FIG14 is a schematic flow chart of a two-dimensional image decompression method according to this embodiment;

FIG15 is a schematic diagram of an IDCT transformation mode shown in this embodiment;

FIG16 is a schematic diagram showing the ordering method of the transposed matrix in the IDCT transformation in this embodiment;

FIG17 is a schematic diagram of IDCT shown in this embodiment;

FIG18 is a schematic diagram of the structure of a two-dimensional image decompression system according to this embodiment;

FIG. 19 is a flowchart of the two-dimensional image decompression system according to this embodiment.

Illustration Description:

Among them, 1801-acquisition module, 1802-decoding module, 1803-storage and calculation module, 1804-quantization module, and 1805-encoding module.

DETAILED DESCRIPTION

The following embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following embodiments do not represent all implementations consistent with the present application. They are only examples of systems and methods consistent with some aspects of the present application as detailed in the claims.

DCT/IDCT is a computationally intensive operation involving a large number of matrix multiplication operations. If it is directly implemented through hardware, a large number of multipliers are required. Multipliers have a relatively large accuracy advantage over adders, but the hardware complexity of multipliers is several times that of adders; the circuit area occupied is large and the power consumption is large; fast DCT/IDCT hardware requires a special storage unit (Read-Only Memory, ROM) to store the transformation matrix data, which cannot effectively control the area. At the same time, when performing multiplication operations, the transformation matrix data needs to be read from the ROM, which will cause a large delay in on-chip calculations and low calculation efficiency. If DCT is implemented using an application-specific integrated circuit (ASIC), it will perform better in terms of processing power, power consumption, and chip area compared to other implementation schemes. However, the implementation method of ASIC is less flexible, and the hardware circuit cannot be changed after the tape-out stage is completed. In other words, only a single fixed-bit width DCT calculation can be implemented, resulting in low image compression and decompression efficiency.

In order to solve the problem of low image compression efficiency, referring to FIG1 , some embodiments of the present application provide a two-dimensional image compression method, including:

S101: Acquire a target image.

The target image is a two-dimensional image, that is, the target image does not contain depth information, but only contains length and width information of two dimensions, for example: geometric figures, raster images, vector images, etc.

In this embodiment, the target image is an image to be compressed.

S102: If the target image is an uncompressed image, decompose the target image into a pixel matrix.

The target image has 16 rows and 16 columns of pixels, with a total of 256 pixels, that is, the target image is a 16×16 pixel matrix.

The 16×16 pixel matrix is divided into four 8×8 pixel matrices. For example: the first 8×8 matrix contains the first 8 rows and first 8 columns of pixels of the 16×16 matrix; the second 8×8 matrix contains the first 8 rows and last 8 columns of pixels of the 16×16 matrix; the third 8×8 matrix contains the last 8 rows and first 8 columns of pixels of the 16×16 matrix; the fourth 8×8 matrix contains the last 8 rows and last 8 columns of pixels of the 16×16 matrix.

Referring to FIG. 2 , the 8×8 pixel matrix in FIG. 2 has 8 rows and 8 columns of pixels, including 64 pixels, and each pixel has a bit width of 8 bits (binary digit).

S103: Perform DCT transformations of different transformation modes according to the bit width and size of the DCT transformation matrix to obtain a frequency domain signal matrix.

DCT requires a predefined transformation matrix to transform the input signal from the time domain to the frequency domain. The transformation matrix contains the basis functions of the DCT transform, which can be discrete forms of cosine functions. In the DCT transform, the input signal is decomposed into a series of linear combinations of cosine functions, and the coefficients of the basis functions form the transformation matrix. Therefore, before performing the DCT transform, the transformation matrix needs to be predefined so that it can be applied to the input signal.

In some embodiments, referring to FIG. 3 , a DCT conversion matrix is predefined, wherein the DCT conversion matrix A has 8 rows and 8 columns of pixels, and the DCT conversion matrix is first divided according to color blocks, wherein each pixel block in the first row has a different color, with a total of 8 color blocks, and the color of the first pixel block in the second row is the same as the color of the eighth pixel block in the first row, that is, the color of the first pixel block in the next row is the color of the last pixel block in the previous row.

The second pixel block of the second row has the same color as the first pixel block of the first row, and the third to eighth pixel blocks of the second row are arranged in the order of the pixel blocks of the first row, and have the same color as the second to seventh pixel blocks of the first row. The other rows are arranged according to the rule between the second row and the first row.

The formula for two-dimensional DCT transform is:

Among them, F(u,v) is the frequency of the coordinate (u,v) after DCT transformation, c(u) and c(v) are compensation coefficients, which can make the DCT transformation matrix an orthogonal matrix, and f(i,j) is the pixel data of the coordinate (i,j).

The two-dimensional DCT transformation formula is a matrix transformation formula, which can be replaced by matrix operations, as shown in the following formula:

Y = ^AXAT ;

Among them, Y is the frequency domain signal matrix, A is the conversion matrix, X is the pixel matrix, and ^AT is the transposed matrix of the conversion matrix.

In order to reuse the same hardware for two matrix multiplications, that is, to multiply the conversion matrix by the transpose of a matrix, an identity transformation is performed according to the rules of matrix operations in linear algebra, see the following formula:

Y = ^AXAT = A( ^AXT ) ^T ;

In some embodiments, the matrix size includes a first size and a second size, wherein the first size is 8×8 and the second size is 16×16, and the matrix bit width includes a first bit width and a second bit width, wherein the first bit width is 8 bits and the second bit width is 16 bits.

Obtain the transposed matrix ^XT of the pixel matrix, see FIG4 .

In some embodiments, the transform mode is obtained according to the size and bit width of the DCT conversion matrix. Referring to FIG. 5 , this embodiment provides four DCT transform modes. Different DCT transform modes can be executed for different matrix sizes and bit widths.

The size of the DCT conversion matrix is a first size, and the bit width of the DCT conversion matrix is a first bit width, and a first DCT transformation mode is obtained.

See Figure 6: The following steps are included:

S601: Input the DCT conversion matrix into the storage unit column by column.

Referring to FIG. 7 , after the conversion matrix A is divided into color blocks, it is stored in the storage and calculation unit.

S602: According to the element positions of the DCT conversion matrix, each column of the transposed matrix is rearranged according to the position of the storage and calculation unit to obtain a first rearranged matrix.

The transposed matrix is input into the storage unit from left to right in columns. Since the weights in the storage unit, that is, the elements of the transformation matrix A, are the transformation matrix that is scrambled and rearranged, each path of the device matrix needs to be rearranged according to the position in the storage unit to ensure the accuracy of the matrix multiplication.

S603: Perform a first matrix multiplication operation on the first rearrangement matrix and the DCT conversion matrix to obtain a first result matrix.

When the transform mode is the first DCT transform mode, obtaining the first result matrix comprises the following steps:

Input each column element of the first rearrangement matrix from bit 0 and multiply it with each column element of the DCT conversion matrix to obtain the first element of eight columns, where the first element is an 8-bit number;

Perform vertical addition on the elements to obtain the second element, which is an 11-bit number.

Perform rounding operation on the second element to obtain the first element;

The elements of the first element are input to the shift adder to obtain a first result matrix.

For example, referring to FIG8 and FIG9, the rearranged column elements are inputted from the <0>th bit, and the <0>th bit of a column is multiplied by the weight of the first column to obtain a column element, i.e., 8 8-bits. The 8 8-bits are added vertically. According to the bit expansion principle of addition, an 11-bit number can be obtained. The decimal places of the 11-bit number are cut off, and the final number is 8-bit. If the number does not reach 8-bit after the decimal places are cut off, the saturation truncation process can be continued. Similarly, until the end of the <8>th bit, the column elements can be inputted completely to obtain an 8-bit number, which is inputted into the shift adder to obtain the element [0,0] of the new matrix B after the multiplication of AX ^T , where matrix B is the first result matrix. By analogy, the second column, the third column, ..., and the eighth column of the transposed matrix are inputted. Similarly, the elements [0,1], [0,2], ..., [0,7] of matrix B can be obtained, and the first row element of matrix B can be obtained. All columns of the transposed matrix X are weighted multiplied with the second column in the storage unit to obtain the second row of the matrix B. All columns of the transposed matrix X are weighted multiplied with the third, ..., and eighth columns in the storage unit to obtain the third, ..., and eighth rows of the matrix B to obtain the matrix B. After obtaining the matrix B, the first matrix multiplication is completed, that is, B = AX ^T .

The size of the DCT conversion matrix is the first size, and the bit width of the DCT conversion matrix is the second bit width, and the second DCT conversion mode is obtained. When the conversion mode is the second DCT conversion mode, obtaining the first result matrix includes the following steps:

Dividing the element positions of the DCT conversion matrix into a first part of positions and a second part of positions;

The first partial position and the second partial position are written with a weight of the second bit width;

Input each column element of the first rearranged matrix from position 0 and multiply it with each column element of the first part position to generate the third element, which is a 4-bit number;

Connect the two third elements end to end to get the first element;

Perform vertical addition on the first element to obtain the second element;

Perform rounding operation on the second element to keep the element equal to the first element;

The first element is input to the shift adder to obtain a first result matrix.

Exemplarily, for a DCT conversion matrix with a bit width of 4 bits, continue to refer to FIG. 7 , write 4-bit weights in the 0th to 3rd positions of all the original 8 weight positions, and write the same 4-bit weights as in the 0th to 3rd positions in the 4th to 7th positions. Unlike the first DCT transformation mode in which the input matrix is input into the storage unit bit by bit, in this mode, two bits are input at a time for multiplication, namely <0> and <4>, <1> and <5>, <2> and <6>, <3> and <7>. Taking <0> and <4> as an example, the <4>th bit is multiplied with the last 4 bits of the weight to obtain 4 bits, and the <0>th bit is multiplied with the first 4 bits of the weight to obtain 4 bits. Connect the two 4 bits end to end to obtain 8 bits. By analogy, a total of 4 8-bit numbers are obtained. Similar to the first DCT transformation mode, vertical addition is performed, and then shift addition is performed to obtain the first result matrix.

When the size of the DCT conversion matrix is 16×16 and the bit width is 8 bits, the third DCT conversion mode is obtained. The size of the DCT conversion matrix is the second size, and the bit width of the DCT conversion matrix is the first bit width, and the third DCT conversion mode is obtained. When the conversion mode is the third DCT conversion mode, obtaining the first result matrix includes the following steps:

Splitting the first-bit-width DCT conversion matrix into matrices of the first size and storing them in the storage and calculation unit;

Splitting the DCT transformation matrix into a matrix of a first size;

Perform vertical addition on the first elements in every two storage and calculation units to obtain the second element, and add the two second elements to obtain two fourth elements;

Perform rounding operation on the fourth element to keep the element as the first element;

The first element is input to the shift adder to obtain a first result matrix.

For example, referring to FIG. 10 , FIG. 11 , and FIG. 12 , the conversion matrix A with a bit width of 8 bits is divided into four 8×8 matrices, which are respectively placed in four storage units cim_00, cim_01, cim_10, and cim_11. The input transposed matrix is also divided into four 8*8 matrices, and X ^T [0][0] and X ^T [0][1] are sequentially input into cim_00 and cim_10 in a column-by-column manner, and X ^T [1][0] and X T [1][1] are sequentially input into cim_00 and cim_10 in a column-by-column manner. [1][1] of ^T is input to cim_01 and cim_11 in column order; similarly to the first and second DCT transform modes, 8×8 matrix multiplication is performed. According to the matrix multiplication operation rules, before entering the shift addition, 8 8 bits in cim_00 and cim_01 need to be added vertically to obtain 11 bits, and then added again to obtain 12 bits. 8 8 bits in cim_10 and cim_11 need to be added vertically to obtain 11 bits, and then added again to obtain 12 bits. After truncating the decimal places and saturation truncations, 8 bits are guaranteed, and each enters the shift adder; finally, the first result matrix B of the 16×16 matrix AX ^T can be obtained.

The conversion matrix A is different from the transposed matrix, that is, the difference between FIG. 11 and FIG. 12 is that the conversion matrix in FIG. 11 is a matrix with color blocks.

When the size of the DCT conversion matrix is 16×16 and the bit width is 4 bits, the fourth DCT conversion mode is obtained. The size of the DCT conversion matrix is the second size, and the bit width of the DCT conversion matrix is the second bit width, and the fourth DCT conversion mode is obtained. For the 16×16 conversion matrix A with a weight bit width of 4 bits, the calculation and arrangement of each 4-bit weight in the four storage units are the same as the second DCT conversion mode. The data feeding method is the same as the third DCT conversion mode, which will not be repeated here.

S604: Perform a first matrix multiplication operation on the first result matrix and the DCT conversion matrix to obtain a frequency domain signal matrix.

After obtaining the first result matrix through different DCT transformation modes, the matrix multiplication operation is performed again. For the first result matrix obtained through the first DCT transformation mode and the second DCT transformation mode, the transposed matrix ^BT of the first result matrix is used as the previous transposed matrix and stored in the storage unit again to complete the DCT operation of the pixel matrix with a conversion matrix width of 8 bits. For the first result matrix obtained through the third DCT transformation mode, similarly, the second matrix multiplication BX ^T is continued to complete the DCT transformation of the input 16×16 matrix.

S104: Perform data quantization on the frequency domain signal matrix to obtain a quantization matrix.

In this embodiment, the frequency domain signal matrix is a set or arrangement of DCT coefficients. Each DCT coefficient corresponds to a specific frequency component in the frequency domain, and the frequency domain signal matrix contains all DCT coefficients, including a direct current (DC) coefficient and multiple alternating current (AC) coefficients.

To compress data, the DCT coefficients are quantized. Quantization is a lossy process that reduces the precision of the coefficients, thereby reducing the amount of data required for storage and transmission. Because the human eye is less sensitive to details, high-frequency coefficients representing image details will be quantized to a greater extent.

The quantization algorithm is shown in the following formula:

Among them, round represents the rounding function, which is used to round the input. G _ij is the frequency domain coefficient of the input, _Qij is the corresponding quantization step size, _Bij is the quantized output, and i and j represent the position of the elements in the matrix.

The matrix size is 8×8 and the quantization table is as follows:

16 11 10 16 twenty four 40 51 61 12 12 14 19 26 58 60 55 14 13 16 twenty four 40 57 69 56 14 17 twenty two 29 51 87 80 62 18 twenty two 37 56 68 109 103 77 twenty four 35 55 64 81 104 113 92 49 64 78 87 103 121 120 101 72 92 95 98 112 100 103 99

The matrix size is 16×16 and the quantization table is as follows:

S105: Encoding the quantization matrix to form a compressed image format, indicating that compression of the target image is completed.

The quantization matrix can be data encoded, also known as RLC (Run-Length-Coding). For example, the quantization matrix is first subjected to Zig-Zag scanning (zigzag scanning), see FIG. 13, and the quantization matrix is converted into a one-dimensional array for arrangement, so that 0s can be put together as much as possible. Since most of the 0s are concentrated in the lower right corner, the order from the upper left corner to the lower right corner is used. After this order transformation, the matrix becomes an integer array, for example, after Zig-Zag scanning:

Compress the 0s in the one-dimensional DCT factor data after Zig-Zag scanning. For example, 57, 45, 0, 0, 0, 0, 23, 0, -30, -16, 0, 0, 1, 0, 0, 0, 0, 0, only 0, 0. It can be converted by RLC to (0, 57), (0, 45), (4, 23), (1, -30); (0, -16), (2, 1); EOB.

(0, 57) means there are 0 "0"s before 57, (1, -30) means there is 1 "0" before -30, and so on. EOB means all the numbers following are 0. It can also be represented by (0, 0), which becomes (0, 57); (0, 45); (4, 23); (1, -30); (0, -16); (2, 1); (0, 0).

The data after run-length coding will be entropy coded, which can use Huffman Coding or Arithmetic Coding. Entropy coding is a lossless compression technology that assigns codes of different lengths according to the probability of data occurrence to achieve optimal data compression. Compressed image data can be obtained.

It can be understood that the compression method provided in this embodiment can perform encoding on four 8×8 matrices or one 16×16 matrix at the same time.

Based on the above two-dimensional image compression method, some embodiments of the present application provide a two-dimensional image decompression method, referring to FIG. 14 , including:

S1401: Acquire a target image.

Similar to the above-mentioned two-dimensional image compression method, the target image is a two-dimensional image, that is, the target image does not contain depth information, but only contains length and width information of two dimensions, for example: geometric figures, raster images, vector images, etc.

In this embodiment, the target image may be a compressed image obtained according to the above two-dimensional image compression method, and the compressed image format may be: JPEG, PNG, GIF, WebP, TIFF.

S1402: If the target image is a compressed image obtained through processing, decoding is performed on the target image to obtain a decoding matrix.

Data encoding facilitates transmission, and decoding is the inverse process of data encoding to obtain a decoding matrix, wherein the decoding matrix is to restore the quantization matrix. The decompression method of this embodiment can choose to decode four 8×8 matrices or one 16×16 matrix at the same time.

S1403: performing IDCT transformations of different transformation modes according to the bit width and size of the IDCT transformation matrix to form a time domain signal matrix, representing the decompression of the target image.

IDCT transformation is the process of converting the signal from the frequency domain back to the time domain. This process requires the inverse matrix of the DCT transformation matrix to perform the inverse transformation operation. Therefore, before performing the IDCT transformation, it is necessary to predefine the IDCT transformation matrix, which is the inverse matrix of the DCT transformation matrix.

In the IDCT transform, the IDCT transform matrix is multiplied by the frequency domain signal obtained after the DCT transform to obtain the original time domain signal or an approximation. The IDCT transform matrix contains the coefficients required to perform the IDCT, and these coefficients have a specific relationship with the coefficients of the DCT transform matrix to ensure that the inverse transform can be performed correctly.

IDCT is the inverse operation of DCT, which converts the frequency data of the image into pixel data. The formula is as follows:

Among them, F(u,v) is the frequency data of the coordinate (u,v), c(u) and c(v) are compensation coefficients that can make the DCT transformation matrix an orthogonal matrix, and f(i,j) is the pixel data of the coordinate (i,j).

Since the formula of DCT transformation is the same, the transformation formula of IDCT is:

X=A ^T YA=A ^T (A ^T y ^T ) ^T ;

Among them, Y is the frequency domain signal matrix, A is the conversion matrix, X is the pixel matrix, and ^AT is the transposed matrix of the conversion matrix.

In some embodiments, the size of the IDCT conversion matrix includes a first size and a second size, wherein the first size is 8×8 and the second size is 16×16, and the matrix bit width includes a first bit width and a second bit width, wherein the first bit width is 8 bits and the second bit width is 16 bits.

The transposed matrix ^XT of the pixel matrix is obtained, and then the transformation mode is obtained according to the size and bit width of the IDCT transformation matrix. Referring to FIG. 15 , this embodiment provides four IDCT transformation modes. Different IDCT transformation modes can be executed for different matrix sizes and bit widths. Among them, when the size of the IDCT transformation matrix is 16×16 and the bit width is 8 bits or 4 bits, the four IDCT transformation modes use the same IDCT transformation mode, i.e., the third IDCT transformation mode.

When the size of the IDCT conversion matrix is the first size and the bit width of the IDCT conversion matrix is the first bit width, the first IDCT conversion mode is obtained. The process of the first IDCT conversion mode is the same as that of the first DCT conversion mode, see Figure 15. By changing the rearrangement of each column in the figure and following the IDCT sorting method, the first IDCT conversion mode can be realized.

When the size of the IDCT conversion matrix is the first size and the bit width of the IDCT conversion matrix is the second bit width, a second IDCT conversion mode is obtained. The process of the second IDCT conversion mode is the same as that of the second DCT conversion mode, see Figure 15. By changing the rearrangement of each column in the figure and following the IDCT sorting method, the first IDCT conversion mode can be realized.

When the size of the IDCT conversion matrix is the second size and the bit width of the IDCT conversion matrix is the first bit width and the second bit width, obtaining a third IDCT conversion mode, when the conversion mode is the third IDCT conversion mode, includes the following steps:

Input the IDCT conversion matrix into the storage unit column by column;

According to the element position of the IDCT conversion matrix, each column of the transposed matrix is rearranged according to the position of the storage unit to obtain a second rearranged matrix;

dividing the second rearranged matrix of the second size into matrices of the first size;

Perform vertical addition on the first elements in every two storage units to obtain the second element, and add the two second elements to obtain two fourth elements, where the fourth elements are 12-bit numbers.

Perform rounding operation on the fourth element to keep the element as the first element;

Inputting the first element into the shift adder to obtain a second result matrix;

A first matrix multiplication operation is performed on the second result matrix and the IDCT conversion matrix to obtain a time domain signal matrix.

Exemplarily, referring to FIG. 17 , the input matrix is also divided into four 8×8 matrices, and Y ^T [0][0] and Y ^T [0][1] are input into cim_00 and cim_01 in columns, and Y ^T [1][0] and Y ^T [1][1] are input into cim_19 and cim_11 in columns; according to the matrix multiplication operation rule, before entering the shift adder, it is necessary to vertically add 8 8-bit cim_00 and cim_10 to obtain 11 bits again to obtain 12 bits, and vertically add 8 8-bit cim_01 and cim_11 to obtain 11 bits again to obtain 12 bits, and after truncating the decimal places and saturation truncations, 8 bits are guaranteed. After entering the shift adder, each of them obtains the element [0,0] of the new matrix _B1 after A ^T Y ^T multiplication, where matrix B ₁ is the second result matrix, see Figure 9, and so on, input the second column, the third column, ..., the eighth column of the transposed matrix, and similarly, the elements [0,1], [0,2], ..., [0,7] of the matrix B can be obtained, and the first row element of the matrix B ₁ can be obtained. Multiply all the columns of the transposed matrix Y ^T with the weights of the second column in the storage unit to obtain the second row of the matrix B _1. Multiply all the columns of the transposed matrix Y ^T with the weights of the third column, ..., the eighth column in the storage unit to obtain the third row, ..., the eighth row of the matrix B ₁ , and obtain the matrix B _1. After obtaining the matrix B ₁ , the first matrix multiplication is completed, that is, B ₁ = A ^T Y ^T .

Referring to FIG. 16 , after obtaining the second result matrix through different IDCT transform modes, the matrix multiplication operation is performed again. For the second result matrix obtained through the first IDCT transform mode, the second DCT transform mode and the third IDCT transform mode, the transposed matrix B ₁ ^T of the second result matrix is used as the previous transposed matrix and is stored in the storage unit again. The second matrix multiplication B ₁ ^T A ^T is continued to be performed to obtain the time domain signal matrix. The IDCT transform of the 16×16 matrix with a matrix bit width of 8 bits and 4 bits can be completed, and the image decompression is completed.

Some embodiments of the present application also provide a two-dimensional image decompression system, see Figures 18 and 19, including: an acquisition module 1801, a decoding module 1802, a storage and calculation module 1803, a quantization module 1804 and an encoding module 1805;

The acquisition module 1801 is used to acquire a target image;

If the target image is an uncompressed image, decompose the target image into a pixel matrix;

The storage and calculation module 1803 is used to perform DCT transformations of different transformation modes according to the bit width and size of the DCT transformation matrix to obtain a frequency domain signal matrix;

The quantization module 1804 is used to perform data quantization on the frequency domain signal matrix to obtain a quantization matrix;

The encoding module 1805 is used to perform encoding on the quantization matrix to form a compressed image format, indicating that the compression of the target image is completed;

If the target image is a compressed image, the decoding module 1802 is used to perform decoding on the target image to obtain a decoding matrix;

The storage and calculation unit 1803 is also used to perform IDCT transformations of different transformation modes according to the bit width and size of the IDCT transformation matrix to form a time domain signal matrix to represent the decompression of the completed target image.

The present application provides a two-dimensional image compression method, a decompression method and a decompression system, and the system includes an acquisition module 1801, a decoding module 1802, a storage and calculation module 1803, a quantization module 1804 and an encoding module 1805; the acquisition module 1801 is used to acquire a target image; if the target image is an uncompressed image, the target image is decomposed into a pixel matrix; the storage and calculation module 1803 is used to perform DCT transformations of different transformation modes according to the bit width and size of the DCT transformation matrix to obtain a frequency domain signal matrix; the quantization module 1804 is used to perform data quantization on the frequency domain signal matrix to obtain a quantization matrix; the encoding module 1805 is used to perform encoding on the quantization matrix to form a compressed image format, indicating that the compression of the target image is completed; if the target image is a compressed image, the decoding module 1802 is used to perform decoding on the target image to obtain a decoding matrix; the storage and calculation unit 1803 is also used to perform IDCT transformations of different transformation modes according to the bit width and size of the IDCT transformation matrix to form a time domain signal matrix, indicating that the decompression of the target image is completed. Hardware technology based on in-memory computing accelerates the efficiency of matrix operations in DCT/IDCT during two-dimensional image compression and decompression. Based on hardware reconfigurable technology, the same memory computing unit can perform DCT/IDCT calculations, matrix quantization, encoding, and decoding of matrices of different sizes according to configuration signals. The multiplier is replaced by bit-by-bit shift addition to achieve a large number of multiplication operations. At the same time, the calculation bit width can be adjusted to solve the problem of low image compression and decompression efficiency.

Similar parts between the embodiments provided in this application can be referenced to each other. The specific implementation methods provided above are only a few examples under the general concept of this application and do not constitute a limitation on the protection scope of this application. For those skilled in the art, any other implementation methods expanded based on the scheme of this application without creative work belong to the protection scope of this application.

Claims (10)

1. A two-dimensional image compression method, comprising: Get the target image; If the target image is an uncompressed image, decomposing the target image into a pixel matrix; Performing DCT transformations of different transformation modes according to the bit width and size of the DCT transformation matrix to obtain a frequency domain signal matrix; Performing data quantization on the frequency domain signal matrix to obtain a quantization matrix; The quantization matrix is encoded to form a compressed image format, indicating that the compression of the target image is completed. 2. The two-dimensional image compression method according to claim 1, characterized in that the size includes a first size and a second size, and the bit width includes a first bit width and a second bit width; The DCT transformation of different transformation modes is performed according to the bit width and size of the DCT conversion matrix to obtain a frequency domain signal matrix, including: Obtaining a transposed matrix of the pixel matrix; Predefined DCT transformation matrix; Acquire the transform mode according to the size and bit width of the DCT transform matrix; The size of the DCT conversion matrix is a first size, and the bit width of the DCT conversion matrix is a first bit width, obtaining a first DCT transformation mode; The size of the DCT conversion matrix is the first size, and the bit width of the DCT conversion matrix is the second bit width, obtaining a second DCT transformation mode; The size of the DCT conversion matrix is the second size, and the bit width of the DCT conversion matrix is the first bit width, obtaining a third DCT transformation mode; The size of the DCT conversion matrix is the second size, and the bit width of the DCT conversion matrix is the second bit width, and a fourth DCT transformation mode is obtained. 3. The two-dimensional image compression method according to claim 2, characterized in that the DCT transformation of different transformation modes is performed according to the bit width and size of the DCT transformation matrix to obtain the frequency domain signal matrix, comprising: When the transform mode is the first DCT transform mode, the DCT conversion matrix is input into the storage unit by column; According to the element position of the DCT conversion matrix, each column of the transposed matrix is rearranged according to the position of the storage and calculation unit to obtain a first rearranged matrix; Performing a first matrix multiplication operation on the first rearrangement matrix and the conversion matrix to obtain a first result matrix; A first matrix multiplication operation is performed on the first result matrix and the conversion matrix to obtain a frequency domain signal matrix. 4. The two-dimensional image compression method according to claim 2, wherein the performing a first matrix multiplication operation on the first rearrangement matrix and the transformation matrix to obtain a first result matrix comprises: Input each column element of the first rearrangement matrix from bit 0, and multiply it with each column element of the DCT conversion matrix to obtain multiple columns of first elements, where the first elements are 8-bit numbers; Perform vertical addition on the first element to obtain a second element, where the second element is an 11-bit number; Performing a rounding operation on the second element to obtain the first element; The first element is input into a shift adder to obtain a first result matrix. 5. The two-dimensional image compression method according to claim 2, wherein the performing a first matrix multiplication operation on the first rearrangement matrix and the transformation matrix to obtain a first result matrix comprises: When the transform mode is the second DCT transform mode, dividing the element positions of the DCT conversion matrix into a first part of positions and a second part of positions; The first partial position and the second partial position are written with a weight of the second bit width; Input each column element of the first rearranged matrix from position 0, and multiply it with each column element of the first partial position to generate a third element, where the third element is a 4-bit number; Connect the two third elements end to end to obtain the first element; Performing vertical addition on the first element to obtain a second element; Performing a rounding operation on the second element to obtain the first element; The first element is input into a shift adder to obtain a first result matrix. 6. The two-dimensional image compression method according to claim 2, wherein the performing a first matrix multiplication operation on the first rearrangement matrix and the transformation matrix to obtain a first result matrix comprises: When the transform mode is the third DCT transform mode, dividing the transposed matrix of the first bit width into a matrix of the first size and storing it in the storage unit; Splitting the DCT transformation matrix into matrices of a first size; Performing vertical addition on the first elements in every two of the storage and calculation units to obtain a second element, and adding two of the second elements to obtain two fourth elements, where the fourth elements are 12-bit numbers; Performing a rounding operation on the fourth element to obtain the first element; The first element is input into a shift adder to obtain a first result matrix. 7. A two-dimensional image decompression method, characterized by comprising: Get the target image; If the target image is a compressed image obtained by processing the method according to any one of claims 1 to 6, decoding is performed on the target image to obtain a decoding matrix; IDCT transformations of different transformation modes are performed according to the bit width and size of the IDCT transformation matrix to form a time domain signal matrix, which represents the decompression of the target image. 8. The two-dimensional image decompression method according to claim 7, characterized in that the size includes a first size and a second size, and the bit width includes a first bit width and a second bit width; The IDCT transformation of different transformation modes is performed according to the bit width and size of the IDCT transformation matrix to form a time domain signal matrix, including: Get the transposed matrix of the pixel matrix; Predefined IDCT transformation matrix; Obtaining the transformation mode according to the size and bit width of the transposed matrix; The size of the IDCT conversion matrix is a first size, and the bit width of the IDCT conversion matrix is a first bit width, and a first IDCT conversion mode is obtained; The size of the IDCT conversion matrix is the first size, and the bit width of the IDCT conversion matrix is the second bit width, obtaining a second IDCT conversion mode; The size of the IDCT conversion matrix is the second size, and the bit width of the IDCT conversion matrix is the first bit width and the second bit width, and a third IDCT transformation mode is obtained. 9. The two-dimensional image decompression method according to claim 8, characterized in that the IDCT transformation of different transformation modes is performed according to the bit width and size of the IDCT transformation matrix to form a time domain signal matrix, comprising: When the transformation mode is the third IDCT transformation mode, the IDCT transformation matrix is input into the storage and calculation unit by column; According to the element positions of the IDCT conversion matrix, each column of the transposed matrix is rearranged according to the position of the storage and calculation unit to obtain a second rearranged matrix; dividing the second rearranged matrix of a second size into matrices of a first size; Performing vertical addition on every two first elements in the storage and calculation units to obtain a second element, and adding two of the second elements to obtain two fourth elements, where the fourth elements are 12-bit numbers; Performing a rounding operation on the fourth element to obtain the first element; Inputting the first element into a shift adder to obtain a second result matrix; A first matrix multiplication operation is performed on the second result matrix and the IDCT conversion matrix to obtain a time domain signal matrix. 10. A two-dimensional image decompression system, characterized by comprising: an acquisition module, a decoding module, a storage module, a quantization module and an encoding module; The acquisition module is used to acquire the target image; If the target image is an uncompressed image, decomposing the target image into a pixel matrix; The storage and calculation module is used to perform DCT transformations of different transformation modes according to the bit width and size of the DCT transformation matrix to obtain a frequency domain signal matrix; The quantization module is used to perform data quantization on the frequency domain signal matrix to obtain a quantization matrix; The encoding module is used to perform encoding on the quantization matrix to form a compressed image format, indicating that the compression of the target image is completed; If the target image is a compressed image, the decoding module is used to decode the target image to obtain a decoding matrix; The storage and calculation unit is also used to perform IDCT transformations of different transformation modes according to the bit width and size of the IDCT transformation matrix to form a time domain signal matrix to represent the decompression of the target image.