Single-pixel imaging one-dimensional signal enhancement method based on coding image similarity

Abstract

The invention discloses a single-pixel imaging one-dimensional signal enhancement method based on coding image similarity, which belongs to the technical field of computational optical imaging, and comprises the following steps of: reconstructing an image by using a one-dimensional signal by using a traditional ghost imaging method; obtaining a coded image, and calculating a similarity index; for any coded image, calculating a weight coefficient according to similarity, and performing weighted average denoising on the one-dimensional signal to obtain an updated one-dimensional signal; using a momentum rule to balance the relative weight of the one-dimensional signal before and after updating; and performing one-dimensional signal updating on all the coded images, and reconstructing the images again by using the updated one-dimensional signals. According to the method, the problem that the signal-to-noise ratio of a one-dimensional signal is reduced due to factors such as insufficient sampling rate and weak illumination intensity during high-speed single-pixel imaging can be effectively solved, and the method is adaptive to coding masks adopting orthogonal projection modes such as a Hadamard base and a Fourier base, and has an important reference value for future high-speed single-pixel imaging.

Description

Technical Field

[0001] The present invention belongs to the technical field of computational optical imaging, and particularly relates to a method for enhancing one-dimensional signals in single-pixel imaging based on the similarity of encoded images. Background Technique

[0002] Single-pixel imaging is an effective computational imaging method that has rapidly developed in recent years. It combines a spatial light modulator and a cost-effective single-point detector, solving the challenge of manufacturing large-scale pixel arrays in invisible bands, thereby significantly expanding the imaging wavelength range at a relatively low cost. One of the ultimate goals in the research direction of single-pixel imaging is to improve the imaging rate of single-pixel imaging. The most important link here is to improve the pattern projection rate, which has higher requirements for sampling devices, making it impossible to perform more sampling times during a single projection, thus limiting the above-mentioned averaging method.

[0003] When the number of samplings is insufficient, the illumination is weak, or the detector noise is slightly large, the quality of the reconstructed image of single-pixel imaging will gradually deteriorate. This deterioration is a deterioration of the signal quality at the root cause, which is caused by the rapid decrease in the signal-to-noise ratio of the one-dimensional voltage signal, and it is often very difficult to correct this degradation in two-dimensional image reconstruction. At present, most research mainly focuses on the improvement of two-dimensional image reconstruction algorithms, the improvement of hardware facilities, and the expansion of application scenarios. Most people usually use a sampling rate much higher than the projection rate to obtain more data points for averaging when dealing with one-dimensional signals, and almost no one mentions further improvement in the processing method of one-dimensional signals in single-pixel imaging. Summary of the Invention

[0004] To solve the above technical problems, the present invention provides a method for enhancing one-dimensional signals in single-pixel imaging based on the similarity of encoded images. By denoising the one-dimensional signal through the similarity between encoded images, it can effectively address the problem of the decrease in the signal-to-noise ratio of one-dimensional signals caused by factors such as insufficient sampling rate and weak illumination intensity during high-speed single-pixel imaging, and is applicable to encoded masks using orthogonal projection modes such as Hadamard basis and Fourier basis, which has important reference value for future high-speed single-pixel imaging.

[0005] To achieve the above purpose, the present invention adopts the following technical solutions:

[0006] A method for enhancing one-dimensional signals in single-pixel imaging based on the similarity of encoded images, the steps include:

[0007] Step 1, reconstruct an image using a one-dimensional signal;

[0008] Step 2, apply a Hadamard encoded mask to the reconstructed image to obtain an encoded image matrix, and calculate the encoded images in the encoded image matrix and another encoded image the mean square error matrix as a similarity index;

[0009] Step 3. For the encoded image in the mean square error matrix find the encoded images similar to the encoded image and use the one-dimensional signals corresponding to the encoded images to perform weighted average denoising on the one-dimensional signal in Step 1 to obtain a preliminarily updated one-dimensional signal;

[0010] Step 4. According to the relative credibility between the one-dimensional signal and the preliminarily updated one-dimensional signal, adopt the momentum rule to balance the weights to obtain a finally available one-dimensional signal;

[0011] Step 5. Sequentially execute Step 3 - Step 4 for all the encoded images and their corresponding one-dimensional signals in the encoded image matrix to obtain all the finally available one-dimensional signals, and obtain a reconstructed image according to the ghost imaging method;

[0012] Step 6. Extract the edge and detail information of the reconstructed image and integrate it into the reconstructed image.

[0013] In a second aspect, the present invention provides an electronic device, including: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the foregoing single-pixel imaging one-dimensional signal enhancement method based on the similarity of encoded images.

[0014] In a third aspect, the present invention provides a computer-readable storage medium, on which executable instructions are stored, and when the instructions are executed by a processor, the processor can implement the foregoing single-pixel imaging one-dimensional signal enhancement method based on the similarity of encoded images.

[0015] The beneficial effects of the present invention are as follows:

[0016] The present invention uses the similarity between encoded images to denoise the one-dimensional signal of single-pixel imaging, without the need to perform a large number of samplings on each projection frame at an extremely high hardware acquisition rate like traditional methods, reducing the pressure on signal acquisition devices for future high-speed single-pixel imaging;

[0017] The processing method of the one-dimensional signal of single-pixel of the present invention has good noise robustness and can still play a good role in denoising the one-dimensional signal in a non-ideal imaging environment with a low signal-to-noise ratio;

[0018] The method for processing one-dimensional single-pixel signals in the present invention is applicable to both the Hadamard basis and the Fourier basis, which are the two most widely used orthogonal bases in single-pixel imaging. It can achieve good denoising effects in both Hadamard Single-pixel Imaging (i.e., HSI) and Fourier Single-pixel Imaging (i.e., FSI), and also has wide applicability to other orthogonal projection basis functions.

[0019] The present invention can significantly remove most of the background noise in the image obtained by single-pixel imaging and protect the overall contour of the image, effectively retaining the edge and detail information of the object. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] Figure 1 is a flowchart of the one-dimensional signal enhancement method for single-pixel imaging based on the similarity of coded images in the present invention;

[0021] Figure 2 is a basic schematic diagram of single-pixel imaging;

[0022] Figure 3 is a schematic diagram of the process for obtaining the coded image;

[0023] Figure 4 is a comparison chart of the reconstructed images and image quality before and after optimization using the present invention at different hardware sampling rates;

[0024] Figure 5 is a comparison chart of the reconstructed images and image quality before and after optimization using the present invention at different signal-to-noise ratios of one-dimensional signals;

[0025] Figure 6 is a comparison chart of the reconstructed images and image quality before and after optimization using the present invention when using the Fourier basis. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0026] The present invention will be further described below in conjunction with the drawings and embodiments.

[0027] As Figure 1 shown, the one-dimensional signal enhancement method for single-pixel imaging based on the similarity of coded images in the present invention specifically includes the following steps:

[0028] Step 1: Reconstruct the initial image using the one-dimensional signal according to the traditional ghost imaging method to obtain the reconstructed image , and the corresponding device schematic diagram is as shown. Among them, a series of sizes of Figure 2 are The Hadamard pattern is loaded onto the digital micromirror array to form structured light modulation of the laser spot. After passing through the focusing lens, it irradiates the object. Then, the total reflected light intensity is collected by the collection lens and the single-pixel detector. The output voltage of the detector corresponding to the i-th Hadamard pattern is a one-dimensional signal , , which is expressed as:

[0029] ,

[0030] where is the scene to be reconstructed, is the spatial distribution of the i-th coding mask, and x and y are the spatial indices of the reconstructed image I and the coding mask . The reconstructed image can be expressed as:

[0031] ,

[0032] where N is the side length of the image, is the i-th one-dimensional signal collected, is the spatial distribution of the i-th coding mask.

[0033] Step 2: First, a series of Hadamard coding masks act on the reconstructed image to obtain the coded image . This process can be described by the following formula, where is the symbol for element-wise multiplication, is the matrix of the reconstructed image, is the matrix of the coding mask. The two numbers in the superscript represent the row number and column of the elements in the matrix, and the corresponding process is as Figure 3 shown.

[0034] ,

[0035] Then, calculate the mean square error matrix between the coded image and another coded image as the similarity index. The process of calculating can be expressed as:

[0036] ,

[0037] Step 3: For any coded image , find the index of the coded images with the minimum mean square error between them and the coded image in the mean square error matrix, and use the corresponding one-dimensional signal Perform weighted average denoising on to obtain a preliminarily updated one-dimensional signal . Specifically, it may include the following two sub-steps:

[0038] Step 3.1: Arrange the th row (i.e., ) of the mean square error matrix in ascending order. The first indices corresponding to the encoded images , which are the encoded images most similar to the encoded image ;

[0039] Step 3.2: Perform weighted average denoising on the one-dimensional voltage signals corresponding to these indices on the one-dimensional signal to obtain a preliminarily updated one-dimensional signal . The specific weighted average conforms to the following formula: ,

[0040] ,

[0041] where is the weight corresponding to the th one-dimensional signal , which is determined by the between the encoded image and the encoded image . The smaller is, the more similar and are. Then, it can be considered that the corresponding one-dimensional voltage signals and are more similar, and the corresponding weight coefficient is larger. Specifically, it conforms to the following formula:

[0042] ,

[0043] where L is the number of the most similar terms, h is the weight stretching factor, and the denominator plays a role in normalization.

[0044] Step 4: According to the relative credibility between the initial one-dimensional signal and the preliminarily updated one-dimensional signal , use the "momentum rule" to balance the weights of the two to obtain a finally available one-dimensional signal . Specifically, it includes the following sub-steps:

[0045] a. According to the initial one-dimensional signal and the preliminarily updated one-dimensional signal Allocate weights based on the relative credibility between them. When using the "momentum rule", in principle, if the signal-to-noise ratio of the initial one-dimensional signal is higher, it should be allocated a higher weight. Conversely, the initially updated one-dimensional signal should be allocated a higher weight. Here, the weight factor is used to adjust the relative weights of the two, and the second updated one-dimensional signal is obtained as follows:

[0046] ,

[0047] wherein, the weight factor is positively correlated with the signal-to-noise ratio of the one-dimensional signal .

[0048] b. Allocate weights according to the relative credibility within the initially updated one-dimensional signal . Most of the one-dimensional signals ~ are concentrated near the mean value, and a few deviate far from the mean value. These small amounts of tend to approach the mean value more easily when updated to , resulting in a larger error and a decrease in their credibility. Therefore, the weight factor is set as the variable according to the deviation degree of each from the mean value, and

[0049] is rewritten as the following expression:

[0050] wherein, the weight factor variable is positively correlated with the signal-to-noise ratio of the one-dimensional signal and negatively correlated with the deviation degree of the one-dimensional signal from the mean value of all one-dimensional signals.

[0051] Step 5: Sequentially execute Step 3 - Step 4 for all encoded images and their corresponding one-dimensional signals in the encoded image matrix to obtain all finally available one-dimensional signals, and obtain the reconstructed image according to the ghost imaging method .

[0052] Step 6: Edge protection. Use the Sobel gradient operator to extract the edge and detail information of the initial reconstructed image , and integrate it into the reconstructed image .

[0053] Embodiment

[0054] Define the ratio of the acquisition card sampling rate to the pattern projection rate as the sampling magnification , the three advantages of the present invention are verified from three perspectives respectively, and the corresponding experimental results are shown in Figure 4 , Figure 5 and Figure 6 respectively. The peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) of the two-dimensional reconstructed images are used as evaluation indicators, and the corresponding values are marked below the pictures.

[0055] Figure 4 shows the reconstruction results before and after optimization using the method of the present invention under the conditions of a one-dimensional signal-to-noise ratio of 0 dB and using the Hadamard basis mode at different sampling ratios . In high-speed single-pixel imaging, a high mode projection rate is required. Insufficient sampling rate of the acquisition card will cause the sampling ratio to drop rapidly. From the results of Figure 4 , when the sampling ratio is relatively high, more points can be collected in each projection frame, so that the noise can be suppressed through the averaging effect, making the denoising effect of the present invention less obvious. However, when the sampling ratio is relatively low, the present invention can effectively suppress the noise in the one-dimensional signal of single-pixel imaging, thereby improving the image quality of the reconstructed two-dimensional image, effectively reducing the background noise in the image, and effectively maintaining the edges and contours of the strokes in the image.

[0056] Figure 5 shows the reconstruction results before and after optimization using the method of the present invention under the conditions of a sampling ratio = 1 and using the Hadamard basis mode at different one-dimensional signal-to-noise ratios. In Figure 5 , when the signal-to-noise ratio is very low, the content of the two-dimensional image of the unprocessed one-dimensional signal is almost completely submerged in the noise, while after being processed by the present invention, the contour of the character can be effectively extracted from the noise.

[0057] Figure 6 shows the reconstruction results before and after optimization using the method of the present invention under the conditions of a sampling ratio = 1 and a one-dimensional signal-to-noise ratio of 0 dB in the Hadamard orthogonal basis mode and the Fourier orthogonal basis mode. The Hadamard basis and the Fourier basis are the most commonly used orthogonal projection modes in single-pixel imaging. In this example, the comparison is also made with the Fourier basis. Figure 6 The results in show that the present invention can take effect on both of these two orthogonal projection basis modes, effectively improving the two-dimensional image quality. The same principle applies to other types of orthogonal bases and will not be elaborated here.

[0058] In a second aspect, the present invention provides an electronic device, comprising: one or more processors; a memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors are caused to implement the foregoing one-dimensional signal enhancement method for single-pixel imaging based on coded image similarity.

[0059] In a third aspect, the present invention provides a computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, enable the processor to implement the foregoing one-dimensional signal enhancement method for single-pixel imaging based on coded image similarity.

[0060] The specific embodiments described above further elaborate on the object, technical solution, and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not used to limit the present invention. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present invention shall be included within the protection scope of the present invention.

Claims

1. One - dimensional signal enhancement method for single - pixel imaging based on the similarity of encoded images, characterized in that the steps Comprising: Step 1, reconstructing an image using a one-dimensional signal; Step 2: Apply the Hadamard coding mask to the reconstructed image to obtain a coded image matrix, and calculate the mean square error matrix between the coded image and another coded image as the similarity index; Step 3: Encode the image , in the mean square error matrix Find and encode the image Similar coded images, using the The one-dimensional signal corresponding to the coded image is subjected to weighted average denoising on the one-dimensional signal in step 1 to obtain a preliminarily updated one-dimensional signal; Step 4, performing weight balancing using a momentum rule according to the relative credibility between the one-dimensional signal and the preliminarily updated one-dimensional signal to obtain a finally available one-dimensional signal; Step 5, successively performing Steps 3 - 4 on all encoded images in the encoded image matrix and their corresponding one-dimensional signals to obtain all finally available one-dimensional signals, and obtaining a reconstructed image according to the ghost imaging method; Step 6, extracting edge and detail information of the reconstructed image and integrating it into the reconstructed image.

2. The single-pixel imaging one-dimensional signal enhancement method based on the similarity of encoded images according to claim 1, characterized in that, The said Step 1 includes: Load a series of Hadamard patterns of size onto the digital micromirror array to form structured light modulation of the light source and then irradiate it on the object. Then collect the total reflected light intensity with a detector. The output voltage of the detector corresponding to the i-th Hadamard pattern is a one-dimensional signal , , expressed as: ， In the formula, is the spatial distribution of the reconstructed image , is the spatial distribution of the i-th Hadamard coding mask; Reconstructed image Expressed as: 。 3. The single-pixel imaging one-dimensional signal enhancement method based on the similarity of encoded images according to claim 1, characterized in that In Step 2, applying a Hadamard encoding mask to the reconstructed image to obtain an encoded image matrix, including: ， Among them, is the symbol for element-wise multiplication, the reconstructed image and the i-th Hadamard pattern The two numbers in the superscript respectively represent the row number and column number of the elements in the reconstructed image matrix and the i-th Hadamard pattern matrix.

4. The method for enhancing a one-dimensional signal in single-pixel imaging based on the similarity of encoded images according to claim 1, wherein In step 2, calculating the mean square error matrix between the encoded image and another encoded image as a similarity metric, includes: ， where x and y are spatial indices of the encoded image.

5. The single-pixel imaging one-dimensional signal enhancement method based on the similarity of encoded images according to claim 1, characterized in that, The said Step 3 includes: Step 3.

1. Arrange the rows of the mean squared error matrix in ascending order, and the first rows, the indices corresponding to the first encoded images are the encoded images that are the most similar to the encoded image; encoded images. Step 3.

2. Based on the one-dimensional voltage signals corresponding to the previous several indices perform weighted average denoising on the one-dimensional signal to obtain a preliminarily updated one-dimensional signal .

6. The method for enhancing a one-dimensional signal of single-pixel imaging based on the similarity of encoded images according to claim 5, wherein In Step 3.2, the weighted average conforms to the following formula: ， Among them, , ,…, are the weight coefficients of the one-dimensional voltage signals corresponding to the first indexes, and are determined according to the mean square error value MSE corresponding to the first indexes : ， where h is a weight stretching factor, and the denominator serves for normalization.

7. The method for enhancing a one-dimensional signal of single-pixel imaging based on the similarity of encoded images according to claim 1, wherein The said step 4 includes using a weight factor to adjust a one-dimensional signal and a preliminarily updated one-dimensional signal to obtain a finally available one-dimensional signal : ， In the formula, the weighting factor is positively correlated with the signal-to-noise ratio of the one-dimensional signal .

8. The method for enhancing a one-dimensional signal of single-pixel imaging based on the similarity of encoded images according to claim 7, characterized in that According to each one-dimensional signal With respect to the degree of deviation from the mean of all one-dimensional signals, the weighting factor Is set as a variable , and the finally available one-dimensional signal Is rewritten as: ， In the formula, the weight factor variable is positively correlated with the signal-to-noise ratio of the one-dimensional signal and is negatively correlated with the degree of deviation of the one-dimensional signal from the mean value of all one-dimensional signals.

9. An electronic device, characterized in that, Comprising: One or more processors; A memory for storing one or more programs; wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method for enhancing a one-dimensional signal in single-pixel imaging based on encoded image similarity according to any one of claims 1 - 8.

10. A computer-readable storage medium, characterized in that, Stored thereon are executable instructions, which when executed by a processor can enable the processor to implement the method for enhancing a one-dimensional signal in single-pixel imaging based on encoded image similarity according to any one of claims 1 - 8.

Images