CN110533733B - Method for automatically searching target depth based on ghost imaging calculation - Google Patents

Abstract

The invention discloses a method for automatically searching for target depth based on calculating ghost imaging, which comprises the steps of firstly restoring a target image at a specified position of a target placed in a system searching range by using a calculating ghost imaging system, then automatically selecting an imaging window for the image by adopting a method for selecting the imaging window based on growth segmentation, and calculating the depth and the image of the image which needs to be restored in each step of iteration by adopting a method for iterating the searching depth based on calculating ghost imaging system and based on image standard deviation until a correct target position is searched. The invention reduces the iteration times of calculating the depth of the ghost imaging system for searching the target; no prior knowledge of the target is required; the imaging window is selected, so that the influence of the recovered image background noise on the standard difference of the calculated image is reduced, and the calculated amount when the target image is recovered is reduced; the depth detection technology is combined with the ghost imaging technology, so that the ghost imaging technology can be practically applied.

Description

Method for automatically searching target depth based on ghost imaging calculation

Technical Field

The invention belongs to the technical field of image processing, and particularly relates to a method for automatically searching a target depth position based on ghost imaging calculation.

Background

The ghost imaging technology is a technology using dual light path imaging, which uses two beams of light separated by a light source, wherein one beam of light is used as an imaging light path and the other beam of light is used as a reference light path, and finally a barrel detector without spatial resolution capability collects light intensity signals reflected or transmitted by a target in the imaging light path and recovers the image of the target by using the correlation of the light intensity signals and the light field intensity of the reference light path. The invention of ghost imaging changes the human knowledge that only an area array detector can image, so that the human can obtain the image of the target by using a cheaper barrel detector. And due to the unique physical and mathematical properties of ghost imaging, it possesses properties of atmospheric turbulence resistance and lensless imaging compared to traditional imaging methods. However, the imaging optical path of ghost imaging is relatively complex to implement, and in the development of two decades, researchers at home and abroad have been trying to improve ghost imaging technology to implement the practical ghost imaging technology.

Computational ghost imaging (pharmaceutical REVIEW a) is a popular Computational imaging technique emerging in recent years. The ghost imaging technology adopts a unique single light path imaging technology, the imaging light path is changed into a mode that a spatial light modulator or a digital micro-mirror array modulates a light field, and the light field recording in the reference light path is changed into a mode that a computer calculates. Calculating the ghost imaging simplifies the optical path of the ghost imaging technique, making practical use of the ghost imaging technique possible.

In recent years, Ghost imaging technology is widely applied to the fields of three-dimensional imaging and target distance detection, for example, in 2012, Zhao CQ proposes a Ghost imaging lidar scheme, and realizes a real space image of a target within an estimated distance with high resolution through a compressed sensing Ghost Imaging (GISC) technology (Ghost imaging virtual space imaging constraints.appl.phys.lett); in 2013, Sun BQ proposed that three-dimensional restoration of a target using four bucket detectors in different positions based on computed ghost imaging from shadow restoration three-dimensional morphology (SFS) technique, which requires only a single light source to solve the three-dimensional morphology of the target compared to the conventional SFS technique (3D computerized imaging with single-pixel detectors.

Although the application of ghost imaging in the field of three-dimensional imaging is greatly promoted by the proposal of the technologies, the technologies require that the distance between a target and a light source in an imaging light path is known in advance, and the three-dimensional shape of the target is recovered by using a focused clear ghost imaging image. When the distance measurement is inaccurate, the image is out of focus, so that the image quality is reduced and the three-dimensional form of the target cannot be recovered.

In order to accurately detect the distance between the light source and the target and obtain a focused clear image, researchers have proposed applying autofocus techniques to computing-based ghost imaging systems. Liu proposes to evaluate the out-of-focus image by using an image evaluation method based on an edge detection algorithm, and searches each depth in a search interval by using a traversal search algorithm (Distance measurement by computational cost imaging, optical-International Journal for Light and electronic Optics). Although this method can search for the correct target depth, the search algorithm is computationally expensive. Yang proposes that the image defocusing degree is evaluated by a method for calculating the signal-to-noise ratio of the image, and the correct target depth in a Search interval is searched by a variable-step traversal Search method (High Quality computerized host Imaging Using an optimal Distance Search method. IEEE Photonics Journal). However, the technical problem of poor practicability is still not solved because the calculation of the signal-to-noise ratio needs the prior knowledge of the target form.

Disclosure of Invention

The invention aims to provide a method for automatically searching for the target depth based on the computed ghost imaging, which shortens the time for computing the ghost imaging system to restore an image in the focusing process and increases the search range for computing the ghost imaging system to automatically search for the target depth.

The technical solution for realizing the purpose of the invention is as follows: a method for automatically searching for target depth based on calculating ghost imaging includes utilizing calculating ghost imaging system to restore target image at specified position on target placed in system search range, then utilizing growth division based imaging window selection method to automatically select imaging window for said image and utilizing image standard deviation based search depth iteration method to calculate depth and image of image to be restored in each step of iteration till correct target position is searched.

Compared with the prior art, the invention has the following remarkable advantages: (1) the iteration times for calculating the target depth searched by the ghost imaging system are reduced; (2) no prior knowledge of the target is required; (3) the imaging window is selected, so that the influence of the recovered image background noise on the standard difference of the calculated image is reduced, and the calculated amount when the target image is recovered is reduced; (4) the depth detection technology is combined with the ghost imaging technology, so that the ghost imaging technology can be practically applied.

The present invention will be described in further detail with reference to the accompanying drawings.

Drawings

FIG. 1 is a diagram of an optical path of a computed ghost imaging system.

FIG. 2 is a flow chart of the growth segmentation based imaging window selection method of the present invention.

Fig. 3 is a flowchart of the image standard deviation-based search depth iterative method of the present invention.

Fig. 4 is a schematic diagram of performing a depth search on an object located in a system search interval by using the present invention, where the object is placed at a position where z is 10cm, and the target position is accurately searched by the method after five iterations.

Detailed Description

The invention relates to a method for automatically searching for target depth based on computational ghost imaging, which uses a computational ghost imaging system to perform preliminary imaging on a target at a specified depth; automatically selecting an imaging window using a growth segmentation based imaging window selection method with an image at a specified depth; and searching the depth position of the target in the system search interval by using a search depth iterative method based on the image standard deviation. In the search depth iteration method, each depth image recovery is based on a method for imaging a target at a specified depth by a computing ghost imaging system and an imaging window.

The method comprises the following specific steps:

step 1, a calculation ghost imaging system is utilized to carry out target image restoration on the position of a target in a system search interval on a specified axis, and the calculation ghost imaging system comprises a laser, a spatial light modulator, a computer and a barrel detector without spatial resolution capability. The target image recovery process is as follows: a computer generates a phase template with random variation in advance and puts the phase template on the reflective spatial light modulator frame by frame; a laser (such as a red light 632.8nm helium neon laser) is used as a light source to emit collimated laser to a spatial light modulator, and the spatial light modulates the laser phase according to a phase template to form speckles and irradiate a projection target; a ghost imaging system was computed using a bucket detector without spatial resolution to receive the transmitted light intensity of the target as shown in FIG. 1. The image of the target is restored according to an imaging formula for calculating ghost imaging, wherein the calculation formula is as follows (1):

G(x,y)＝<I ⁽ⁱ⁾ ·P ⁽ⁱ⁾ (x,y)>-<I ⁽ⁱ⁾ ><P ⁽ⁱ⁾ (x,y)> (1)

<…>to average. To be provided with<I ⁽ⁱ⁾ ·P ⁽ⁱ⁾ (x,y)>For the purpose of example only,<I ⁽ⁱ⁾ >and<P ⁽ⁱ⁾ (x,y)>the calculation method is the same.

In the formula (2), n is the number of times of signal acquisition of the single-pixel camera. The spatial light modulator is an initial plane of the system, and the position of the plane on the optical axis of the system is recorded as z being 0. And x and y are plane coordinates of the system axis where z is L, and L is any length in the system search range. G (x, y) is the restored image result of the computed ghost imaging system. I is ⁽ⁱ⁾ To calculate the light intensity detected by a single-pixel camera in a ghost imaging system when the computer is delivering the ith frame. P ⁽ⁱ⁾ (x, y) to compute the ith frame speckle at z-L in a ghost imaging system, P is defined ⁽ⁱ⁾ The range taken for (x, y) is the illumination range of the speckle in the computed ghost imaging system on a cross-section perpendicular to the optical axis of the computed ghost imaging system. P ⁽ⁱ⁾ (x, y) does not require probe acquisition and can be calculated from equation (3).

E _z＝0 The complex amplitude of the light field at z-0 on the system axis is the phase template generated in advance by the computer. λ is the wavelength of the laser. ζ and η are plane coordinates on the system axis where z is 0.

For the section with the search range z ═ L, R (L and R are two boundaries of the system search range, respectively), the depth z ═ L + (R-L)/2 is defined as the specified depth of the step, and a target image is restored at the specified depth.

Step 2, selecting and calculating an imaging window of the ghost imaging system by using the image at the specified depth obtained in the step 1, wherein the imaging window is defined as P ⁽ⁱ⁾ (x, y) is within the range. P used due to restoration of target image ⁽ⁱ⁾ When the (x, y) is the whole speckle image, excessive background noise can be introduced when the target image is subsequently restored, and the target depth search is influenced, so that the imaging window is automatically selected by adopting an imaging window selection method based on growth segmentation, the influence caused by the background noise is reduced, and the calculation efficiency of the subsequently restored image is improved. The flow of the imaging window selection method based on growth segmentation is shown in fig. 2, and the specific implementation process is as follows:

the image restored by formula (1) in step 1 is divided into a plurality of regions, which are called super pixels, using m × m pixels (e.g., 5 × 5 pixels) as a basic unit.

And calculating the average gray value g of all pixels contained in each super pixel, and selecting the super pixel with the maximum g as the initial super pixel.

Setting a threshold parameter T as an average gray value of the image, and searching boundary superpixels in four directions (up, down, left and right) in the image by using a search iteration method with the starting superpixel as a starting position, wherein the boundary superpixels are defined as superpixels located at the boundary of a new imaging window. The search iteration method is as follows (only one direction is explained, and the other directions are the same): for the ith iteration, the average gray value of the pixels of which the ith superpixel in the current search iteration direction contains all pixels starting from the starting superpixel is represented as g (i). If g (i ≦ T, it indicates that this superpixel is the boundary superpixel in that direction and stops the iteration. Otherwise, this superpixel is not the boundary superpixel in that direction, and the (i + 1) th iteration is continued.

When the boundary super pixels in the four directions are found, a closed area formed by connecting the row where the boundary super pixels searched in the upper and lower directions are located and the column where the boundary super pixels searched in the left and right directions are located is a new imaging window. Based on calculating P in ghost imaging system ⁽ⁱ⁾ The position and the size of a new imaging window in the initial imaging window are obtained by the mapping relation of (x, y) and G (x, y), and the subsequent recovery of the computed ghost imaging target image is adoptedP in the New imaging Window ⁽ⁱ⁾ (x, y) restoring the target image at each depth.

And 3, searching the depth position of the target in the search interval by using a search depth iterative method based on the image standard deviation. The image standard deviation function is a reference-free image quality evaluation function, when the target is located in an imaging interval of the computed ghost imaging system, and the imaging state of the system changes from out-of-focus to in-focus to out-of-focus, the value of the image standard deviation function gradually increases and decreases along with the change of the state, and the function value takes the maximum value when the system is in focus. Therefore, the characteristic of the image standard deviation function can be used to match with a searching method to search the maximum position of the function, so as to achieve the purpose of detecting the depth of the target.

Several parameters are initialized as output parameters of the method: z ═ BL, BR ] is the search interval of the method, BL and BR are the two boundary depths of the search interval. The first iteration requires the recovery of the image at depth z1 ═ 2 (BL + BR). And defining the depth precision delta z required to be searched, and taking the value as the axial resolution of the computed ghost imaging system. With reference to fig. 3, for the ith iteration, the specific process is as follows:

calculating the depth z of the target image needing to be recovered in iterative search _i The calculation formula is shown as formula (4):

BL and BR are the two boundary depths of the search interval for this iterative search.

Recovering z ═ z using equation (1) _i And z ═ z _i The image at + δ z, the restored image are respectively marked as I _zi And I _zi+δz And calculating the image standard difference of the two restored images at the same time, and recording as S _zi And S _zi+δz The calculation formula is shown in formula (5):

S＝∑ _p ∑ _q (I(x,y)-u) ² (5)

i (x, y) is the gray value of the pixel at coordinate (x, y) in the restored image, u is the average gray value of the whole image, p and q are the number of pixels in rows and columns in the image, respectively, and Σ is the accumulated sign in mathematics.

Comparison S _zi And S _zi+δz If S is _zi >S _zi+δz The optimum points are stated to lie in z ═ BL and z ═ z _i Otherwise the optimal point is between z BR and z zi + δ z. Only the optimum points are described here as z ═ BL and z ═ z _i The optimal point lies in z ═ BR and z ═ z _i Search iterations between + δ z are similar.

Calculating the iterative judgment parameter ZDL ═ z _i BL if ZDL>2 δ z, meaning that in the next iteration the target depth does not converge within δ z, then the parameters need to be updated: let BR be z _i And i is i +1, and entering the next iteration. Otherwise, stopping the iterative search, wherein the target depth given by the search depth iterative method is z ═ z (z) _i +BL)/2。

A schematic diagram of the iterative search process is shown in fig. 4. Since the minimum axial resolving power of the ghost imaging system is calculated as deltaz, z is calculated from equation (4) to simplify the operation _i If not an integer multiple of δ z, z in the iteration _i Take down the z not more than the maximum calculated by equation (4) _i Is an integer multiple of δ z, e.g. z is calculated from equation (4) for an iteration _i At 11.125cm, z is the number of iterations _i Is 11 cm. In the example iterative search process given in fig. 4, the target is placed at z 10 cm. The initialized BL is 5cm and BR is 21 cm. The first iteration calculates the image standard deviation S of the restored image at the position where z is 13cm _z1 Image standard deviation S from restored image at z 13cm _z2+δz Due to S _z1 >S _z1+δz Updating BR to 13 cm; and the second iteration calculates the image standard deviation value S of the restored image at the position where z is 9cm _z1 Image standard deviation S from restored image at z 9.25cm _z2+δz Due to S _z2 <S _z2+δz Updating BL to be 9.25 cm; the third iteration calculates the image standard deviation S of the restored image at the position where z is 11cm _z1 Image standard deviation value S of restored image at z of 11.25cm _z2+δz Due to S _z1 >S _z1+δz Updating BR to 11 cm; the fourth iteration calculates the image standard deviation S of the restored image at the position where z is 10cm _z1 Image standard deviation S from restored image at z 10.25cm _z4+δz Due to S _z4 >S _z4+δz Updating BR to 10 cm; the fifth iteration calculates the image standard deviation S of the restored image at the position where z is 9.5cm _z1 Image standard deviation value S of restored image at 9.75cm from z _z2+δz Due to S _z5 >S _z5+δz And ZDR is more than or equal to 2 deltaz, so the target depth given by the search iteration method is 9.825cm, the difference value with the target actual depth does not exceed deltaz, and the target depth is considered to be correctly searched.

Claims (4)

1. A method for automatically searching for target depth based on computed ghost imaging is characterized in that: firstly, restoring a target image at a specified position of a target placed in a system search range by using a ghost imaging calculation system, then automatically selecting an imaging window for the image by adopting a growth segmentation-based imaging window selection method, and calculating the depth and the image of the image which needs to be restored in each step of iteration by adopting a ghost imaging calculation system-based image standard deviation search depth iteration method until a correct target position is searched;

the target image recovery process is as follows: a computer generates a phase template with random variation in advance and puts the phase template on the reflective spatial light modulator frame by frame; the laser is used as a light source to emit collimated laser to the spatial light modulator, and the spatial light modulates the laser phase according to the phase template to form speckles and irradiate the projection target; receiving the transmitted light intensity of the target by using a barrel detector without spatial resolution capability;

the image of the target is restored according to the imaging formula of the computational ghost imaging system, i.e.

G(x，y)＝<I ⁽ⁱ⁾ ·P ⁽ⁱ⁾ (x，y)>-<I ⁽ⁱ⁾ ><P ⁽ⁱ⁾ (x，y)> (1)

To calculate the mean value, to<I ⁽ⁱ⁾ ·P ⁽ⁱ⁾ (x，y)>For the purpose of example only,<I ⁽ⁱ⁾ >and<P ⁽ⁱ⁾ (x，y)>the calculation method is the same;

n in the formula (2) is the signal acquisition frequency of the single-pixel camera; the spatial light modulator is an initial plane of the system, and the position of the plane on the optical axis of the system is recorded as z equal to 0; x and y are plane coordinates of a position where z is equal to L on the system axis, and L is any length in the system searching range; g (x, y) is a restored image result of the computed ghost imaging system; i is ⁽ⁱ⁾ Calculating the light intensity detected by a single-pixel camera in the ghost imaging system when the computer puts the ith frame; p ⁽ⁱ⁾ (x, y) to compute the ith frame speckle at z-L in a ghost imaging system, P is defined ⁽ⁱ⁾ (x, y) is the illumination range of the speckle in the computed ghost imaging system on a cross-section perpendicular to the optical axis of the computed ghost imaging system; p is ⁽ⁱ⁾ (x, y) does not require detector acquisition and is calculated from equation (3):

E _z＝0 the complex amplitude of the optical field at the position where z is 0 on the system axis is the phase template generated in advance by the computer, lambda is the wavelength of the laser, zeta is the plane coordinate at the position where z is 0 on the system axis;

for the section with the search range z ═ L, R ], L and R are two boundaries of the system search range, respectively, and the depth z ═ L + (R-L)/2 is defined as the specified depth at which a target image is restored.

2. The method of claim 1, wherein the imaging window of the ghost imaging system is computed by using the image selection at the specified depth, and the method for selecting the imaging window based on the growth segmentation is implemented as follows:

dividing the restored image into a plurality of regions by taking m pixels as basic units, wherein the regions are called super pixels;

calculating the average gray value g of all pixels contained in each super pixel, and selecting the super pixel with the maximum g as an initial super pixel;

setting a threshold parameter T as an average gray value of the image, and searching boundary super pixels in the upper, lower, left and right directions of the image by using a search iteration method by using an initial super pixel as an initial position, wherein the boundary super pixels are defined as super pixels positioned at the boundary of a new imaging window;

when the boundary super pixels in the four directions are found, a closed area formed by connecting a row where the boundary super pixels searched in the upper and lower directions are located and a column where the boundary super pixels searched in the left and right directions are located is a new imaging window; based on calculating P in ghost imaging system ⁽ⁱ⁾ The position and the size of a new imaging window in the initial imaging window are obtained through the one-to-one mapping relation between (x, y) and G (x, y), and P in the new imaging window is adopted for the subsequent recovery of the computed ghost imaging target image ⁽ⁱ⁾ (x, y) restoring the target image at each depth.

3. The method of claim 2, wherein: the search iteration method in one direction is as follows, and the other direction methods are the same: for the ith iteration, the ith super pixel in the current search iteration direction with the starting super pixel as the starting position contains the average gray value of all pixels and is represented by g (i), if g (i) is less than or equal to T, the super pixel is the boundary super pixel in the direction and the iteration is stopped, otherwise, the super pixel is not the boundary super pixel in the direction, and the (i + 1) th iteration is continued.

4. The method of claim 1, wherein: the process of searching the depth position of the target in the search interval by using the search depth iterative method based on the image standard deviation is as follows:

several parameters are initialized as output parameters: z ═ BL, BR ] is the search interval, BL and BR are the two boundary depths of the search interval;

the first iteration needs to restore the image with the depth z1 ═ 2 (BL + BR), the depth precision δ z needed to be searched is defined, and the value is taken as the axial resolution of the computed ghost imaging system;

for the ith iteration, the specific process is as follows:

calculating the depth z of the target image needing to be recovered in iterative search _i The calculation formula is shown as formula (4):

using G (x, y) ═ G<I ⁽ⁱ⁾ ·P ⁽ⁱ⁾ (x，y)>-<I ⁽ⁱ⁾ ><P ⁽ⁱ⁾ (x，y)>Restoring z to z _i And z ═ z _i The image at + δ z, the restored image are respectively marked as I _zi And I _zi+δz And calculating the standard deviation of the two restored images at the same time, and recording as S _zi And S _zi+δz The calculation formula is shown in formula (5):

S＝∑ _p ∑ _q (I(x，y)-u) ² (5)

i (x, y) is a gray value of a pixel with coordinates (x, y) in the recovered image, u is an average gray value of the whole image, p and q are the pixel numbers of rows and columns in the image respectively, and sigma is an accumulated symbol in mathematics;

comparison S _zi And S _zi+δz If S is _zi ＞S _zi+δz The optimum points are stated to lie in z ═ BL and z ═ z _i Otherwise, the optimal point is located between z ═ BR and z ═ zi + δ z; only the optimum points are described here as z ═ BL and z ═ z _i The optimal point lies in z ═ BR and z ═ z _i Search iterations between + δ z are similar;

calculating the iterative judgment parameter ZDL ═ z _i BL | if ZDL > 2 δ z, meaning that the target depth does not converge within δ z in the next iteration, then the parameters need to be updated: let BR be z _i Entering the next iteration, wherein i is i + 1; otherwise, stopping the iterative search, wherein the target depth given by the search depth iterative method is z ═ z (z) _i +BL)/2。

Images