Image compression system and method based on memristor

Abstract

The present invention provides a memristor-based image compression system and method, the image compression system comprising: a first storage subarray, a second storage subarray, an input encoding module, an output encoding module, a storage module, and a frequency signal encoding module; image information is converted into a processable signal by the input encoding module, and a Walsh-Hadamard transform operator is mapped to the storage subarray; after completing the image compression forward transform, the frequency information encoding module compresses the obtained frequency information; the image compression system performs an inverse transform of the image compression, and the current value after the image compression inverse transform is output as compressed image information after passing through the output encoding module, thus completing the image compression. The image compression in the present invention uses the Walsh-Hadamard transform. Since the transform operator only includes +1 and -1, only a memristor device with 1-bit precision is required, which greatly reduces the performance requirements of the device and also reduces the complexity of the device control coding.

Description

Technical Field

[0001] The present invention belongs to the field of microelectronic devices, and more specifically, relates to a memristor-based image compression system and method. Background Art

[0002] Image compression is a crucial technology in image processing, used to reduce image storage space and transmission bandwidth. Many image compression techniques essentially rely on a large number of orthogonal transformations, which are well-suited to the characteristics of memristive memory-computing systems. However, due to the differences between different image compression technologies, each has its own advantages and disadvantages when used in conjunction with memristive memory-computing systems.

[0003] The discrete cosine transform (DCT) is a common image compression technology, but because its operators include floating-point numbers and negative numbers, encoding, mapping, and storage present considerable challenges, especially for memristor arrays. First, at the encoding and mapping level, the maximum precision of existing memristor devices is 7 bits, far less than 32 bits. As the scale of DCT operators increases, the number of bits required for the device continues to increase, and existing memristor arrays cannot meet the needs of accurate encoding and mapping. In terms of storage, even if current memristor arrays meet the requirements of encoding and mapping DCT operators, multi-order regulation is very complex. Therefore, while ensuring image compression quality, it is very important to minimize the complexity and power consumption of array operations. Summary of the Invention

[0004] In view of the defects of the prior art, the purpose of the present invention is to provide an image compression system and method based on memristors, aiming to solve the problem in the prior art that the mapping coding algorithm is very complex and causes high system power consumption.

[0005] The present invention provides a memristor-based image compression system, comprising: a first storage subarray, a second storage subarray, an input encoding module, an output encoding module, a storage module, and a frequency signal encoding module; the first storage subarray and the second storage subarray are of the same size and interconnected, the first storage subarray and the second storage subarray corresponding to a positive input encoding region and a negative input encoding region, respectively, for storing operator elements of image compression transformation; an input end of the input encoding module is connected to the positive input encoding region and the negative input encoding region, for converting a digital image signal into an analog signal processable by the first storage subarray and the second storage subarray; the output encoding module is used to amplify and quantize the current analog signal after operation by the first storage subarray and the second storage subarray into a voltage digital signal; the input end of the storage module is connected to the output end of the output encoding module, for storing intermediate data when the first storage subarray and the second storage subarray perform operations; the input end of the frequency signal encoding module is connected to the output end of the storage module, for processing the frequency information matrix after the image compression positive transformation and for subsequent input of the positive input encoding region and the negative input encoding region, thereby achieving information compression.

[0006] Furthermore, a Walsh-Hadamard transform operator mapping is stored on the first storage sub-array and the second storage sub-array;

[0007] 1 is mapped to the high-conductance state, 0 is mapped to the low-conductance state, +1 is the high-conductance state minus the low-conductance state, and -1 is the low-conductance state minus the high-conductance state.

[0008] Furthermore, the structures of the first storage sub-array and the second storage sub-array are both a cross-bar structure, a transistor-memristor cascade structure, or a single transistor-multi-memristor cascade structure.

[0009] Furthermore, the devices in the first storage sub-array and the second storage sub-array are resistive random access memory, phase change memory, random access memory, NOR Flash device or NAND Flash device.

[0010] The present invention also provides an image compression method based on the above-mentioned image compression system, comprising the following steps:

[0011] Inputting image information as a voltage and converting it into an analog signal that can be processed by the first storage sub-array and the second storage sub-array through an input coding module;

[0012] The Walsh-Hadamard transform operator mapping is stored in the first storage subarray and the second storage subarray and a matrix-vector multiplication operation is performed; wherein the voltage input signal of the positive input coding region is positive, and at the corresponding position, the voltage input signal of the negative input coding region has the same amplitude as that of the positive input coding region but opposite sign;

[0013] After performing the matrix-vector multiplication operation, the output result passes through the output encoding module and the current operation result is stored.

[0014] Furthermore, after the image compression forward transform is complete, it needs to be processed by the frequency signal encoding module. Because the human eye is sensitive to high-frequency information and tends to ignore low-frequency information, the frequency signal encoding module retains most of the high-frequency information and removes most of the low-frequency information, thus completing the information compression. Frequency signal encoding methods include but are not limited to mask matrix processing and threshold setting methods.

[0015] Furthermore, the mapping method of the Walsh-Hadamard transform operator on the memristor array is as follows:

[0016] During mapping, 1 is mapped to a high-conductance state, 0 is mapped to a low-conductance state, +1 is the high-conductance state minus the low-conductance state, and -1 is the low-conductance state minus the high-conductance state.

[0017] Furthermore, the specific operation for implementing negative number arithmetic on the first and second storage subarrays is as follows: 0 is stored in the first storage subarray, i.e., a low-conductance state, and a positive signal V is input; 1 is stored in the corresponding position of the second storage subarray, i.e., a high-conductance state, and a negative signal V is input. The two arrays are added together to implement the negative number arithmetic of 0×V-1×V=-V. This approach avoids the need for a subtractor, thereby reducing system power consumption.

[0018] The image compression method employed in this invention utilizes the Walsh-Hadamard transform. Because the transform operators consist solely of +1 and -1, only a single-bit precision memristor is required. This significantly reduces the performance requirements for the device and the complexity of the device control coding. Given the current limitations of device performance, this method offers significant advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 is a structural diagram of a memristor-based image compression system provided by an embodiment of the present invention;

[0020] Figure 2 This is a schematic diagram of the process operation of image compression transformation provided by an embodiment of the present invention;

[0021] Figure 3 Schematic diagram of the storage and operation format of the image compression transformation operator provided by an embodiment of the present invention;

[0022] Figure 4 It is a schematic diagram of processing the frequency matrix obtained after the forward transformation of image compression by the image compression system provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0024] Current memristor-based image compression transformation implementations include discrete cosine transforms and wavelet transforms. On the one hand, these transform operators include floating-point or complex numbers, making their mapping and encoding algorithms on memristive devices very complex. On the other hand, the order of these transform operators increases significantly with image resolution, placing high demands on the device's memory. The Walsh-Hadamard transform, an image transformation module used in this image compression system, requires only two operators, +1 and -1. This transform can be well integrated with memristive device systems to address the high complexity of existing image compression operator encoding, mapping, and storage.

[0025] like Figure 1 As shown, the present invention provides an image compression system based on a memristor, comprising: a first storage subarray and a second storage subarray of the same size and connected to each other, an input encoding module, an output encoding module, a storage module, and a frequency signal encoding module;

[0026] The two memristor arrays are of exactly the same size, corresponding to the size of the selected operator, and are used to store operator elements for image compression transformation. The two arrays correspond to the positive input encoding area and the negative input encoding area respectively;

[0027] The input encoding module is first used to sample the digital signal of the image, quantize it to the required accuracy, and then maintain and encode it into an analog signal that can be processed by the memristor array, including but not limited to modules with similar functions such as digital-to-analog converters and current-to-voltage converters;

[0028] The output encoding module is first used to sample the current analog signal after the memristor array operation, amplify it to a suitable range, quantize it to the required accuracy, and then maintain and encode it into a voltage digital signal, including but not limited to a transimpedance amplifier, an analog-to-digital converter, or a module with similar functions;

[0029] The storage module is used to store the intermediate data after the memristor array performs operations. Since the forward and inverse transformations of the image compression algorithm involve two matrix-vector multiplications, the intermediate data of the first matrix-vector multiplication needs to be cached and used for the second matrix-vector multiplication.

[0030] The frequency signal encoding module processes the frequency information matrix after the image compression forward transform. Because the human eye is sensitive to high-frequency information and tends to ignore low-frequency information, the frequency information encoding module retains most of the high-frequency information and removes most of the low-frequency information, thus completing the information compression. Frequency signal encoding methods include, but are not limited to, mask matrix processing and threshold setting methods. The mask matrix method multiplies the frequency information matrix by the mask matrix to retain the desired frequency information points. The threshold setting method determines the compression threshold based on the compression requirements and only retains frequency information points above the threshold. This method is relatively more accurate, but also more complex.

[0031] The present invention also provides an image compression method based on a memristor, comprising the following steps:

[0032] The image information is input as a voltage and passed through the input encoding module to be converted into an analog signal that can be processed by the memristor array;

[0033] The Walsh-Hadamard transform operator mapping is stored on two memristor arrays of equal size, and matrix-vector multiplication is performed. The voltage input signal of the positive input coding area is positive, and at the corresponding position, the voltage input signal of the negative input coding area has the same amplitude as the positive input coding area, but with opposite signs.

[0034] After performing the matrix-vector multiplication operation, the output is passed through the output encoding module and stored. After the forward transform for image compression is completed, the image must also pass through the frequency signal encoding module. Because the human eye is sensitive to high-frequency information and tends to ignore low-frequency information, the frequency signal encoding module retains most of the high-frequency information and removes most of the low-frequency information, thus completing the information compression. Frequency signal encoding methods include, but are not limited to, mask matrix processing and threshold setting.

[0035] The mapping method of the Walsh-Hadamard transform operator on the memristor array is:

[0036] During mapping, since the image compression operator only includes +1 and -1, only a 1-bit precision memristor is required. 1 is mapped to a high-conductance state, and 0 is mapped to a low-conductance state. Thus, +1 is the high-conductance state minus the low-conductance state, and -1 is the low-conductance state minus the high-conductance state.

[0037] The specific operations for implementing negative number operations on two memristor arrays are:

[0038] One array stores 0, or a low-conductance state, and receives a positive signal V. The other array stores 1, or a high-conductance state, and receives a negative signal V. The two arrays are added together to achieve the negative number operation of 0 × V - 1 × V = - V. This solution avoids the need for a subtractor, thus reducing system power consumption.

[0039] like Figure 2 As shown, the image compression algorithm in this invention primarily consists of two parts: a forward transform and an inverse transform. Each transform involves two matrix-vector multiplications. During the forward transform, the original image's pixel matrix is first right-multiplied and then left-multiplied by the Walsh-Hadamard transform matrix. These two matrix-vector multiplications yield the image's frequency information. After the forward transform is complete, the frequency matrix is processed by a frequency matrix encoding module.

[0040] Because the human eye is sensitive to high-frequency information and tends to ignore low-frequency information, the frequency information encoding module retains most of the high-frequency information and removes most of the low-frequency information, thus achieving information compression. The processed frequency matrix is then input into the encoding module for an inverse compression transform. During the inverse transform, the information compression frequency matrix also undergoes two matrix-vector multiplications. First, it is multiplied on the right by the Walsh-Hadamard transform matrix, and then on the left to obtain the compressed image information. This completes the image compression process.

[0041] Example 1

[0042] like Figure 3 The figure shows a method for mapping an image compression operator onto a memristor array, using the 16×16 Walsh-Hadamard transform as an example. Because the operator matrix involves negative numbers, which memristor devices cannot map, the mapping method of the present invention is introduced. The 16×16 Walsh-Hadamard transform operator is split into two 16×16 memristor arrays for storage: the first storage subarray and the second storage subarray. Since the image compression operator only includes +1 and -1, only 1-bit precision memristor devices are required for mapping. 1 is mapped to a high conductance state, and 0 is mapped to a low conductance state. Therefore, +1 is the high conductance state minus the low conductance state, and -1 is the low conductance state minus the high conductance state. Therefore, in the first storage subarray, +1 is stored as 1, and -1 is stored as 0. In the second storage subarray, +1 is stored as 0, and -1 is stored as 1. The first storage subarray corresponds to the positive-coding input region, and the second storage subarray corresponds to the negative-coding input region. Figure 3 V1, V2, V3, etc. in the memory are image pixel information input through the input coding module. The voltage signals input at the corresponding positions of the two memory arrays have the same amplitude but opposite polarity, thereby realizing negative number operation on the memory array.

[0043] like Figure 4As shown, the present invention provides one example method of a frequency matrix encoding module. After the image compression forward transform is completed, the output needs to be processed by the frequency matrix encoding module. Because the human eye is sensitive to high-frequency information and tends to ignore low-frequency information, the frequency information encoding module retains most of the high-frequency information and eliminates most of the low-frequency information, thereby completing the information compression. This frequency signal encoding method includes but is not limited to mask matrix processing and threshold setting methods.

[0044] A method for mask matrix processing is provided in Example 1 of the present invention. The size of the mask matrix is consistent with the size of the frequency matrix, and the specific value of the mask matrix is specified according to the target of image compression. The size of the frequency matrix in Example 1 of the present invention is 8×8, so the size of the mask matrix is also 8×8. After the positive transformation of image compression, most of the high-frequency information is concentrated in the upper left of the frequency matrix, so we retain the frequency information in the upper left as much as possible. The mask matrix we selected has 6 spatial values of 1 in the upper left, and the rest are 0. After the mask matrix is dot-multiplied with the frequency matrix, only the 6 frequency information in the upper left of the frequency matrix is retained, thereby completing the compression processing of the frequency information.

[0045] In summary, the present invention has the following advantages:

[0046] The present invention implements negative numbers by storing 0, or a low-conductance state, on one array and inputting a positive signal V. The other array stores 1, or a high-conductance state, and inputting a negative signal V. The two arrays are then added together to achieve the negative number operation of 0×V-1×V=-V. This solution avoids the use of a subtractor, thereby reducing system power consumption.

[0047] The Walsh-Hadamard transform, an image transformation module used in the image compression system of this invention, requires only a single-bit memristor device because its operators consist solely of +1 and -1. This significantly reduces the performance requirements for the device and the complexity of the device control coding. Given the current limitations of device performance, this approach offers significant advantages.

[0048] It will be easily understood by those skilled in the art that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A memristor-based image compression system, characterized in that: include: A first storage subarray, a second storage subarray, an input encoding module, an output encoding module, a storage module, and a frequency signal encoding module; The first storage subarray and the second storage subarray have the same size and are connected to each other. The first storage subarray and the second storage subarray correspond to the positive input coding area and the negative input coding area, respectively, and are used to store operator elements of image compression transformation. The input end of the input coding module is connected to the positive input coding area and the negative input coding area, and is used to convert the digital signal of the image into an analog signal that can be processed by the first storage sub-array and the second storage sub-array; The output encoding module is used to amplify and quantize the current analog signal after the operation of the first storage sub-array and the second storage sub-array into a voltage digital signal; The input end of the storage module is connected to the output end of the output coding module, and is used to store intermediate data when the first storage sub-array and the second storage sub-array perform operations; The input end of the frequency signal encoding module is connected to the output end of the storage module, and is used to process the frequency information matrix after the positive transformation of the image compression, and is used for the subsequent input of the positive input encoding area and the negative input encoding area to achieve information compression; Walsh-Hadamard transform operator mapping is stored in the first storage subarray and the second storage subarray; 1 is mapped to a high conductivity state, 0 is mapped to a low conductivity state, +1 is a high conductivity state minus a low conductivity state, and -1 is a low conductivity state minus a high conductivity state; In the first storage subarray, +1 is stored as 1 and -1 is stored as 0; in the second storage subarray, +1 is stored as 0 and -1 is stored as 1; the first storage subarray corresponds to the positive coding input area and the second storage subarray corresponds to the negative coding input area.

2. The image compression system according to claim 1, wherein: The structures of the first storage sub-array and the second storage sub-array are both a cross-bar structure, a transistor-memristor cascade structure, or a single transistor-multi-memristor cascade structure.

3. The image compression system according to claim 1 or 2, wherein: The devices in the first storage sub-array and the second storage sub-array are resistive random access memory, phase change memory, self-selected transfer torque random access memory, NOR Flash device or NAND Flash device.

4. An image compression method implemented based on the image compression system according to any one of claims 1 to 3, characterized in that: The steps include: Inputting image information as a voltage and converting it into an analog signal that can be processed by the first storage sub-array and the second storage sub-array through an input coding module; The Walsh-Hadamard transform operator mapping is stored in the first storage subarray and the second storage subarray and a matrix-vector multiplication operation is performed; wherein the voltage input signal of the positive input coding region is positive, and at the corresponding position, the voltage input signal of the negative input coding region has the same amplitude as that of the positive input coding region but opposite sign; After performing the matrix-vector multiplication operation, the output result passes through the output encoding module and the current operation result is stored.

5. The image compression method according to claim 4, wherein: When the forward transform of image compression is completed, it also needs to be processed by the frequency signal encoding module.

6. The image compression method according to claim 4, wherein: The mapping method of the Walsh-Hadamard transform operator on the memristor array is as follows: During mapping, 1 is mapped to a high-conductance state, 0 is mapped to a low-conductance state, +1 is the high-conductance state minus the low-conductance state, and -1 is the low-conductance state minus the high-conductance state.

7. The image compression method according to any one of claims 4 to 6, wherein: The specific operations for implementing negative number operations on the first storage sub-array and the second storage sub-array are: The first storage subarray stores 0, i.e., a low-conductance state, and a positive signal V is input. The corresponding position of the second storage subarray stores 1, i.e., a high-conductance state, and a negative signal V is input. The two arrays are added together to realize the negative number operation of 0×V-1×V=-V.

Images