CN104410789A - Staring super-resolution imaging device and method - Google Patents

Abstract

The present invention provides a staring super-resolution imaging device and method, which are used to solve the problem of low resolution of existing imaging devices. The device comprises an imaging device lens group (1), a detector (2), a detector driving platform (3), a driver (4), a storage unit (5), an image processing circuit (6) and an image display circuit (7); The detector (2) is arranged in the focal plane on the optical path of the lens group (1) of the imaging device; the driving part of the driver (4) drives the detector driving platform (3) to generate movements of random values in size and direction. The steps of realizing the imaging method: setting the command of the number of changes; the driver (4) drives the detector (4) to perform random shaking of the corresponding number of times; the optical signal is imaged on the detector (4) through the lens group (1) of the imaging device, and obtained and changed Low-resolution images corresponding to the number of frames; use variational Bayesian algorithm to reconstruct low-resolution images. Suitable for video reconstruction and satellite capture.

Description

Staring type super-resolution imaging device and method

technical field

The invention belongs to the technical field of super-resolution imaging, and in particular relates to a staring super-resolution imaging device and imaging method for multi-frame image acquisition and reconstruction, which can be used for video reconstruction and satellite shooting.

Background technique

Improving the resolution of the optical imaging system is the relentless pursuit of image science research and its engineering applications. In addition to improving the resolution by improving the performance of imaging system-related components, such as optical system focal length, aperture and detectors, you can also choose Adding suitable optical devices to the imaging system can improve the resolution, but adjusting the structure of the imaging system will lead to the complexity of the system and a large increase in tasks to a certain extent, so how to restore or reconstruct super-resolution based on existing imaging equipment Image has become an urgent need in many image application fields today. In scientific research, people gradually focus on applying modern optics and digital image processing technology to improve the resolution of the imaging system.

Improving the resolution of an imaging device can be achieved by reconstructing a single-frame image or multiple-frame images. Due to the limited amount of information contained in a single-frame image and the lack of new information in the reconstruction process, the effect of improving the resolution is not ideal. The super-resolution reconstruction algorithm based on multi-frame images uses multiple low-resolution images of the same scene to obtain a frame of super-resolution images of the scene, because the amount of information contained in multi-frame images is greater than that of single-frame images, image sequences They contain similar but not identical complementary information and certain prior information, which makes it possible to restore real and effective super-resolution images.

At present, there are three ways to acquire multi-frame images: the first one is to acquire multi-frame low-resolution images by moving the lens. Shenzhen Yatu Digital Video Technology Co., Ltd. disclosed a system for obtaining multi-frame low-resolution images through lens translation in the application document of patent application number 201010505956.9. It is more complex, and the optical design is limited by the micro-displacement mechanism, so the versatility is poor; the second is to add a parallel plate to the optical path, and obtain multiple frames of images by swinging or rotating the position of the plate, the patent of Beihang University with application number 201210451785.5 In the application, a method of adding parallel plates in the optical path to obtain multi-frame images is disclosed. This method can achieve the effect of super-resolution real-time imaging of dynamic scenes, which is a relatively practical way. However, it requires high processing precision for parallel plates, and at the same time, the change of the optical path will lead to the complexity of the system and a large increase in the workload. Therefore, in practical applications, the cost is too high and the difficulty is too great. The number of frames of the same scene limits the improvement of resolution; the third is to obtain multi-frame images by means of moving detectors. Tsinghua University discloses a method of The method is a device for obtaining multi-frame images, and obtains a series of low-resolution images with relative micro-rotations by controlling the transfer mechanism or micro-rotating the photodetector array. Compared with the first method, the device has high practicability; compared with the second method, because no new optical elements are introduced, the system will not be complicated, and the manufacturing cost is lower. However, the precise control of displacement in the third method requires high hardware requirements, and the pre-set micro-rotation angle limits the range of acquired multi-frame images and the number of frames of images collected from the same scene, resulting in unsatisfactory reconstruction results.

Contents of the invention

The purpose of the present invention is to overcome the defects of poor versatility, complex process, and poor reconstruction effect in the prior art, and provides a staring super-resolution imaging device and method, which obtains multiple frames of images by randomly moving the detector, and is used for The problem of low resolution of the existing imaging device is solved.

In order to achieve the above object, technical solutions of the present invention include:

The lens group 1 of the imaging device is used to collect the optical signal and obtain the collected optical signal;

The detector 2 is used to receive the collected optical signal, form an image on it, and output the formed image to the storage unit 5, the image processing circuit 6 and the image display circuit 7 in sequence;

The detector driving platform 3 is used to load the detector 2 and drive it to move;

The driver 4 is used to drive the detector to drive the platform 3 to move, and form multiple frames of images on the detector 2;

The storage unit 5 is used to store multiple frames of images obtained by moving the detector 2 in real time;

The image processing circuit 6 is used to reconstruct the multi-frame images in the storage unit 5 to obtain super-resolution images;

Image display circuit 7, for outputting a reconstructed super-resolution image;

The detector 2 is arranged on the focal plane of the imaging device lens group 1, and is fixed on the side of the detector driving platform 3 facing the imaging device lens group 1; a rigid connection is adopted between the driver 4 and the detector driving platform 3 , so that the movement of the two during work is consistent, and the size and direction of the movement are random values.

The detector 2 adopts a charge coupled device CCD or a complementary metal oxide semiconductor CMOS or a charge injection device CID.

The movement of the detector driving platform 3 is random shaking in the horizontal direction and vertical direction in the plane perpendicular to the collection light, and its shaking range in the horizontal direction is less than 0.23% of the width of the detector. The jitter range in the vertical direction is less than 0.23% of the length of the detector.

The number of random shakes of the detector driving platform 3 is the same as the number of image frames obtained by the detector 2 .

The driver 4 is a piezoelectric ceramic driver.

The imaging method for realizing staring type super-resolution imaging comprises the following steps:

Step 1: According to the required imaging effect, set the change frequency command of the driving part of the driver;

Step 2: Using the imaging lens group 1 to collect optical signals to obtain the collected optical signals;

Step 3: According to the command of the number of changes set, the driver 4 drives the detector 2 to perform random jitters of corresponding times in the horizontal and vertical directions respectively. The jitter range is less than 0.23% of the length and width of the detector 2. The collected optical signal is Imaging on the detector 2 to obtain multi-frame low-resolution images corresponding to the number of frames and the number of changes;

Step 4: Utilize the storage unit 5 to store the multi-frame low-resolution images obtained above in real time;

Step 5: Utilize the variational Bayesian algorithm in the image processing circuit 6 to reconstruct the multi-frame low-resolution images in the storage unit 5 to obtain corresponding super-resolution images;

Step 6: Utilize the image display circuit 7 to output the above-mentioned super-resolution image obtained through calculation and reconstruction.

The variational Bayesian algorithm adopted in the fifth step above includes the following steps:

Step 5.1 Establish a reconstruction constraint model to obtain the posterior probability function P(x/y _k ):

Reconstruction constraint model y _k ＝DH _k C(s _k )x+n _k ＝B _k (s _k )x+ _nk , where x represents the high-resolution image of the scene, and its number of pixels is PN; y _k represents the obtained Low-resolution image, the number of pixels is N; P represents the improvement of the spatial resolution of the image by the algorithm; k represents the number of image frames obtained that is not greater than the total number of frames of the low-resolution image; D is the downsampling matrix of N×PN, H _k is the PN×PN motion matrix, C(s _k ) is the PN×PN motion matrix generated by the motion vector s _k , n _k is the N×1 noise, and the downsampling, blurring and deformation effects of the imaging system can be combined form an N×PN system matrix B _k (s _k ); _nk , H _k and s _k of each low-resolution image can be different, and P(x/y _k ) can be obtained according to the reconstruction constraint model, and then Find its negative logarithm -lnP(y _k /x);

Step 5.2 Establish a priori constraint model to obtain the prior probability density function P(x):

Establish a feature space containing image spatial scale information and directional information composed of the second-order partial derivatives of Laplacian pyramid and Gaussian pyramid, and use the prediction of Laplacian pyramid as the initial estimate, and obtain the image containing image by accelerating block matching method The estimated gradient prior knowledge of high-frequency information, that is, to obtain P(x), and then obtain its negative logarithm -lnP(x);

Step 5.3 According to the Bayesian maximum a posteriori probability estimation theory, in order to obtain the optimal high-resolution image estimation value x, it is necessary to first obtain the value of x that makes the posterior probability P(x/y _k ) take the maximum value; because y _k is the obtained known low-resolution image, so P(y _k ) is a constant, so the maximum posterior probability of high-resolution image is: $\stackrel{\‾}{x}=\arg \underset{x}{\max }\ln P(x/{\mathrm{the\; y}}_k)=\arg \underset{x}{\min }(-\ln P({\mathrm{the\; y}}_k/x)-\ln P\left(x\right)),$ In the formula Represents the optimal estimated super-resolution image; using the calculation results obtained in the above two steps, the optimal estimated super-resolution image is obtained using the steepest descent method in the maximum a posteriori probability framework , which is the final result of the super-resolution image.

Compared with the prior art, the present invention has the following advantages:

1. The present invention adopts the method of multi-frame image reconstruction to obtain the output image. Compared with the method of direct imaging with ordinary cameras, images with higher resolution can be obtained under the condition of using the same detector.

2. The imaging device of the present invention obtains multi-frame images by moving the detector. Compared with the method of moving the lens or adding optical elements in the optical path in the prior art, the process is simple, the cost is low, and it can achieve The gaze effect increases the versatility of the device.

3. The magnitude and direction of the detector shaking displacement in the present invention are random. Compared with the fixed movement of the detector in the prior art, the super-resolution image can be obtained when the motion parameters are unknown, and it is suitable for generating random displacement by itself. shooting devices, such as satellites.

4. The number of random vibrations of the detector in the present invention can be set by the command of the number of changes. Compared with the number of moves of the detector in the prior art is a fixed value, the more the number of changes, the corresponding increase in the obtained low-resolution images , through later reconstruction, the resolution of the final output image is higher.

5. In the imaging method of the present invention, the variational Bayesian algorithm is applied in low-resolution image reconstruction, which introduces distribution estimation, enhances the stability of the algorithm to noise, and effectively improves the image resolution.

Description of drawings

Fig. 1 is a structural schematic diagram of the present invention.

Fig. 2 is a schematic diagram of a simple detector displacement method in the present invention.

Fig. 3 is a schematic diagram of an optional image acquisition when the detector moves randomly four times and the displacement is half a pixel in the present invention.

Fig. 4 is a schematic diagram of a super-resolution reconstruction in which the low-resolution image is four frames in the present invention.

detailed description

In order to make the purpose of the present invention, the technical problem to be solved and the technical solution clearer, the present invention will be further described in detail below in conjunction with the accompanying drawings.

With reference to Fig. 1, the present invention comprises imaging device lens group 1, detector 2, detector driving platform 3, driver 4, storage unit 5, image processing circuit 6 and image display circuit 7; In order to ensure that detector 2 can successfully receive images And move synchronously with the detector driving platform 3, the detector 2 is located on the optical path of the lens group 1 of the imaging device and in its focal plane, the detector 2 adopts a charge-coupled device CCD or a complementary metal oxide semiconductor CMOS or a charge injection device The CID is fixed before the detector driving platform 3; the driver 4 can be a piezoelectric ceramic driver, which is arranged on one side of the detector driving platform 3, and the shell is fixedly connected with the body to ensure that it does not move relative to itself during the driving process; The driver 4 is rigidly connected to the detector platform 3, such as rod connection, welding or bolt connection, to ensure the synchronous movement of the detector platform 3 and the detector 2; the detector 2 is connected to the storage unit 5, the image processing circuit 6 and the image display circuit 7 in turn form circuit connections for transferring and processing images.

The imaging principle of the present invention is: according to the required imaging effect, set the change frequency command of the drive part of the driver; use the imaging lens group 1 to aim at the scene target to collect the optical signal, and obtain the collected optical signal; when the imaging device starts to work , according to the set number of changes, the driving part of the driver 4 drives the detector driving platform 3 to move, thereby controlling the random vibration of the detector 2 in the focal plane of the lens group, and the displacement of the detector 2 in the horizontal and vertical directions respectively Less than 0.23% of the length and width of the detector, the magnitude of the displacement is a random value; detector 2 obtains low-resolution images corresponding to the number of frames and the set number of changes, the number of frames is represented by L, and the obtained low-resolution images of each frame The high-rate image is stored in the middle of the storage unit 5 of the camera in real time, until the number of image frames stored in the storage unit 5 reaches the L frames corresponding to the set number of times of change instruction; after shooting L frames of images for a scene, the image digital information is transmitted to The image processing circuit 6 reconstructs it to obtain a super-resolution image, and finally the image display circuit 7 outputs and displays the reconstructed super-resolution image.

The principle of L frame image reconstruction in the above-mentioned image processing circuit 6 is as follows:

The algorithm used in the image processing circuit 6 is the variational Bayesian algorithm. The core of the variational Bayesian reconstruction algorithm is to use the prior knowledge of the form of the unknown probability density and the value range of the unknown parameter from the training sample itself. Information, calculate the posterior probability P(y _k /x).

Rebuild constraint model:

y _k ＝DH _k C(s _k )x+n _k ＝B _k (s _k )x+n _k (1)

Among them, D is the downsampling matrix of N×PN, H _k is the motion matrix of PN×PN, C(s _k ) is the motion matrix of PN×PN generated by the motion vector s _k , n _k is the noise of N×1, and the imaging The system's downsampling, blurring and deformation effects can be combined into an N×PN system matrix B _k (s _k ). _nk , H _k and s _k of each frame of low-resolution images may be different. Among them, x represents the high-resolution image of the scene, and its number of pixels is PN; y _k represents the obtained low-resolution image, and its number of pixels is N; P represents the improvement of the spatial resolution of the image by the algorithm;

Under the condition of known y _k , the posterior probability of x can be written as:

$\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\frac{\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span>}{\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; )</span>}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

Among them, P(y _k /x) is the conditional probability density function of observing y _k when x is known; P(x) and P(y _k ) represent the prior probability of x and y _k respectively. According to Bayesian maximum a posteriori probability estimation theory, in order to obtain the optimal estimated super-resolution image, it is necessary to find the x that maximizes the posterior probability P(x/y _k ). Because y _k is known, P(y _k ) is a constant, and the logarithmic function is a monotonically increasing function, so the maximum a posteriori probability estimate can be expressed as follows:

$\stackrel{<span\; class="notranslate">\; \&OverBar;</span>}{\mathrm{<span\; class="notranslate">\; x</span>}}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arg</span>}\underset{\mathrm{<span\; class="notranslate">\; x</span>}}{\mathrm{<span\; class="notranslate">\; max</span>}}\mathrm{<span\; class="notranslate">\; ln</span>}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; /</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arg</span>}\underset{\mathrm{<span\; class="notranslate">\; x</span>}}{\mathrm{<span\; class="notranslate">\; min</span>}}<span\; class="notranslate">\; (</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; ln</span>}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>{\mathrm{<span\; class="notranslate">\; the\; y</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; /</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; ln</span>}\mathrm{<span\; class="notranslate">\; P</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; x</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 3</span>}<span\; class="notranslate">\; )</span>$

Denotes the best estimated super-resolution image.

As can be seen from formula (3), the steps of L frame image reconstruction in the above-mentioned image processing circuit 6 are as follows:

Step 1: Establish a reconstruction constraint model to obtain -lnP(y _k /x), and the reconstruction constraint model is shown in formula (1);

Step 2: Establish a priori constraint model to obtain -lnP(x). A feature space containing image spatial scale information and directional information composed of the second-order partial derivatives of the Laplacian pyramid and the Gaussian pyramid is established, and the Laplacian pyramid prediction is used as the initial estimate, and the accelerated block matching method is used to obtain the feature space containing The estimated gradient prior knowledge of image high-frequency information, that is, to find -lnP(x);

Step 3: Integrate the results obtained in the above steps 1 and 2 into the maximum a posteriori probability framework, and use the steepest descent method to obtain the optimal estimated super-resolution image x, which is the final result of the super-resolution image.

The variational Bayesian algorithm puts the unknown super-resolution image and motion parameters in the same framework for joint estimation, and introduces certain uncertainty into the estimation process through the variance of the probability distribution of the unknown super-resolution image and motion parameters. , which enhances the stability of the algorithm to noise; moreover, the Bayesian estimation method is used in the algorithm to estimate the distribution of unknown parameters instead of the point estimation of the traditional algorithm, which effectively suppresses the expansion of the estimation error in the algorithm and improves the resolution of the image. improve.

Referring to Fig. 2, it is a schematic diagram of a simple detector displacement mode in the present invention, the driving part of the driver 4 impels the detector driving platform 3 to move, so that the detector 2 at the focal plane of the imaging lens group of the camera is randomly moved in the focal plane. Shaking, assuming that the coordinates of the center point of the detector 2 are (0, 0) during the first shaking, the first frame of image y ₁ is obtained, and the next movement method can be divided into the following steps:

1) Taking the first frame of image as a reference, move the detector 2 horizontally for a distance c ₂ , and vertically move the detector 2 for a distance d ₂ to obtain the second frame of low-resolution image y ₂ . The coordinates of its center are (c ₂ , d ₂ ), the image is transferred to the storage unit 5 for temporary storage.

2) Taking the first frame of image as a reference, move the detector 2 horizontally for a distance c ₃ and vertically move the detector 2 for a distance d ₃ to obtain the third frame of low-resolution image y ₃ , and the coordinates of its center are (c ₃ , d ₃ ), the image is transferred to the storage unit 5 for temporary storage.

3) By analogy, the detector 2 can obtain L-3 frames of images from the fourth to the L-th shaking, combined with the above-mentioned three frames of images, a total of L frames of images are obtained, which are sequentially stored in the storage unit 5 in real time .

4) The L frames of images obtained in the above steps and stored in the storage unit 5 are transmitted to the image processing circuit 6 together.

In order to make the reconstruction algorithm using L frames of low-resolution images as the data source achieve the super-resolution reconstruction effect, it is necessary to ensure that the displacement in the horizontal and vertical directions between the selected images of each frame is at the sub-pixel level, that is, the moving distance c _k and d _k are not the size of an integer pixel. By controlling the random jitter of the detector 2, the L frames of low-resolution images can respectively contain the information of different parts of the original scene, but there is information redundancy between the low-resolution images, so the overlapping area can be processed by the image processing circuit 6 Reconstruct to obtain a super-resolution image, and finally output it through the image display circuit 7.

Referring to Fig. 3, it is a schematic diagram of a possible image acquisition when the detector moves four times at random in the present invention when the displacement is half a pixel. Fig. 3(a) shows the position where the detector shakes for the first time, and Fig. 3(b ) means that the detector has shifted half a pixel to the left relative to the position of the first jitter. Figure 3(c) shows that the detector has shifted up by half a pixel on the basis of Figure 3(b), and Figure 3(d) shows that the detection On the basis of Figure 3(c), the device is shifted by half a pixel to the right. Carry out super-resolution reconstruction process to the obtained four-frame low-resolution image with reference to Fig. 4, be the low-resolution image among the present invention is a kind of super-resolution reconstruction schematic diagram of four frames, in the reconstruction process Fig. 4 (a) The pixels of each low-resolution image are re-registered to obtain a super-resolution image with significantly improved resolution as shown in Figure 4(b).

Claims (8)

1. A staring type super-resolution imaging device, comprising: The lens group (1) of the imaging device is used to collect the optical signal and obtain the collected optical signal; The detector (2) is used to receive the collected optical signal, form an image on it, and output the formed image to the storage unit (5), image processing circuit (6) and image display circuit (7) in sequence; The detector driving platform (3) is used to load the detector (2) and drive it to move; The driver (4) is used to drive the detector to drive the platform (3) to move, and form multiple frames of images on the detector (2); The storage unit (5) is used for real-time storage of multi-frame images obtained by moving the detector (2); An image processing circuit (6), used to reconstruct the multi-frame images in the storage unit (5), to obtain a super-resolution image; Image display circuit (7), for outputting a reconstructed super-resolution image; It is characterized in that: the detector (2) is arranged on the focal plane of the lens group (1) of the imaging device, and is fixed on the side of the detector driving platform (3) facing the lens group (1) of the imaging device; the driver (4) A rigid connection is adopted between the detector driving platform (3) so that the movement of the two during work is consistent, and the size and direction of the movement are random values. 2 . The staring super-resolution imaging device according to claim 1 , wherein the detector ( 2 ) adopts a charge coupled device (CCD) or a complementary metal oxide semiconductor (CMOS) or a charge injection device (CID). 3. The staring type super-resolution imaging device according to claim 1, characterized in that: the movement of the detector drive platform (3) is performed in a horizontal direction and a vertical direction in a plane perpendicular to the collection light. random jitter, and its jitter range in the horizontal direction is less than 0.23% of the detector width dimension, and its jitter range in the vertical direction is less than 0.23% of the detector length dimension. 4. The staring super-resolution imaging device according to claim 3, characterized in that: the number of random shakes of the detector driving platform (3) is the same as the number of image frames obtained by the detector (2). 5. The staring super-resolution imaging device according to claim 1, characterized in that: the driver (4) is a piezoelectric ceramic driver. 6. utilize the device of claim 1 to carry out the method for staring type super-resolution imaging, it is characterized in that: comprise the steps: 1) According to the required imaging effect, set the change frequency command of the driver; 2) Using the imaging lens group to collect the optical signal to obtain the collected optical signal; 3) According to the command of the number of changes set, the driver (4) drives the detector (2) to perform random jitters in the horizontal direction and the vertical direction respectively, so that the collected optical signal is imaged on the detector (2), Obtain multi-frame low-resolution images with the same number of frames as the number of changes; 4) Utilize the storage unit (5) to store the multi-frame low-resolution images obtained above in real time; 5) Utilize the variational Bayesian algorithm to reconstruct the multi-frame low-resolution images in the storage unit (5) to obtain corresponding super-resolution images; 6) Outputting the above-mentioned super-resolution image obtained through calculation and reconstruction through the image display circuit (7). 7. The method for staring-type super-resolution imaging according to claim 6, characterized in that: the random dithering in the step 3), its dithering range in the horizontal direction is less than 0.23% of the detector width dimension, and in the vertical direction The jitter range in the vertical direction is less than 0.23% of the length of the detector. 8. The method for staring type super-resolution imaging according to claim 6, characterized in that: step 5) utilizes the variational Bayesian algorithm in the image processing circuit (6) to reconstruct the storage unit (5) Multi-frame low-resolution images in , proceed as follows: 5.1) Establish a reconstruction constraint model to obtain the posterior probability function P(x/y _k ): Reconstruction constraint model y _k ＝DH _k C(s _k )x+n _k ＝B _k (s _k )x+ _nk , where x represents the high-resolution image of the scene, and its number of pixels is PN; y _k represents the obtained Low-resolution image, the number of pixels is N; P represents the improvement of the spatial resolution of the image by the algorithm; k represents the number of image frames obtained that is not greater than the total number of frames of the low-resolution image; D is the downsampling matrix of N×PN, H _k is the PN×PN motion matrix, C(s _k ) is the PN×PN motion matrix generated by the motion vector s _k , n _k is the N×1 noise, and the downsampling, blurring and deformation effects of the imaging system can be combined form an N×PN system matrix B _k (s _k ); _nk , H _k and s _k of each low-resolution image can be different, and P(x/y _k ) can be obtained according to the reconstruction constraint model, and then Find its negative logarithm -lnP(y _k /x); 5.2) Establish a priori constraint model to obtain the prior probability density function P(x): Establish a feature space containing the image spatial scale information and directional information composed of the second-order partial derivatives of the Laplacian pyramid and the Gaussian pyramid, and use the Laplacian pyramid prediction as the initial estimate, and obtain the image containing image through the accelerated block matching method The estimated gradient prior knowledge of high-frequency information, that is, to obtain P(x), and then obtain its negative logarithm-lnP(x); 5.3) According to Bayesian maximum a posteriori probability estimation theory, in order to obtain the optimal high-resolution image estimation value It is necessary to obtain the value of x that makes the posterior probability P(x/y _k ) take the maximum value; because y _k is the obtained known low-resolution image, so P(y _k ) is a constant, so the high-resolution image The maximum posterior probability is: $\stackrel{\&OverBar;}{x}=\arg \underset{x}{\max }\ln P(x/{\mathrm{the\; y}}_k)=\arg \underset{x}{\min }(-\ln P({\mathrm{the\; y}}_k/x)-\ln P\left(x\right)),$ In the formula Represents the optimal estimated super-resolution image; using the calculation results obtained in the above two steps, the optimal estimated super-resolution image is obtained using the steepest descent method in the maximum a posteriori probability framework That is, the final result of the super-resolution image.