Regional single-pixel imaging method based on optical coding and multiplexing

Abstract

The invention discloses a regional single-pixel imaging method based on optical coding and multiplexing. According to the method, the code division multiple access principle in mobile communication is used for reference, the target to be detected can be reconstructed under the condition of low sampling rate, and higher spatial resolution can be obtained in a specific area. The method comprises the following steps: firstly, carrying out space-time coding on each region of a random illumination mode by utilizing a Hadamard matrix, and projecting the coded illumination mode to a target scene through a digital micromirror device; secondly, a single-pixel detector receives the total light intensity reflected or transmitted by the target and decodes the total light intensity, and the independent light intensity value of each area in a certain mode can be obtained; and reconstructing each region based on a compressed sensing algorithm so as to realize regional single-pixel imaging. According to the invention, on the basis of uniform sampling density of traditional single-pixel imaging, space-time coding is introduced to realize redistribution of the sampling density in different regions of the target, and the spatial resolution of a specific region is effectively improved while the overall imaging capability is maintained.

Description

Technical Field

[0001] The invention relates to a regional single-pixel imaging method based on light coding and multiplexing, and belongs to the field of single-pixel imaging. Background Art

[0002] Single-pixel imaging is an indirect imaging technique closely related to computational ghost imaging. Its basic principle is to spatially modulate a light beam using a series of illumination patterns and project it onto a target. The light signal reflected or transmitted by the target is collected by a single detector with no spatial resolution. Finally, the target image is reconstructed by correlating the measured light intensity with the corresponding illumination pattern.

[0003] To obtain a high-resolution reconstructed image, single-pixel imaging typically requires a number of samples at least equal to the total number of pixels in the target image. As image size increases, the number of samples required can reach tens of millions or more, significantly limiting its practicality. To alleviate this problem, single-pixel imaging methods based on compressed sensing have gained widespread application, aiming to effectively reconstruct images using undersampled measurement data. Furthermore, image reconstruction with non-uniform resolution, where high-resolution reconstruction is performed on key areas of the target image while reducing the resolution of non-key areas, is a common practice in practice for balancing imaging speed and quality.

[0004] Previous studies have shown that because single-pixel imaging does not rely on a fixed Cartesian pixel arrangement, image resolution does not need to be consistent across the entire field of view. This flexibility allows single-pixel imaging systems to dynamically adjust sampling strategies based on specific needs, allowing for more efficient acquisition of information in key areas.

[0005] Currently, single-pixel image reconstruction with non-uniform resolution relies primarily on designing illumination patterns with varying spatial resolution. However, when the total sampling rate is low, the imaging quality of such methods degrades significantly, making it difficult to meet the dual requirements of efficient imaging and high-fidelity reconstruction. Summary of the Invention

[0006] This paper discloses a regional single-pixel imaging method based on optical coding and multiplexing. Drawing on the principle of code division multiple access in mobile communications, this method introduces a spatiotemporal orthogonal coding mechanism to dynamically adjust the sampling density of different regions, significantly reducing the overall sampling count while maintaining image quality in key areas.

[0007] Unlike methods that achieve non-uniform resolution image reconstruction by designing illumination patterns with varying spatial resolutions, this present invention employs a random illumination pattern with consistent spatial resolution. By introducing a spatiotemporal orthogonal encoding mechanism, regional multiplexing and encoding of the target to be measured are performed, thereby achieving single-pixel image reconstruction with non-uniform resolution. Even at relatively low overall sampling rates, this method can achieve high-quality reconstruction by concentrating sampling resources on key areas. This is the most significant technical advantage of this present invention over existing methods.

[0008] The technical solution of the present invention is to first introduce orthogonal spatiotemporal coding into the light beam modulated by the illumination mode to achieve optical multiplexing of different target regions. Then, a single-pixel detector collects and decodes the total light intensity signal reflected by the target in each coding state. After decoding, each region is reconstructed based on a compressed sensing algorithm. The main steps include:

[0009] (1) The target image to be measured is divided into several key and non-key sub-regions. The position and size of each sub-region are pre-set according to subjective needs or prior knowledge before sampling, and the corresponding region is also divided accordingly in the random illumination pattern used for projection;

[0010] (2) A time-space encoder is constructed based on the Hadamard matrix to perform spatiotemporal orthogonal encoding on the random illumination pattern;

[0011] (3) In a single-pixel imaging device, a digital micromirror device (DMD) is used to project a series of random illumination patterns that are orthogonally encoded in time and space onto the target to be measured, and the detector collects the total reflected light intensity of all sub-areas after time-sequential multiplexing;

[0012] (4) Decoding the collected results using a decoder to obtain the light intensity of each sub-area of ​​the target under the corresponding lighting mode;

[0013] (5) Combining the decoded regional light intensity and lighting pattern, image reconstruction is performed using a compressed sensing algorithm.

[0014] The time-space encoder in step (2) is composed of a coding mask derived from a Hadamard matrix. The Hadamard matrix H is an orthogonal matrix containing only +1 and -1 elements, which satisfies HH T =NE, where E is the identity matrix and N is the matrix order. Since the light intensity signal is non-negative, the coding mask is generated by the N-order unipolar Hadamard matrix, denoted as H N . H N It is obtained by replacing all -1 elements in the N-order Hadamard matrix with 0. NIt can be considered as consisting of N chips (i.e., N columns of the matrix), each chip is used to encode a specific spatial region in the lighting pattern. By dividing each chip into N time slots (corresponding to N rows of the matrix), a time-space encoder consisting of N time slots and N spatial regions is constructed. Each column of the encoder represents the transmission status of a spatial region in each time slot, and each row represents the transmission status of all spatial regions in a given time slot.

[0015] The single-pixel detector in step (3) collects the corresponding multiplexed light intensity in each time slot. Taking H4 as an example, the multiplexed light intensity measured in four time slots for any random illumination pattern can be combined and represented as a four-dimensional row vector:

[0016] V i =[I1+I2+I3+I4 I1+I3 I1+I2 I1+I4]

[0017] Among them, I1, I2, I3, and I4 respectively represent the reflected light intensity of each area of ​​the target under the current lighting mode.

[0018] The decoder generation formula used in step (4) is:

[0019]

[0020] Among them, D N For an N-order decoder, H N The inverse matrix of .

[0021] The single-pixel imaging reconstruction algorithm in step (5) adopts a compressed sensing reconstruction algorithm based on total variation regularization, and completes image undersampling and reconstruction under the premise of achieving a 50% sampling rate for the minimum coding area generated in step (1).

[0022] The advantages of the present invention are:

[0023] (1) The present invention achieves single-pixel image reconstruction with regionally variable resolution by spatially redistributing the sampling density. This differs from previous illumination modes that vary spatial resolution by design, which still use a uniform sampling density across the entire field of view, which can easily lead to a decrease in overall reconstruction quality under conditions of low total sampling rates. The present invention concentrates sampling resources in key areas, thereby improving the quality of image details.

[0024] (2) Improved system robustness. By introducing spatiotemporal orthogonal coding, not only is the inherent contradiction between sampling rate and resolution in single-pixel imaging effectively alleviated, but the anti-noise characteristics of code division multiple access technology are also utilized to significantly improve the stability and reliability of the system in complex environments.

[0025] In summary, this paper proposes a regional single-pixel imaging method based on optical coding and multiplexing, which breaks through the limitation of traditional single-pixel imaging systems that impose a uniform sampling rate across the entire field of view. Through spatiotemporal coding, the system can image large-scale scenes at a low overall sampling rate while achieving higher spatial resolution in selected key areas. In addition, this method is compatible with a variety of image reconstruction algorithms and has good versatility and scalability. It has a positive impact on the development of efficient and low-resource computational imaging. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Schematic diagram of the overall process of an embodiment of the present invention.

[0027] Figure 2 Schematic diagram of a single-pixel imaging system. In the figure, 1 is a digital micromirror device (DMD), 2 is a target to be measured, 3 is a convex lens, and 4 is a single-pixel detector.

[0028] Figure 3 Schematic diagram of fourth-order spatiotemporal coding of an illumination pattern in an embodiment of the present invention.

[0029] Figure 4 Schematic diagram of the light intensity vector obtained after multiplexing in an embodiment of the present invention.

[0030] Figure 5 Schematic diagram of non-uniform resolution imaging results reconstructed according to the method in an embodiment of the present invention.

[0031] Figure 6 It is the signal-to-noise ratio and multi-scale structural similarity index of each region in the reconstructed image in the embodiment of the present invention.

[0032] Figure 7 This is a multi-scale structural similarity change curve of the key area under different noise levels in an embodiment of the present invention. DETAILED DESCRIPTION

[0033] In order to more clearly illustrate the technical solution of the present invention, the specific implementation of the present invention will be described below with reference to the accompanying drawings.

[0034] The present invention proposes a regional single-pixel imaging method based on optical coding and multiplexing, and its system architecture is based on a typical single-pixel imaging system, such as Figure 2As shown, it mainly includes: a digital micromirror device, a convex lens, a single-pixel detector and a target to be measured. First, a series of random binary illumination patterns are spatiotemporally encoded, and the encoded illumination patterns are projected onto the target to be measured through the digital micromirror device. In the specific implementation process, the digital micromirror device receives a binary pattern control signal from the controller, controls each micromirror unit to switch to the "on" or "off" state, thereby forming a light intensity distribution corresponding to the pattern. After the target to be measured is illuminated by the coded illumination pattern, its reflected light is converged through the convex lens and received by the single-pixel detector. In each time slot, the detector collects the multiplexed total light intensity signal. Subsequently, through the decoding operation matching the encoding matrix, the independent light intensity information of each sub-area under the illumination pattern can be extracted respectively. Combining the measurement results under all coded patterns, the target scene image is reconstructed. The specific operations are as follows:

[0035] 1. Divide the random binary illumination pattern into several sub-regions. The location and size of each region are pre-set before sampling based on subjective needs or prior knowledge. Since the key region is usually small, it is set to a coding region with the minimum number of pixels to ensure higher sampling density and better image quality during the sampling and reconstruction process.

[0036] 2. A time-space encoder is constructed based on the Hadamard matrix to perform spatiotemporal orthogonal encoding on random binary lighting patterns.

[0037] Taking the fourth-order Hadamard matrix as an example, the process of spatiotemporal encoding of a random binary lighting pattern is as follows: Figure 3 As shown. A fourth-order unipolar Hadamard matrix can be regarded as consisting of 4 code chips (i.e., 4 columns of the matrix), each code chip corresponds to a spatial region, and is used to encode the illumination pattern of the region. By dividing each code chip into 4 time slots (i.e., 4 rows of the matrix), a space-time encoder consisting of 4 time slots and 4 spatial regions can be constructed. In this encoding structure, each column of the matrix represents the transmission state of a spatial region in different time slots, and each row represents the transmission state of each region in a certain time slot. By multiplying the code chip state in each time slot with the corresponding region of a random binary illumination pattern, spatial encoding of the pattern in different time slots can be achieved. Finally, after encoding, an illumination pattern can generate 4 corresponding time slot encoding patterns, which are used to be projected onto the target to be measured in sequence to achieve time multiplexing.

[0038] 3. Place the DMD and single-pixel detector in front of the target to be measured, and sequentially load the encoded illumination patterns into the DMD. Each encoded pattern is projected onto the target scene, and the detector collects the total reflected light intensity from all regions after time-sequential multiplexing.

[0039] When using a fourth-order encoder for spatiotemporal encoding, the target to be measured is divided into four coding areas, denoted as O1, O2, O3, and O4. Under the action of an illumination pattern, the corresponding reflected light intensities of these four areas are denoted as I1, I2, I3, and I4. As the encoding pattern switches within the four time slots, the reflected light intensities of different areas are linearly combined according to the encoding matrix, and the corresponding total light intensity signal is collected by a single-pixel detector in each time slot. Finally, each illumination pattern P i The multiplexed light intensity results obtained in four time slots can form a four-dimensional row vector, denoted as V i .like Figure 4 As shown, it can be expressed as the following matrix operation:

[0040]

[0041] in, Represents the area to be measured Q x The corresponding lighting pattern part. L x For region O x The corresponding number of pixels. is the transpose of the fourth-order monopole Hadamard matrix.

[0042] When the encoding area O2 is set as the key area, that is, it has the least number of pixels L2. In order to obtain good reconstruction results under the compressed sensing reconstruction algorithm based on total variation regularization, the required number of illumination patterns is set to 50% of the number of pixels in the key area. Therefore, the total number of sampling times required by the system is:

[0043]

[0044] Wherein, M is the total number of sampling times required to reconstruct the entire target to be measured, and N is the order of the encoding matrix. In this embodiment, a fourth-order encoding matrix is ​​selected, so N=4.

[0045] 4. Use the decoder to decode the collected results and obtain the light intensity of each sub-area of ​​the target under the corresponding lighting mode.

[0046] The generation formula of the decoder is:

[0047]

[0048] Among them, D N For an N-order decoder, is the inverse matrix of the N-order monopole Hadamard matrix.

[0049] For each lighting mode P i The four-dimensional row vector V formed iBy using the fourth-order decoder D4 for decoding, the reflected light intensity value of each sub-area of ​​the target to be measured under this lighting mode can be restored, which is specifically expressed as:

[0050] V i D4=[I1 I2 I3 I4]

[0051] After decoding, the actual sampling rate of the smallest coding area O2 is 50% of its pixel count. The sampling rates of the other three coding areas are all lower than 50% and different from each other due to the large number of pixels, thus achieving an effective redistribution of sampling density among the areas.

[0052] 5. Combining the decoded light intensity of each area and the corresponding lighting mode, the compressed sensing reconstruction algorithm based on total variation regularization is used to complete the image undersampling reconstruction, such as Figure 5 shown.

[0053] Since the sampling density assigned to each region is no longer uniform, but presents different sparsity levels according to the importance of the region, the final reconstructed image exhibits different resolutions in space. Figure 6 The signal-to-noise ratio (SNR) and multi-scale structural similarity (MS-SSIM) indicators of each region in the reconstructed image are displayed.

[0054] When different encoders (N=2, 4, 8) are used, the MS-SSIM value of the minimum coding area O2 shows different changing trends under the action of Gaussian white noise of different intensities, such as Figure 7 As shown in the figure, when the signal-to-noise ratio (SNR) drops below 20 dB, the MS-SSIM value of the second-order encoder drops significantly, indicating that the low-order encoder is less robust to noise interference. In contrast, although the MS-SSIM value of the high-order encoder also decreases with decreasing SNR, the decrease is smaller and the overall level remains high, indicating that it has stronger noise resistance during image reconstruction.

[0055] Both qualitative observation and quantitative evaluation results prove that the sampling density redistribution mechanism proposed in the present invention through spatiotemporal coding is effective. It can not only realize single-pixel image reconstruction with non-uniform resolution, but also has strong robustness.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for regional single-pixel imaging based on optical coding and multiplexing, characterized by: A random illumination pattern with uniform spatial resolution is adopted. By introducing a spatiotemporal orthogonal coding mechanism, the target to be measured is multiplexed and encoded in different regions. The sampling density between different regions of the target to be measured is flexibly adjusted, thereby achieving single-pixel image reconstruction with non-uniform spatial resolution. Even under low total sampling rate conditions, high-quality reconstruction effects can be achieved by concentrating sampling resources on key areas.

2. The method according to claim 1, characterized in that The imaging process includes the following steps: (1) Sub-region division and temporal and spatial orthogonal coding (1a) The target to be measured is pre-divided into several key and non-key sub-areas, whose spatial positions and sizes are set according to subjective needs or prior information, and corresponding areas are also divided accordingly in the random lighting pattern used for projection; (1b) A time-space encoder is constructed based on the Hadamard matrix. The encoder is composed of a coding mask derived from the Hadamard matrix. The Hadamard matrix H is an orthogonal matrix containing only +1 and -1 elements, which satisfies HH T =NE, where E is the unit matrix and N is the matrix order; since the light intensity signal is non-negative, the coding mask is generated by the N-order unipolar Hadamard matrix, denoted as H N , H N Replace all -1 elements in the N-order Hadamard matrix with 0; N Considered as consisting of N chips, corresponding to the N columns of the matrix, each chip is used to encode a specific spatial region in the lighting pattern; by time-dividing each chip and assigning it to N time slots, corresponding to the N rows of the matrix, a time-space encoder consisting of N time slots and N spatial regions is constructed; each column of the encoder represents the transmission status of a spatial region in each time slot, and each row represents the transmission status of all spatial regions in a certain time slot; (1c) Design H in the temporal-spatial encoder for the key and non-key sub-regions N The specific content of the matrix realizes the spatiotemporal orthogonal encoding of the random lighting pattern; (2) Lighting projection and light intensity recording In a single-pixel imaging device, a digital micromirror device (DMD) projects a series of random illumination patterns that are orthogonally encoded in time and space onto the target. A single-pixel detector records the total intensity of the light field reflected by the target in each time slot. These total intensity measurements are the result of time-sequential multiplexing of all sub-regions of the target. (3) Data decoding and image reconstruction (3a) According to the formula Generate a decoder, where D N For an N-order decoder, H N The inverse matrix of (3b) Decoding the measured total light intensity value using a decoder to obtain the light intensity value of each sub-area of ​​the target under the corresponding lighting mode; due to the effect of the time-space encoder, although the number of light intensity measurement values ​​of the key and non-key sub-areas is the same, the sampling density of the key sub-area is higher, that is, the sampling rate is higher; (3c) A compressed sensing reconstruction algorithm based on total variation regularization is used to reconstruct the image of each sub-region from the light intensity value of each sub-region. Since the sampling rate of the key sub-region is higher, the reconstructed details are better than those of the non-key sub-region, achieving image reconstruction with non-uniform spatial resolution effect.

Images