Image enhancement method and system of spatial vision neural network model

Abstract

The invention relates to an image enhancement method for constructing a spatial vision neural network model. The method comprises the following steps: firstly, carrying out normalization processing on a binarized noise-containing two-dimensional image to enable the binarized noise-containing two-dimensional image to meet small parameter limitation brought by activation of an FHN neuron adiabatic approximation condition; secondly, the space signal is subjected to convolution through an FHN neuron template, and a space neuron pulse potential sequence two-dimensional image coding signal is obtained; constructing an anisotropic spread function based on improved bidirectional adaptive selection as a visual receptive field model with a direction selection mechanism, and carrying out convolution on a neuron pulse coded signal and the constructed receptive field model; and finally, an optimal binary image result is found. According to the method, the direction selectivity and the neuron pulse issuing mechanism of the biological vision system are effectively combined, so that the biological vision system conforms to the physiological characteristics of real vision neuron pulse triggering and signal selectivity, and the ideal image denoising effect is achieved.

Description

Technical Field

[0001] The invention belongs to the field of image processing, and in particular relates to an image enhancement method and system of a spatial visual neural network model. Background Art

[0002] In the process of image acquisition, interference signals are often mixed in. For example, the sensor material properties, working environment, electronic components and circuit structure will introduce interference noise. Therefore, image denoising is an important part of image processing, and the quality of its processing results is related to the subsequent image feature extraction, image analysis and image recognition classification. At the same time, visual physiological experiments and model simulations show from a macro perspective that the visual system has corresponding directional selectivity when observing or identifying targets, and does not respond equally to stimuli in all directions.

[0003] For example, the visual perception system perceives the light intensity changes in the horizontal or vertical direction the most, but perceives the light intensity in the oblique direction the least. In addition, from a microscopic perspective, there are clusters of neurons sensitive to different directions in the biological visual system. The method and system can produce a stronger response to stimulation of a specific contour or motion direction, and a weaker response in other directions.

[0004] Traditional neurons use their own pulse characteristics to perform denoising, which requires dimensionality reduction, which is not in line with the physiological mechanism of visual space. The image has no spatial structural correlation during the dimensionality reduction process, which leads to the loss of spatial correlation information. Dimensionality reduction also causes pixel drift due to the time lag effect. Summary of the invention

[0005] The purpose of the present invention is to provide an image enhancement method for a spatial visual neural network model, so that the present invention solves the problem that the image has no spatial structural correlation during the dimensionality reduction process, resulting in the loss of spatial correlation information, and the pixel drift phenomenon caused by the time lag effect caused by dimensionality reduction.

[0006] To solve the above problems, the first aspect of the present invention provides an image enhancement method of a spatial visual neural network model, which includes: normalizing a noisy binary image; constructing a 3*3 spatial FHN neural network convolution kernel template based on the processed binary image; and obtaining a spatial neuron pulse potential signal corresponding to the binary image by passing a noisy signal through the FHN neural network convolution kernel: Traversing the spatial neuron pulse potential signal of the binary image, selecting the current center value point with 8 surrounding directions in turn; judging the difference between the center value of the current center value point and the 8 surrounding directions through the corresponding local variance function or substitution function, and constructing the visual receptive field diffusion coefficient with a direction selection mechanism; Substituting the visual receptive field diffusion coefficient into the diffusion equation; calculating the mean and local variance of the eight directions relative to the center value point; and obtaining the diffusion coefficient by selecting from the local variance function or the substitute function by judging the difference between the mean and the center value; The spatial neuron pulse potential signal of the binary image is subjected to multiple spatial convolution filtering by using the visual receptive field diffusion coefficient; in each filtering, multiple optimal algorithm iterations are performed by using a quantum particle swarm algorithm to obtain a processed image.

[0007] A second aspect of the present invention provides an image enhancement system of a spatial visual neural network model, comprising: The pulse potential signal generating unit is configured to perform normalization processing on the noisy binary image; construct a 3*3 spatial FHN neural network convolution kernel template based on the processed binary image; obtain a spatial neuron pulse potential signal corresponding to the binary image by passing the noisy signal through the FHN neural network convolution kernel: A diffusion coefficient unit is configured to traverse the spatial neuron pulse potential signal of the binary image, and sequentially select a current center value point with 8 surrounding directions; according to the difference value between the center value of the current center value point and the 8 surrounding directions, determine the corresponding local variance function or substitution function to construct a visual receptive field diffusion coefficient with a direction selection mechanism; A selection unit is configured to substitute the visual receptive field diffusion coefficient into a diffusion equation; calculate the mean and local variance of eight directions relative to the center value point; and select the diffusion coefficient from the local variance function or the substitute function by determining the difference between the mean and the center value; The graphics processing unit is configured to perform multiple spatial convolution filtering on the spatial neuron pulse potential signal of the binary image through the visual receptive field diffusion coefficient; in each filtering, multiple optimal algorithm iterations are performed through the quantum particle swarm algorithm to obtain a processed image.

[0008] The third aspect of the present invention provides an electronic device, which includes a processor; and a memory, which stores computer instructions of an image enhancement method of a spatial visual neural network model. When the computer instructions are executed by the processor, the electronic device executes the image enhancement method of the spatial visual neural network model in the present invention.

[0009] The fourth aspect of the present invention provides a computer-readable storage medium, a computer-readable storage medium, comprising program instructions for an image enhancement method of a spatial visual neural network model, and when the program instructions are executed by a processor, the method for implementing the image enhancement method of the spatial visual neural network model according to the present invention is implemented.

[0010] The above technical solution of the present invention has the following beneficial technical effects: By constructing a bidirectional adaptive filter as an anisotropic diffusion equation filter as a visual receptive field selection mechanism, it is possible to select the difference direction, thereby satisfying the directional selectivity of the visual perception system and effectively improving the previous disadvantage that it is difficult to filter out mutation noise signals with large differences.

[0011] Based on the FHN neuron model, a 3*3 neuron network convolution kernel template is constructed. Direct spatial potential pulse encoding does not require dimensionality reduction, retains the adjacent correlation of spatial information, and avoids the occurrence of offset and time lag. The neuron spatial filter kernel conforms to the two-dimensional spatial characteristics of vision and generates corresponding pulse potentials for the input signal. The neuron network can retain the main characteristics of neuron pulse emission and has neighborhood spatial correlation, so that the image potential encoding is more in line with the real visual physiological characteristics.

[0012] The quantum particle swarm optimization iterative algorithm is used to efficiently and conveniently obtain the optimal parameters and output results. The optimal fitness function is selected as the quantitative evaluation indicator in each iteration process, and different diffusion functions are adaptively selected according to the characteristics of the center value, which is conducive to retaining the spatial structure characteristics of the image in a specific direction and the correlation of the pixel values ​​in this direction, as well as the reliability of noise removal.

[0013] By constructing the FHN neuron spatial network convolution kernel, highlighting the information features, re-encoding the potential, selecting the diffusion function to differentially select 8 directions at the center point with a direction selection mechanism, and constructing a two-way selection adaptive visual receptive field model, the real signal is effectively retained, the noise signal is selectively removed, and high-quality images are effectively output. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] Figure 1 It is a schematic flow chart for illustrating an image enhancement method of a spatial visual neural network model in one embodiment of the present invention; Figure 2 It is a schematic diagram for explaining the composition of an image enhancement system of a spatial visual neural network model in one embodiment of the present invention; Figure 3 It is a schematic diagram for illustrating the composition of an image enhancement method of a spatial visual neural network model in another embodiment of the present invention; Figure 4 It is a schematic diagram for illustrating a processing of a bidirectional adaptive selection model of a visual receptive field neural network in one embodiment of the present invention; Figure 5 It is a QPSO image processing optimization curve diagram used to illustrate one embodiment of the present invention; Figure 6 It is a comparison diagram for illustrating one embodiment of the present invention and other methods; Figure 7 It is a schematic structural diagram of an electronic device provided by one embodiment of the present invention. DETAILED DESCRIPTION

[0015] In order to make the purpose, technical scheme and advantages of the present invention clearer, the present invention is further described in detail below in conjunction with specific embodiments and with reference to the accompanying drawings. It should be understood that these descriptions are only exemplary and are not intended to limit the scope of the present invention. In addition, in the following description, the description of well-known structures and technologies is omitted to avoid unnecessary confusion of the concept of the present invention.

[0016] The present invention takes the following factors into consideration: first, the traditional neuron model needs to reduce the two-dimensional signal to a one-dimensional signal for denoising. In the process of dimensionality reduction, the correlation of the spatial structure is lost, and pixel drift is also caused by the time lag effect caused by the time dependency of the single direction in the process of dimensionality reduction and denoising.

[0017] Secondly, the traditional image denoising method has no selection mechanism for noise signals. The traditional partial differential equation image denoising method can remove image noise while maintaining image details such as edges and textures, but it will produce a step effect in the smooth area of ​​the image. Although the difference between the surrounding elements and the central elements is taken into account, it is not effective for granular noise or strong noise. It does not consider the direction selection characteristics of the human visual system and the visual physiological characteristics of the neuron action potential pulse emission. The image denoising results are usually difficult to meet the requirements consistent with human subjective evaluation. Therefore, combining the direction selection characteristics of the visual system with the pulse emission mechanism of neurons, an improved visual receptive field anisotropic diffusion function is constructed, and an image enhancement method of a spatial visual neural network model is proposed to make it conform to the binary image denoising working mechanism of visual physiological characteristics.

[0018] In one aspect of the embodiments of the present invention, Figure 1 As shown, a method for image enhancement of a spatial visual neural network model is provided, which includes the following steps: Step S101, this step includes normalizing the noisy binary image. Based on the processed binary image, a 3*3 spatial FHN neural network convolution kernel template is constructed. The noisy signal passes through the FHN neural network convolution kernel to obtain a spatial neuron pulse potential signal corresponding to the binary image.

[0019] The noisy binary image is normalized, a 3*3 spatial FHN neural network convolution kernel template is constructed, and the noisy signal is pulse-coded by the neural network convolution kernel to obtain a two-dimensional image coding signal of a spatial neuron pulse potential sequence.

[0020] Step S102, this step includes traversing the spatial neuron pulse potential signal of the binary image, and selecting the current center value point with 8 surrounding directions in turn. According to the difference between the center value of the current center value point and the 8 surrounding directions, it is determined through the corresponding local variance function or substitution function to construct a visual receptive field diffusion coefficient with a direction selection mechanism.

[0021] Construct a visual receptive field diffusion coefficient with a direction selection mechanism, and automatically select the local variance and the substitution function according to the difference between the current center value and the surrounding 8 directions. The local variance is for the center value with small difference, and the substitution function is for the diffusion coefficient selection made by the center point with large difference. The noise is often abrupt, and the substitution function can effectively remove isolated noise points.

[0022] Step S103, this step includes substituting the diffusion coefficient of the visual receptive field into the diffusion equation. Calculating the mean and local variance of the eight directions relative to the center value point. By determining the difference between the mean and the center value, the diffusion coefficient is obtained from the local variance function or the alternative function.

[0023] The constructed improved bidirectional adaptive selected anisotropic diffusion function is substituted into the diffusion equation, and the mean and local variance of eight directions relative to the center point are calculated. The local variance or alternative function is selected as the diffusion coefficient by judging the difference between the mean and the center value, and spatial convolution filtering is performed on the pulse sequence two-dimensional image coding signal.

[0024] Step S104, this step includes performing multiple spatial convolution filtering on the spatial neuron pulse potential signal of the binary image through the diffusion coefficient of the visual receptive field. In each filtering, multiple optimal algorithm iterations are performed through the quantum particle swarm algorithm to obtain a processed image.

[0025] Initialize the quantum particle swarm population, initialize the best fitness function, use the peak signal-to-noise ratio for the fitness function, calculate the peak signal-to-noise ratio of the two-dimensional image obtained by the anisotropic diffusion equation in the iteration, update the best fitness function, determine whether the optimal peak signal-to-noise ratio is obtained, and then decide the number of iterations. Use the quantum particle swarm algorithm to perform multiple optimal algorithm iterations on the results after each processing to obtain the best peak signal-to-noise ratio and the processed image.

[0026] In one embodiment of the image enhancement method of the spatial visual neural network model of the present invention, obtaining the spatial neuron pulse potential signal corresponding to the binary image in step S101 includes: pulse encoding the noisy signal through the neural network convolution kernel to obtain the spatial neuron pulse potential signal corresponding to the binary image.

[0027] In another embodiment of the image enhancement method of the spatial visual neural network model of the present invention, the step S102 determines the difference between the center value of the current center value point and the surrounding eight directions, and constructs the visual receptive field diffusion coefficient with a direction selection mechanism through the corresponding local variance function or substitution function, including: Get the difference between the center value of the current center value point and the surrounding 8 directions, and judge whether the difference value is less than the set threshold. If so, construct the visual receptive field diffusion coefficient with direction selection mechanism through the local variance function. If not, construct the visual receptive field diffusion coefficient with direction selection mechanism through the substitution function.

[0028] In another embodiment of the image enhancement method of the spatial visual neural network model of the present invention, in step S104, multiple optimal algorithm iterations are performed through the quantum particle swarm algorithm in each filtering.

[0029] Initialize the quantum particle swarm population, initialize the best fitness function, use the peak signal-to-noise ratio as the fitness function, calculate the peak signal-to-noise ratio of the two-dimensional image obtained by the anisotropic diffusion equation in each filtering, update the best fitness function, determine whether the optimal peak signal-to-noise ratio is obtained, and then decide the number of iterations.

[0030] In one embodiment of the image enhancement method of the spatial visual neural network model of the present invention, the mathematical expression of the FHN neuron is shown in Formula 1: Formula 1 In the formula, u is the fast-changing membrane voltage of the neuron. v is the neuron's slow-changing recovery voltage. ε(0<ε=1) is the time constant that determines the firing rate of the neuron. a is the threshold value for transition from single potential well to double potential well, a and b are all constants, and together they determine the shape of the double potential well.

[0031] In one embodiment of the image enhancement method of the spatial visual neural network model of the present invention, the receptive field model of the diffusion equation is shown in the following formula 2: Formula 2 In the formula u 0 ( x , y ) is the initial image, div is the divergence operator, t is the scale factor, |▽ u | is the modulus of the image gradient, c 1 represents the local variance diffusion coefficient function,c 2 represents the surrogate diffusion coefficient function, u 1 and u 2 Indicates which diffusion coefficient function coefficient is selected, the value is 0 or 1, and the sum is 1. The threshold of the difference between the center point value and the mean value of the eight directions is 0.2. If it is less than 0.2, the local variance is selected, otherwise the alternative function is selected as the coefficient function.

[0032] A second aspect of the present invention, as Figure 2 As shown, an image enhancement system of a spatial visual neural network model is provided, which includes: The pulse potential signal generating unit 101 is configured to perform normalization processing on the noisy binary image. Based on the processed binary image, a 3*3 spatial FHN neural network convolution kernel template is constructed. The spatial neuron pulse potential signal corresponding to the binary image is obtained by passing the noisy signal through the FHN neural network convolution kernel: The diffusion coefficient unit 201 is configured to traverse the spatial neuron pulse potential signal of the binary image, and sequentially select the current center value point with 8 surrounding directions. According to the difference between the center value of the current center value point and the 8 surrounding directions, the corresponding local variance function or substitution function is judged to construct the visual receptive field diffusion coefficient with a direction selection mechanism.

[0033] The selection unit 301 is configured to substitute the diffusion coefficient of the visual receptive field into the diffusion equation, calculate the mean and local variance of the eight directions relative to the center value point, and select the diffusion coefficient from the local variance function or the substitute function by judging the difference between the mean and the center value.

[0034] The graphics processing unit 401 is configured to perform multiple spatial convolution filtering on the spatial neuron pulse potential signal of the binary image through the diffusion coefficient of the visual receptive field. In each filtering, multiple optimal algorithm iterations are performed through the quantum particle swarm algorithm to obtain a processed image.

[0035] In another embodiment of the image enhancement system of the spatial visual neural network model of the present invention, the pulse potential signal generating unit 101 is further configured to: pulse encode the noisy signal through the neural network convolution kernel to obtain a spatial neuron pulse potential signal corresponding to the binary image.

[0036] In another embodiment of the image enhancement method of the spatial visual neural network model of the present invention, Figure 3 As shown, the corresponding steps include: Step (1) for the image in the field of view I ( i , j) ( i = 1,2,… M ; j = 1,2,… N ,variable i and j The same below) is normalized, I ( i , j ) = S ( t )+ξ( t ), S ( t ) is the input signal, ξ ( t ) is Gaussian white noise. Its expression is shown in formula (1): (1) in, f is the normalized image; I ( i, j ) is the original noisy image; max( I ( i , j ))express I ( i , j )’s maximum value; M is the number of rows of the image; N is the number of columns of the image.

[0037] Step (2) constructs a 3*3 FHN neural network convolution kernel template, and pulse encodes the noisy signal through the neural network convolution kernel to obtain a two-dimensional image encoding signal of a spatial neuron pulse potential sequence.

[0038] The FHN neuron model retains the characteristics of the regeneration excitation mechanism of excitable nerve cells and is used to describe the existence of the excitation threshold of nerve and optic nerve electrical pulses, the generation of pulse chains under the action of external current, and other characteristics. The mathematical expression of the FHN neuron is shown in formula (2): (2) In the formula, u is the fast-changing membrane voltage of the neuron; v is the neuron’s slow-changing recovery voltage; ε (0<ε=1) is the time constant that determines the firing rate of the neuron; a is the threshold value for transition from single potential well to double potential well, a and b are all constants, and together they determine the shape of the double potential well. The stochastic resonance mechanism of the neuron itself works together with the external periodic signal and noise to produce a synergistic effect, thereby achieving signal enhancement, detection, denoising and encoding.

[0039] The coding characteristics of neurons are determined by their own threshold characteristics. If the membrane potential is lower than the threshold, no pulses will be emitted. Only when the membrane potential is higher than the threshold, pulses will be emitted. The mathematical expression of pulse emission is shown in formula (3): (3) in, th is the threshold value; f ( i , j ) is the normalized external noisy signal of formula (2). After encoding, the image can be converted from the external electrical pulse sequence into a symbol sequence.

[0040] Step (3) Perona and Malik The anisotropic diffusion equation PM is proposed as formula (4): (4) in: u 0 (x,y) is the initial image, div is the divergence operator, t is the scale factor, |▽ u | is the modulus of the image gradient, c (|▽ u |) is the diffusion coefficient function, also known as the gradient modulus function, which satisfies: c (|▽ u |) is about |▽ u |Non-negative monotonic function, |▽ u |→0, c (|▽ u |)=0;|▽ u |→∞ , c (|▽ u |) = 1. Usually c ( x )= k 2 / ( x 2 + k 2 )or c ( x )=exp(-( x 2 / k 2 )). In the formula, k is the decreasing rate constant. c (|▽ u |)Diffusion coefficients of different scales are generated according to different local gradients.

[0041] The diffusion coefficient function is improved by using the normalized local mean and gradient method. f ( i , j ) is the normalized image, f ( i , j ) is the center of any point, and the local grayscale mean in a square area of ​​3×3 is: (5) The local variance is as follows: (6) When the image is not contaminated by noise, the average value of each pixel in the smooth area is 2 ( i , j ) is very close to the pixel gray value, so the local variance is small; while the gray value of isolated noise pixels or edge points varies greatly, and the gray value of some pixels is greatly different from the average value in the region, so the local variance δ 2 ( i , j ) is very big.

[0042] If there is noise, there will be pixels in the smooth area and edge area with local variance δ 2 ( i , j ) are both increased, so in this case the local variance δ 2 ( i , j ) can no longer effectively measure the spatial details, so the diffusion coefficient function is reconstructed using an alternative function ω( i , j ) Replace the central value based on the maximum statistical value replacement algorithm.

[0043] The absolute values ​​of the differences between the eight directions and the center point are set as shown in formula (7): (7) The substitution function ω( i , j ) As shown in formula (8): (8) In the formula δ represents the statistical difference, Θ represents the pixel value corresponding to the maximum frequency statistic replacing the central value function.

[0044] The diffusion coefficient function of the local variance is as follows: (9) Another scale diffusion coefficient substitution function is as follows: (10) Schematic diagram of the processing of the bidirectional adaptive selection model of the visual receptive field neural network, as shown in Figure 4 As shown, the two diffusion coefficient functions are combined to construct the anisotropic diffusion equation receptive field model of the bidirectional selection adaptive filter as shown in formula (11): (11) In the formula u 0 ( x,y ) is the initial image, div is the divergence operator, t is the scale factor, |▽ u | is the modulus of the image gradient, c 1 represents the local variance diffusion coefficient function, c 2 represents the surrogate diffusion coefficient function, u 1 and u 2 Indicates which diffusion coefficient function coefficient is selected, the value is 0 or 1, and the sum is 1. The threshold of the difference between the central 11 points and the mean of the 8 directions is 0.2. If it is less than 0.2, the local variance is selected, otherwise an alternative function is selected as the coefficient function.

[0045] The discretization numerical iteration form is as follows, considering the eight neighborhoods around the pixel point, as shown in formula (12): (12) in G m It is a symbol representing 8 directions and has the characteristic of direction selection. c Represents the diffusion coefficient function, which has two scales and is judged based on the mean. m The value range is 1-8. G 1 to G 4 Represents N, W, E, S four vertical and horizontal directions, G 5 to G 8 Represent WN, EN, WS, ES respectively. 4 oblique directions. In order to ensure the stability of iteration, 0≤λ≤0.25.

[0046] Step (4) In order to select the best result after multiple iterations, the quantum particle swarm algorithm is used to optimize each result after the receptive field filtering in step (3), and the peak signal-to-noise ratio of the image quality evaluation index is used as the optimal fitness function. Initialize the population and set the maximum number of iterations of the quantum particle swarm. T max 、Number of outer wheels R , population size M and dimensions D dim As well as the optimization range of each dimension, the particle position vector within the set range X i =( X i1 , X i2 ,..., X iDdim )( i =1,2,..., M ) to randomly assign initial values.

[0047] Initialize the best fitness function and calculate the fitness function value of each particle fitness ( X ), the non-global optimal fitness value is replaced by the best fitness value of each generation fitness ( X ) is replaced and recorded as pbest ( i )( i =1,2,..., M ), set the global optimal fitness function, denoted as gbest Through the local optimum of each particle pbest ( i ) is replaced by the maximum value. Figure 5 As shown in the figure, after QPSO optimization, the convergence value of fitness can be obtained to about 19.1db in about 12 times, and the best output image can be found by rapid iteration, and the signal-to-noise ratio tends to stabilize at the maximum value of 19.1db.

[0048] in, fitness ( X ) is the peak signal-to-noise ratio, which is an objective standard for evaluating images. The principle of this method is to calculate the mean square error between the original image and the processed image, relative to (2 n -1) 2 The logarithmic value of is shown in (13): (13) MSE is the mean square error between the original image and the processed image, as shown in formula (14): (14) In the formula f * ( i , j ) and f ( i , j ) represent the filtered image and the ideal image respectively. The better the denoising effect is, the smaller the MSE is.

[0049] Step (5) Output of the best results and parameters, according to the best global gbest The value is used as the basis for obtaining the optimal parameters and outputting the optimal image.

[0050] Figure 6 Comparison diagram of the present invention and other methods. It can be seen from Figure (c) that most of the noise can be filtered out after wavelet denoising, but there is still granular noise. Catte Filtering out noise will also cause the binary image to produce a fogging effect. This is because Catte The algorithm introduces Gaussian filtering which causes blurring.

[0051] Figure (e) is processed by traditional dimensionality reduction method and then enters FHN model processing. When processing the two-dimensional signal after dimensionality reduction, some structural correlation will be lost, thus losing details. We can see from Figure (e) that after processing, the pixel shift phenomenon in the structure will be caused. Figure (f) is the result of the algorithm flow in this paper. From a subjective point of view, it is more in line with the visual characteristics and almost maintains the characteristics of the original binary image.

[0052] The third aspect of the present invention provides an electronic device, comprising: a processor and a memory, computer instructions for an image enhancement method of a spatial visual neural network model and / or computer instructions for processing state data of an intelligent agent using an action strategy model. When the computer instructions are executed by the processor, the electronic device executes an image enhancement method according to the spatial visual neural network model and / or an image enhancement method of a spatial visual neural network model in the present invention.

[0053] The fourth aspect of the present invention provides a computer-readable storage medium, comprising program instructions of an action strategy model and / or program instructions for processing state data of an intelligent agent using an action strategy model. When the program instructions are executed by a processor, an image enhancement method based on a spatial visual neural network model and / or an image enhancement method based on a spatial visual neural network model in the present invention is implemented.

[0054] As a non-volatile computer-readable storage medium, it can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as the spatial vision system in the embodiment of the present invention. The program instructions / modules corresponding to the image enhancement method of the spatial visual neural network model. One or more program instructions are stored in a non-volatile computer-readable storage medium, and when executed by a processor, the image enhancement method of the spatial visual neural network model and the method of controlling an intelligent agent using an action strategy model in any of the above method implementations are executed.

[0055] The non-volatile computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system and application programs required for at least one function; the data storage area may store an image enhancement method based on a spatial visual neural network model, and data created by the use of a method processing device for controlling an intelligent agent using an action strategy model, etc.

[0056] In addition, the non-volatile computer-readable storage medium may include high-speed random access memory and may also include non-volatile memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device.

[0057] In some embodiments, the non-volatile computer-readable storage medium optionally includes a memory remotely disposed relative to the processor, and the remote memory can be connected to the image enhancement method of the spatial visual neural network model and the method processing device for recognizing language using the action strategy model through a network. Examples of the above network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0058] An embodiment of the present invention also provides a computer program product, which includes a computer program stored on a non-volatile computer-readable storage medium, and the computer program includes program instructions. When the program instructions are executed by a computer, the computer executes any one of the above-mentioned image enhancement methods of the spatial visual neural network model.

[0059] Figure 7 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, such as Figure 7 As shown, the device includes: one or more processors 710 and a memory 720, Figure 7 A processor 710 is taken as an example. The device of the image enhancement method of the spatial visual neural network model may also include: an input device 730 and an output device 740. The processor 710, the memory 720, the input device 730 and the output device 740 may be connected via a bus or other means.

[0060] Figure 7The example of the connection via bus is taken. The memory 720 is the above-mentioned non-volatile computer-readable storage medium. The processor 710 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 720, that is, the above-mentioned method is implemented. The input device 730 can receive input digital or character information, and generate key signal input related to user settings and function control of the information delivery device. The output device 740 may include display devices such as display screens.

[0061] The above-mentioned product can execute the method provided by the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided by the embodiment of the present invention.

[0062] As an implementation mode, the electronic device can be applied to an encryption and decryption platform, including: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the image enhancement method processor of the at least one spatial visual neural network model to: The method comprises: normalizing a noisy binary image; constructing a 3*3 spatial FHN neural network convolution kernel template based on the processed binary image; and obtaining a spatial neuron pulse potential signal corresponding to the binary image by passing a noisy signal through the FHN neural network convolution kernel: Traversing the spatial neuron pulse potential signal of the binary image, selecting the current center value point with 8 surrounding directions in turn; judging the difference between the center value of the current center value point and the 8 surrounding directions through the corresponding local variance function or substitution function, and constructing the visual receptive field diffusion coefficient with a direction selection mechanism; Substituting the visual receptive field diffusion coefficient into the diffusion equation; calculating the mean and local variance of the eight directions relative to the center value point; and obtaining the diffusion coefficient by selecting from the local variance function or the substitute function by judging the difference between the mean and the center value; The spatial neuron pulse potential signal of the binary image is subjected to multiple spatial convolution filtering by using the visual receptive field diffusion coefficient; in each filtering, multiple optimal algorithm iterations are performed by using a quantum particle swarm algorithm to obtain a processed image.

[0063] The electronic device according to the embodiments of the present invention exists in various forms, including but not limited to: (1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, feature phones, etc.; (2) Ultra-mobile personal computer devices: These devices belong to the category of personal computers, have computing and processing functions, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices; (3) Portable entertainment devices: These devices can display and play multimedia content. Such devices include: audio and video players, handheld game consoles, e-books, as well as smart toys and portable car navigation devices; (4) Server: A device that provides computing services. The server consists of a processor, hard disk, memory, system bus, etc. The server is similar to a general computer architecture, but because it needs to provide highly reliable services, it has higher requirements in terms of processing power, stability, reliability, security, scalability, manageability, etc. (5) Other electronic devices with data interaction functions.

[0064] The device implementation described above is merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, i.e., they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the present implementation scheme. Those of ordinary skill in the art may understand and implement it without creative effort.

[0065] Through the description of the above implementation modes, those skilled in the art can clearly understand that each implementation mode can be implemented by means of software plus a necessary general hardware platform, or of course by hardware. Based on such an understanding, the above technical solution can essentially or in other words be embodied in the form of a software product that contributes to the prior art. The computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and includes a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each implementation mode or some parts of the implementation mode.

[0066] Finally, it should be noted that the above implementation modes are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned implementation modes, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned implementation modes, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various implementation modes of the present invention.

Claims

1. An image enhancement method for a spatial visual neural network model, characterized in that: It includes: Normalize the noisy binary image; Based on the processed binary image, a 3*3 spatial FHN neural network convolution kernel template is constructed; Obtaining a spatial neuron pulse potential signal corresponding to the binary image by passing the noisy signal through the FHN neural network convolution kernel; Traversing the spatial neuron pulse potential signal of the binary image, selecting the current center value point with 8 surrounding directions in turn; judging the difference between the center value of the current center value point and the 8 surrounding directions through the corresponding local variance function or substitution function, and constructing the visual receptive field diffusion coefficient with a direction selection mechanism; Substituting the visual receptive field diffusion coefficient into the diffusion equation; Calculate the mean and local variance of the eight directions relative to the central value point; and obtain the diffusion coefficient by determining the difference between the mean and the central value and selecting from the local variance function or the substitute function; The spatial neuron pulse potential signal of the binary image is subjected to multiple spatial convolution filtering by using the visual receptive field diffusion coefficient; in each filtering, multiple optimal algorithm iterations are performed by using a quantum particle swarm algorithm to obtain a processed image.

2. The image enhancement method of the spatial visual neural network model according to claim 1, characterized in that: The obtaining of the spatial neuron pulse potential signal corresponding to the binary image comprises: The noisy signal is pulse-coded by a neural network convolution kernel to obtain a spatial neuron pulse potential signal corresponding to the binary image.

3. The image enhancement method of the spatial visual neural network model according to claim 1, characterized in that: According to the difference between the center value of the current center value point and the surrounding eight directions, the corresponding local variance function or substitution function is used to construct the visual receptive field diffusion coefficient with a direction selection mechanism, including: Obtain the difference between the center value of the current center value point and the surrounding 8 directions, and determine whether the difference value is less than the set threshold. If so, construct a visual receptive field diffusion coefficient with a direction selection mechanism through a local variance function; if not, construct a visual receptive field diffusion coefficient with a direction selection mechanism through a substitution function.

4. The image enhancement method of the spatial visual neural network model according to claim 1, characterized in that: In each filtering, multiple optimal algorithm iterations are performed by using a quantum particle swarm algorithm; Initialize the quantum particle swarm population, initialize the best fitness function, use the peak signal-to-noise ratio as the fitness function, calculate the peak signal-to-noise ratio of the two-dimensional image obtained by the anisotropic diffusion equation in each filtering, update the best fitness function, determine whether the optimal peak signal-to-noise ratio is obtained, and then decide the number of iterations.

5. The image enhancement method of the spatial visual neural network model according to claim 1, characterized in that: The mathematical expression of the FHN neuron is shown in Formula 1: Formula 1 In the formula, u is the fast-changing membrane voltage of the neuron; v is the neuron’s slow-changing recovery voltage; ε (0<ε=1) is the time constant that determines the firing rate of the neuron; a is the threshold value for transition from a single potential well to a double potential well, a and b are all constants, and together they determine the shape of the double potential well.

6. The image enhancement method of the spatial visual neural network model according to claim 1, characterized in that: The receptive field model of the diffusion equation is shown in the following formula 2: Formula 2 In the formula u 0( x , y ) is the initial image, div is the divergence operator, t is the scale factor, |▽ u | is the modulus of the image gradient, c 1 represents the local variance diffusion coefficient function, c 2 represents the alternative diffusion coefficient function, u 1 and u 2 represents the choice of diffusion coefficient function coefficient, which takes the value of 0 or 1, and the sum is 1. The threshold of the difference between the center point value and the mean of the eight directions is 0.

2. If it is less than 0.2, the local variance is selected, otherwise the alternative function is selected as the coefficient function.

7. An image enhancement system based on a spatial visual neural network model, characterized in that: It includes: a pulse potential signal generating unit, which is configured to perform normalization processing on a noisy binary image; Based on the processed binary image, a 3*3 spatial FHN neural network convolution kernel template is constructed; the spatial neuron pulse potential signal corresponding to the binary image is obtained by passing the noisy signal through the FHN neural network convolution kernel: A diffusion coefficient unit is configured to traverse the spatial neuron pulse potential signal of the binary image, and sequentially select a current center value point with 8 surrounding directions; according to the difference value between the center value of the current center value point and the 8 surrounding directions, determine the corresponding local variance function or substitution function to construct a visual receptive field diffusion coefficient with a direction selection mechanism; A selection unit is configured to substitute the visual receptive field diffusion coefficient into a diffusion equation; calculate the mean and local variance of eight directions relative to the center value point; and select the diffusion coefficient from the local variance function or the substitute function by determining the difference between the mean and the center value; The graphics processing unit is configured to perform multiple spatial convolution filtering on the spatial neuron pulse potential signal of the binary image through the visual receptive field diffusion coefficient; in each filtering, multiple optimal algorithm iterations are performed through the quantum particle swarm algorithm to obtain a processed image.

8. The image enhancement system of the spatial visual neural network model according to claim 1, characterized in that: The pulse potential signal generating unit is further configured as: The noisy signal is pulse-coded by a neural network convolution kernel to obtain a spatial neuron pulse potential signal corresponding to the binary image.

9. An electronic device, characterized in that: include: processor; as well as A memory storing computer instructions of an image enhancement method of a spatial visual neural network model, wherein when the computer instructions are executed by the processor, the electronic device executes the method according to any one of claims 1 to 6.

10. A computer-readable storage medium, characterized in that: A program instruction for an image enhancement method comprising a spatial visual neural network model, which, when executed by a processor, implements the method according to any one of claims 1 to 6.

Images