A method for image denoising based on memristor

Abstract

The present invention discloses a method for image denoising based on a memristor, which is specifically as follows: step 1: designing a denoising branch based on a memristor; step 2: if the image is a one-dimensional image, constructing a denoising circuit for the one-dimensional image based on the denoising branch, and if the image is a two-dimensional image, constructing a denoising circuit for the two-dimensional image based on the denoising branch; step 3: mapping the grayscale value of the image to an input current, and importing it into a corresponding denoising circuit; step 4: restoring the current mapping output by the denoising circuit to the grayscale value of the image; step 5: binarizing the grayscale value in step 4 and restoring it to the image, thereby obtaining a denoised image. The denoising method adopted by the present invention provides a new idea for the application of memristors in image denoising.

Description

Technical Field

[0001] The invention belongs to the technical field of image processing. Background Art

[0002] Due to the unique switching mechanism, natural memory function, low energy consumption, continuous input and output characteristics, nanometer size and compatibility with CMOS technology, memristors have broad application prospects in the field of image processing. At present, memristors are mainly used for storage and brain-like computing. The only applications in images are mostly limited to handwritten digit recognition, lacking other applications. Summary of the invention

[0003] Purpose of the invention: In order to solve the problems existing in the above-mentioned prior art, the present invention provides an image noise reduction method based on memristor.

[0004] Technical solution: The present invention provides an image noise reduction method based on a memristor, which specifically includes the following steps:

[0005] Step 1: Design a noise reduction branch based on memristor;

[0006] Step 2: If the image is a one-dimensional image, a noise reduction circuit for the one-dimensional image is constructed based on the noise reduction branch; if the image is a two-dimensional image, a noise reduction circuit for the two-dimensional image is constructed based on the noise reduction branch;

[0007] Step 3: Map the grayscale value of the image to the input current and import it into the corresponding noise reduction circuit;

[0008] Step 4: Restore the current mapping output by the noise reduction circuit to the image grayscale value;

[0009] Step 5: Binarize the grayscale value in step 4 and restore it to an image to obtain a denoised image.

[0010] Furthermore, the noise reduction branch in step 1 includes a memristor and a current source, the current source is connected to the negative electrode of the memristor, and the positive electrode of the memristor is grounded.

[0011] Furthermore, the noise reduction circuit for the two-dimensional image in step 2 includes H*W noise reduction branches, H represents the height of the image, that is, the number of rows of the noise reduction array, W represents the width of the image, that is, the number of columns of the noise reduction array, the negative electrodes of the memristors of two adjacent noise reduction branches in each row are connected through a resistor, and the negative electrodes of the memristors of two adjacent noise reduction branches in each column are connected through a resistor, the current value of the image pixel point is input into the corresponding negative electrode of the memristor, and the current output from the positive electrode of the memristor is collected.

[0012] Furthermore, the noise reduction circuit of the one-dimensional image in the step 2 includes n noise reduction branches arranged vertically, where n is the number of pixels of the one-dimensional image, and the negative electrodes of the memristors of two adjacent noise reduction branches are connected through a resistor, the current value of the image pixel is input into the corresponding negative electrode of the memristor, and the current output from the positive electrode of the memristor is collected.

[0013] Furthermore, in step 3, the grayscale value of the image is mapped to an input current with a value between [1uA, 3uA].

[0014] Beneficial effects: The present invention will not remove characteristic images when removing noise. The present invention has the characteristics of multidisciplinary integration, involving software, simulation, circuit design and optimization. The noise reduction method adopted provides a new idea for the application of memristor in image noise reduction. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 The noise reduction branches in the present invention, wherein Figure (a) is a circuit diagram of two branches, and Figure (b) is a circuit diagram of five branches;

[0016] Figure 2 The graphs of the currents in the two branches changing with time are shown below;

[0017] Figure 3 is the graph of the current contrast of the two branches changing with time;

[0018] Figure 4 Simulate a single current versus time graph for the five-branch circuit;

[0019] Figure 5 Simulate a single current contrast versus time plot for the five branch circuits;

[0020] Figure 6 Simulate the time-varying graph of the connected current for the five branch circuits;

[0021] Figure 7 Simulate the time-varying graph of the connected current contrast for five branch circuits;

[0022] Figure 8 This is a circuit diagram of a noise reduction array of the present invention;

[0023] Fig. 9 This is an image denoising test effect diagram of the present invention, wherein (a) is an input image, and (b) is an image denoised by a two-dimensional image denoising circuit. DETAILED DESCRIPTION

[0024] The accompanying drawings, which constitute a part of the present invention, are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations on the present invention.

[0025] The present invention provides a method for hardware-implementing a noise reduction algorithm in a memristor circuit, comprising the following steps:

[0026] Step 1: Design a noise reduction branch based on memristor;

[0027] Step 2: If the image is a one-dimensional image, a noise reduction circuit for the one-dimensional image is constructed based on the noise reduction branch; if the image is a two-dimensional image, a noise reduction circuit for the two-dimensional image is constructed based on the noise reduction branch;

[0028] Step 3: Map the grayscale value of the image to the input current and import it into the corresponding noise reduction circuit;

[0029] Step 4: Restore the current mapping output by the noise reduction circuit to the image grayscale value;

[0030] Step 5: Binarize the grayscale value in step 4 and restore it to an image to obtain a denoised image.

[0031] The noise reduction circuit in step 1 includes a current source I1 (hereinafter referred to as In) and a memristor M1. I1 is connected to the non-doped TiO2 end (ie, the negative electrode) of the memristor M1, and the doped TiO2-x end (hereinafter referred to as the positive electrode) of M1 is grounded to form a branch.

[0032] The denoising circuit of the one-dimensional image in step 2 is to expand one branch into n branches according to the size n of the one-dimensional image, such as Figure 1 As shown in (a) in FIG. 1 , the circuit is expanded into two branches, which are connected by a resistor R (hereinafter referred to as the distance resistor); and expanded into five branches, as shown in FIG. Figure 1 As shown in (b), each branch is connected through a resistor R, and the two ends of R1 (the resistor of the first branch and the second branch) are connected to the node between I1 and M1 and the node between I2 and M2. Similarly, a noise reduction circuit is constructed in a one-dimensional plane. The resistance value of the resistor and the initial resistance state of the memristor are adjusted, and the optimal solution of the image noise reduction circuit is obtained through the change of contrast.

[0033] The negative electrode of the memristor is connected to a current source, and the oxygen vacancies are shifted and rearranged under the action of the current. The doped area of ​​the memristor is grounded, and the current flows from the non-doped side of the negative electrode to the doped side. The oxygen vacancies are attracted to the positive electrode side, the width of the TiO2-x area decreases, the width of the TiO2 area increases, the conductance of the memristor decreases, the resistance value increases, and the current flowing through the memristor decreases.

[0034] Figure 2It is a graph showing the change of two branch currents over time in the present invention, wherein the M1 memristor is connected to a 3uA current source on the left, the M2 memristor is connected to a 1uA current source on the left, and a resistor R is connected in the middle. It can be obtained by the superposition theorem that when the 3uA current source acts alone, the current flowing through the memristor decreases, the branch current remains unchanged, and the potential on the left side of M1 is higher than the potential of M2, so the resistance of M1 increases faster, so the current flowing through M1 decreases, and the current flowing through M2 increases. Similarly, when the 1uA current source acts alone, the current flowing through M2 decreases, and the current flowing through M1 increases. Since the current source connected to the left of M1 is 3uA, the current changes faster and the current increased by M2 is larger, so the current at the left end of M1 decreases, and the current at the left end of M2 increases. According to the above characteristics, the output current of the image with low input current (1uA) can be increased (the grayscale value becomes brighter) and tends to the output current of the image with high input current (3A). As long as the low current representing noise can be distinguished from the current representing feature information while increasing the low current representing noise, image noise reduction can be achieved.

[0035] Figure 3 This is a graph showing the change of the current contrast of the two branches in the present invention over time. Contrast refers to the measurement of different brightness levels between the brightest white and the darkest black in the light and dark areas of an image. The larger the difference range, the greater the contrast, and the smaller the difference range, the smaller the contrast. Define the input current of M1 (3uA, grayscale value is 255) / M2 input current (1uA, grayscale value is 0) as the input contrast of the two branches. As time goes by, the contrast of the output current of M1 / the output current of M2 drops to 1.3, indicating that the grayscale values ​​of adjacent pixels are approaching each other after image processing.

[0036] pass Figure 2 and Figure 3 The degree of convergence between the grayscale values ​​represented by the two branch inputs can be compared. The greater the contrast reduction, the closer the two grayscale values ​​are, and the better the noise reduction effect. It can be concluded that the current on the memristor connected to the branch with a larger input current will decrease, and the current on the memristor connected to the branch with a smaller input current will increase.

[0037] Figure 4A single current variation over time diagram is simulated for the five branch circuits, where M1, M3 and M5 are connected to a 3uA (grayscale value is 255) current source on the left, and M2 and M4 are connected to a 1uA (grayscale value is 0) current source on the left, which is used to simulate the noise reduction effect of the memristor on isolated noise. The bottom dotted line is the current flowing through M3, and the bottom solid curve is the current flowing through M4. The two current curves eventually overlap. From the analysis of the figure, it can be seen that the current of M1, M3 and M5 decreases (the grayscale value becomes darker), and the current of M2 and M4 increases (the grayscale value becomes brighter). The branch currents of M2 and M4 slowly approach the branch currents of M1, M3 and M5, and finally blend together, that is, at this time, the 3uA current representing the image background and the 1uA current representing the noise are in the same current range.

[0038] Figure 5 A single current contrast variation diagram over time is simulated for five branch circuits. From the analysis of the diagram, it can be seen that the contrast of M3 / M4 (Background / Noise) decreases and approaches 1, that is, the noise and the image background have been merged together at this time, proving that the noise reduction circuit of the present invention can achieve image noise reduction.

[0039] Figure 6 The simulation of the connected current changes over time for the five branch circuits. M1 and M5 are connected to a 3uA (grayscale value is 255) current source on the left, and M2, M3 and M4 are connected to a 1uA (grayscale value is 0) current source on the left, which is used to simulate the noise reduction effect of the memristor on the characteristic image. The solid line is the current flowing through M1 and M5 and overlaps together, and the top dotted line is the current flowing through M2. From the analysis of the figure, it can be seen that the current flowing through M1 and M5 decreases (the grayscale value becomes darker), and the current flowing through M2, M3 and M4 increases (the grayscale value becomes brighter). The branch currents of M2, M3 and M4 are slowly approaching the branch currents of M1 and M5, and finally do not blend together, which can be easily distinguished. That is, at this time, the 3uA current representing the image background and the 1uA current representing the characteristic image are in two current intervals.

[0040] Figure 7 The figure shows the change of the contrast of the connected current of the five branch circuits over time. From the analysis of the figure, it can be seen that the contrast of M1 / M2 (Background / Feature) drops to about 1.4, that is, at this time the feature image and the background do not overlap and can be easily distinguished, proving that the noise reduction circuit of the present invention can remove the noise while still distinguishing the feature image from the image background.

[0041] according to Figure 2~Figure 7 It can be stated that the noise reduction circuit of the present invention will not remove the characteristic image when removing the noise.

[0042] like Figure 8As shown, taking a 15×15 two-dimensional image denoising circuit as an example, the input current sources are all connected to the negative electrodes of memristors M1_1 to M1_15. The positive electrodes of M1_1 to M1_15 are directly grounded. The negative electrodes of the memristors in each branch are connected through resistors R, thus forming the first row of the two-dimensional image denoising circuit. The second row of the circuit is built according to the above rules. At the same time, a resistor R is connected to the negative electrodes of M1_1 and M2_1, and this is extended to M1_15 and M2_15. By analogy, a 15×15 memristor array can be built. The image gray value in the interval [1uA, 3uA] is mapped to the corresponding current value, which is input by the current source to the corresponding branch memristor to represent the gray value of the corresponding point of the image.

[0043] As Fig. 9 shown, the image feature image is "世". The image noise is divided into two types. The first type is the black dots around "世", and the second type is the white dots inside "世". In the 15×15 memristor array, the image is divided into three parts: the strokes of the feature image "世", the image background, and the noise. 1uA represents the gray value 0, and 3uA represents the gray value 255. The 3uA current represents the image background, and the 1uA current represents the strokes and noise of the feature image "世". The 3uA current is pulled down by the 1uA current and gradually tends to the 1uA current curve, and the 1uA current is also correspondingly pulled up. The 1uA current of "世" is a connected region, so the rising current will not blend with the 3uA current and can ultimately be distinguished; the isolated 1uA current will be pulled up and blend with the 3uA current. After all the current values are exported and mapped back to the image gray value, and after binarization and restoration to the image, the method of hardware implementing the denoising algorithm in the memristor circuit in the present invention is realized. It should be particularly noted that when a 3uA (gray value 255) current is added inside the 1uA "世", this current value will also be regarded as noise, affected by the surrounding 1uA current values, and ultimately blend with the 1uA current, and image denoising can also be achieved.

[0044] In addition, it should be noted that those of ordinary skill in the art should recognize that the above embodiments are only used to verify the present invention, rather than as a limitation to the present invention. As long as it is within the scope of the present invention, changes and deformations of the above embodiments will fall within the protection scope of the present invention. In addition, it should be noted that in the above specific embodiments, the various specific technical features described can be combined in any suitable manner without contradiction. To avoid unnecessary repetition, the present invention will not separately describe various possible combination methods.

Claims

1. A memristor-based image denoising method, characterized in that: The specific steps include: Step 1: Design a noise reduction branch based on memristor; Step 2: If the image is a one-dimensional image, a noise reduction circuit for the one-dimensional image is constructed based on the noise reduction branch; if the image is a two-dimensional image, a noise reduction circuit for the two-dimensional image is constructed based on the noise reduction branch; Step 3: Map the grayscale value of the image to the input current and import it into the corresponding noise reduction circuit; Step 4: Restore the current mapping output by the noise reduction circuit to the image grayscale value; Step 5: Binarize the grayscale value in step 4 and restore it to an image to obtain a denoised image; The noise reduction branch in step 1 includes a memristor and a current source, the current source is connected to the negative electrode of the memristor, and the positive electrode of the memristor is grounded; The noise reduction circuit for the two-dimensional image in step 2 includes H*W noise reduction branches, where H represents the height of the image, that is, the number of rows of the noise reduction array, and W represents the width of the image, that is, the number of columns of the noise reduction array. The negative electrodes of the memristors of two adjacent noise reduction branches in each row are connected through a resistor, and the negative electrodes of the memristors of two adjacent noise reduction branches in each column are connected through a resistor. The current value of the image pixel point is input into the corresponding negative electrode of the memristor, and the current output from the positive electrode of the memristor is collected.

2. The image denoising method based on memristor according to claim 1, characterized in that: The noise reduction circuit for the one-dimensional image in step 2 includes n noise reduction branches arranged vertically, where n is the number of pixels of the one-dimensional image. The negative electrodes of the memristors of two adjacent noise reduction branches are connected through a resistor, the current value of the image pixel is input into the corresponding negative electrode of the memristor, and the current output from the positive electrode of the memristor is collected.

3. The image denoising method based on memristor according to claim 1, characterized in that: In step 3, the grayscale value of the image is mapped to an input current between [1uA, 3uA].

Images