CN105469416A - Rotary coded aperture imaging system based depth estimation method - Google Patents

Abstract

The invention belongs to the technical field of computer vision. A rotary coded aperture mode is provided, so that obtained depth information is accurate and high in precision, and full-focus restored image quality is improved; and meanwhile, an imaging system is convenient in usage mode and high in stability. For achieving the purpose, the technical scheme adopted by the invention is that a rotary coded aperture imaging system based depth estimation method comprises the following steps: proposing a performance evaluation standard of the rotary coded aperture imaging system; optimizing a rotary coded aperture by using a genetic algorithm and a coordinate descent method according to the evaluation standard; and performing depth estimation and full-focus image restoration operations by virtue of the optimized rotary coded aperture. The method is mainly applied to the image processing.

Description

Based on the depth estimation method of rotary coding aperture imaging system

Technical field

The invention belongs to technical field of computer vision.Specifically, relating generally to Digital Image Processing and calculate shooting field, is the depth estimation method based on rotary coding aperture imaging system.

Background technology

Modern society is more and more higher for the requirement of the image image quality of digital imaging device, and expects to obtain full picture rich in detail.But optical imaging system develops into and all cannot thoroughly solve defocus blur problem now.The basic imaging principle of traditional optical imaging device is that concave-convex lens changes direction to light, uses this characteristic to design imaging device.Therefore the distance between the object plane of subject to object lens has certain selectivity, if subject is on the focal plane being positioned at imaging system, so imaging plane obtain the most clearly as; Otherwise if shot object is away from focal plane, have with it certain distance, the image that so imaging plane obtains will thicken.Before introducing coding aperture technology, the method for head it off is mostly need to use extra device, improves focusing system characteristic to reach higher focusing precision.

Coloured image and the combination of corresponding depth map can bring more fully image restoration information to be applied to such as picture editting, based in the field such as image rendering and augmented reality of the degree of depth.But depth information, compared to the luminosity information that can obtain with the optical device of maturation, is difficult to obtain.Therefore, based on the depth estimation method of blurred picture out of focus due to lower to the requirement of optical device and there is the characteristics such as higher robustness and be widely used in the fields such as the depth recovery of image.

Coding aperture technology learns a skill as a kind of emerging calculating shooting, and the point spread function in itself by modulating imaging system realizes expanding performance based on the depth estimation method of digital camera with the depth of field improving visual light imaging.Wherein, three inherent parameters of camera imaging, i.e. the shape of aperture, imaging focal length, and camera lens is to the distance of imaging sensor, is the important indicator realizing depth information recovery.Coding aperture technology mainly carries out coding reconstruct work to the lens aperture of camera, improves the discrete information of the degree of depth comprised in its point spread function, obtains depth estimation result accurately, and realizes the recovery of total focus picture rich in detail by the depth map information of image.Forefathers based in fuzzy estimation of Depth work out of focus, aperture of singly encoding, coding aperture to or colored aperture in, there is estimation of Depth and contradict with total focus image restoration, practical application complexity or the problem such as color image information distortion.In order to better use this technology, and solve Problems existing in these coding aperture patterns, the present invention uses rotary coding aperture pattern to carry out improving the work such as estimation of Depth and total focus image restoration performance.

Summary of the invention

For overcoming the deficiencies in the prior art, provide a kind of rotary coding aperture pattern, the depth information of acquisition is accurate, and precision is high, and promote total focus restored image quality, meanwhile, the use-pattern of imaging system is convenient, and degree of stability is high.For achieving the above object, the technical scheme that the present invention takes is, based on the depth estimation method of rotary coding aperture imaging system, comprise the steps: the Performance evaluation criterion proposing rotary coding aperture imaging system, and use genetic algorithm and coordinate descent to optimize rotary coding aperture according to this evaluation criterion, after realize estimation of Depth and total focus image restoration work by the rotary coding aperture optimized.

Wherein:

1) Performance evaluation criterion of rotary coding aperture imaging system is proposed,

(1) fuzzy imaging model out of focus:

$f=f_0\⊗k^d+\mathrm{\η}$

Wherein, f represents imaging system and takes the blurred picture out of focus obtained, f ₀represent original picture rich in detail, k ^drepresent the point spread function of imaging system at different depth d place, describe white Gaussian noise, X represents Gaussian distribution, represent the standard deviation of Gaussian noise;

The imaging model out of focus of (2) three rotary coding aperture system:

F _i＝F ₀·K _i ^d+N _i，i＝1，2，3

Wherein, this model is model expression in a frequency domain: F in (1) _irepresent that rotary coding aperture imaging system takes the blurred picture out of focus obtained, F ₀represent original image, K _i ^drepresent the point spread function of different depth, N _irepresent that i-th rotary coding aperture obtains the white Gaussian noise in shooting image, i represents number of revolutions, and namely three rotations can produce three different blurred pictures out of focus; This model corresponds to the mathematical model of three blurred pictures out of focus of the shooting gained of rotary coding aperture imaging system;

(3) energy function minimum value is used to solve maximum a posteriori probability problem (MaximumAPosteriori) to realize the performance optimization to rotary coding aperture.

$E\left(\hat{d}\right|F_1,F_2,F_3,\mathrm{\σ})=\underset{F_0}{min}\underset{i=1,2,3}{\Σ}\left|\right|{\hat{F}}_0\·{K_i}^{\hat{d}}-F_i|{|}^2+||C\·{\hat{F}}_0|{|}^2$

Wherein, the jamtosignal (NoisetoSignalRatios) of representative image priori, can be expressed as σ ²/ A, wherein, A uses the average power spectra of multiple natural images to represent, ξ represents frequency, and use energy function minimum value to mean to make the difference between ambiguous estimation image and original blurred picture minimum, namely make estimation procedure precision the highest, depth recovery result is the most accurate;

(4) use dimension receive warp area method obtain estimate picture rich in detail.Namely to a set estimating depth energy function is used first to solve estimation picture rich in detail namely exist condition under, can obtain,

${\hat{F}}_0=\frac{\underset{i}{\Σ}{F}_i\·{{\stackrel{\‾}{K}}_i}^{\hat{d}}}{\underset{i}{\Σ}{\left|{{\stackrel{\‾}{K}}_i}^{\hat{d}}\right|}^2+{\left|C\right|}^2},i=1,2,3$

Wherein, the complex conjugate of K, refer at set estimating depth the frequency domain representation of the imaging system point spread function that place is corresponding, the ∧ of the letter top in the present invention all represents that the representative variable of this letter is one and calculates estimated value;

(5) bring the formula in step (4) and step (2) into step (3), reconstruct energy function, obtains one about estimating depth through readjusting and simplifying heat-supplied function,

$\begin{array}{c}E\left(\hat{d}\right|{K_1}^{d^{*}},{K_2}^{d^{*}},{K_3}^{d^{*}},\mathrm{\σ})\\ =\underset{\mathrm{\ξ}}{\Σ}A\frac{{\left|K_1^d{K}_2^{d^{*}}-K_2^d{K}_1^{d^{*}}\right|}^2+{\left|K_1^d{K}_3^{d^{*}}-K_3^d{K}_1^{d^{*}}\right|}^2+{\left|K_2^d{K}_3^{d^{*}}-K_3^d{K}_2^{d^{*}}\right|}^2}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+C^2}\\ +{\mathrm{\σ}}^2\·\underset{\mathrm{\ξ}}{\Σ}\[\frac{C^2}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}+1\]\end{array}$

Wherein, d ^*represent the standard depth be positioned at corresponding to imaging system, d represents place's degree of depth of each level of visible images;

(6) the additive white Gaussian noise parameter in spatial domain is under the condition of σ, is positioned at d to rotary coding aperture ^*the estimation of Depth at place is carried out performance evaluation and can be obtained,

$\begin{array}{c}R(K_1,K_2,K_3|d^{*},\mathrm{\σ})\\ =\underset{d\∈D/d^{*}}{\min }E\left(d\right|{K_1}^{d^{*}},{K_2}^{d^{*}},{K_3}^{d^{*}},\mathrm{\σ})-E\left(d^{*}\right|{K_1}^{d^{*}},{K_2}^{d^{*}},{K_3}^{d^{*}},\mathrm{\σ})\end{array}$

Wherein, D={c ₁d ^*, c ₂d ^*..., c _ld ^*each degree of depth level corresponding to representative system, { c _ican be taken as 0.1,0.15 ..., 1.5}, step-length is 0.05.

(7) above result is normalized to the Performance evaluation criterion that can obtain rotary coding aperture:

$\left\{\begin{array}{c}M(K_1,K_2,K_3,d,d^{*})\\ ={\[\frac{1}{n}\underset{\mathrm{\ξ}}{\Σ}\frac{{\left|K_1^d{K}_2^{d^{*}}-K_2^d{K}_1^{d^{*}}\right|}^2+{\left|K_1^d{K}_3^{d^{*}}-K_3^d{K}_1^{d^{*}}\right|}^2+{\left|K_2^d{K}_3^{d^{*}}-K_3^d{K}_2^{d^{*}}\right|}^2}{|K_1^d{|}^2+|K_2^d{|}^2+|K_3^d{|}^2+C^2}\]}^{\frac{1}{2}}\\ R=\underset{d\∈D/d^{*}}{min}M(K_1,K_2,K_3,d,d^{*})\end{array}\right.$

The larger energy function evaluating estimation of Depth that characterizes of R value changes more precipitous, makes estimation of Depth performance stronger for the robustness of picture noise or color texture.

Design and use genetic algorithm optimization to obtain the rotary coding aperture of low resolution; Coordinate descent is used to improve the resolution of rotary coding aperture.

Estimation of Depth is recovered and total focus image restoration process, is to receive deconvolution and aberration method and carry out the work of depth recovery and total focus image restoration by tieing up to it.

Technical characterstic of the present invention and effect:

1) compared to coding aperture to the mode of the replacing aperture needed in use, the use-pattern of rotary coding aperture imaging system is more simple and convenient, obtains image more effectively stable.

2) mode for three times of rotary coding aperture rotated makes the discrete performance of the frequency spectrum of aperture better, therefore, more obvious in the discrete degree of point spread function corresponding to each degree of depth level.In the computation process of estimating depth, the discrete performance of better frequency spectrum makes the performance of the performance of the estimated result of the degree of depth in the accuracy of identification and the resolution of the degree of depth more outstanding, and computing power is robust more.

3) obtaining on the basis estimating depth map more accurately, use dimension receive warp area method restore the effect of total focus picture rich in detail can be better.Compared to the mode of previous coding aperture, the complementary performance of frequency spectrum of rotary coding aperture imaging system is more outstanding, and the frequency coverage of imaging frequency spectrum is larger, and it is more that this makes the spectrum information of visible images retain, also mean the better effects if recovering picture rich in detail, improve the quality of restored image.

4) because restored method of the present invention is simple, easy realization, working time is short, therefore, ambiguity solution can be restored performance and directly be integrated in the hardware of digital camera the work making camera or other imaging devices directly can carry out estimation of Depth and image restoration, decrease the dependence of native system to external unit.

Accompanying drawing illustrates:

Fig. 1 is that the present invention uses genetic algorithm and coordinate descent to optimize the result of rotary coding aperture, wherein, resolution be 11 × 11 aperture result be use genetic algorithm generate low resolution optimize aperture, be the result that coordinate descent improves aperture resolution afterwards, 47 × 47 is the present invention's rotary coding apertures used;

Fig. 2 is the spectral characteristic of rotary coding aperture;

Fig. 3 is the result using rotary coding aperture system variant depth model to be carried out to estimation of Depth recovery;

A () is discrete stair, ladder respectively, continuous inclined-plane, circular conical surface three-dimensional model, model surface half attachment complex texture (left side), half attachment simple textures (right side).

B () is the original depth-map that above four models are corresponding respectively.

The c estimating depth figure corresponding respectively to four models that () coding aperture is right.

D () rotates the estimating depth figure corresponding respectively to four models of circular iris.

The estimating depth figure corresponding respectively to four models of (e) rotary coding aperture.

Following table be coding aperture to, rotate RMSE (root-mean-square error) statistics between circular iris, each estimating depth figure of rotary coding aperture and master pattern depth map.

Fig. 4 is the result using rotary coding aperture system to restore the estimation of Depth of 3 dimension scenes and total focus picture rich in detail.

(1a) (1b) is clear figure and the depth map of the original scene of indoor, outdoor scene respectively.

(2a) (2b) is the right total focus restored image of the coding aperture of corresponding scene and estimating depth figure respectively.

(3a) (3b) is total focus restored image and the estimating depth figure of the rotary coding aperture of corresponding scene respectively.

Following table is the statistics of the RMSE (root-mean-square error) of total focus restored image and the clear figure of original scene and the PSNR (Y-PSNR) of estimating depth figure and original scene depth map, and wherein, db is measurement unit.

Embodiment

The technical scheme that the present invention takes is, based on the depth estimation method of rotary coding aperture imaging system.Propose the Performance evaluation criterion of rotary coding aperture imaging system, and use genetic algorithm and coordinate descent to optimize rotary coding aperture according to this evaluation criterion, after realize estimation of Depth and total focus image restoration work by the rotary coding aperture system optimized.When rotary coding aperture imaging system carries out estimation of Depth to visible images, need by point spread function corresponding to each degree of depth level, namely the otherness of the point spread function at different depth place is the leading indicator parameter of resolution depth.Therefore, realize the key issue that the degree of depth accurately estimates and be that the point spread function of rotary coding aperture imaging system has outstanding discrete performance, corresponding in frequency spectrum, mean that the frequency spectrum of rotary coding aperture needs to have more zero crossing.The frequency spectrum zero crossing of the point spread function that different depth is corresponding is having different positions, and when these zero crossings divide immediately as much as possible, the estimation of depth information will be all the more accurate.Comprising the steps: 1) Performance evaluation criterion that proposes rotary coding aperture imaging system is applied to estimation of Depth and recovers and total focus image restoration:

(1) fuzzy imaging model out of focus:

$f=f_0\⊗k^d+\mathrm{\η}$

Wherein, f represents imaging system and takes the blurred picture out of focus obtained, f ₀represent original picture rich in detail, k ^drepresent the point spread function of imaging system at different depth d place, white Gaussian noise is described.

The imaging model out of focus of (2) three rotary coding aperture system:

F _i＝F ₀·K _i ^d+N _i,i＝1,2,3

Wherein, this model is model expression in a frequency domain in (1).F _irepresent that rotary coding aperture imaging system takes the blurred picture out of focus obtained, F ₀represent original image, K _i ^drepresent the point spread function of different depth, N _irepresent that i-th rotary coding aperture obtains the white Gaussian noise in shooting image, i represents number of revolutions, and namely three rotations can produce three different blurred pictures out of focus.This model corresponds to the mathematical model of three blurred pictures out of focus of the shooting gained of rotary coding aperture imaging system.

(3) energy function minimum value is used to solve maximum a posteriori probability problem (MaximumAPosteriori) to realize the performance optimization to rotary coding aperture.

$E\left(\hat{d}\right|F_1,F_2,F_3,\mathrm{\σ})=\underset{F_0}{min}\underset{i=1,2,3}{\Σ}\left|\right|{\hat{F}}_0\·{K_i}^{\hat{d}}-F_i|{|}^2+||C\·{\hat{F}}_0|{|}^2$

Wherein, the jamtosignal (NoisetoSignalRatios) of representative image priori, can be expressed as σ ²/ A, wherein, A can use the average power spectra of multiple natural images to represent, namely ξ represents frequency.Use energy function minimum value to mean to make the difference between ambiguous estimation image and original blurred picture minimum, namely make estimation procedure precision the highest, depth recovery result is the most accurate.

(4) use dimension receive warp area method obtain estimate picture rich in detail.Namely to a set estimating depth energy function is used first to solve estimation picture rich in detail namely exist condition under, can obtain,

${\hat{F}}_0=\frac{\underset{i}{\Σ}{F}_i\·{{\stackrel{\‾}{K}}_i}^{\hat{d}}}{\underset{i}{\Σ}{\left|{{\stackrel{\‾}{K}}_i}^{\hat{d}}\right|}^2+{\left|C\right|}^2},i=1,2,3$

Wherein, the complex conjugate of K, refer at set estimating depth the frequency domain representation of the imaging system point spread function that place is corresponding.

(5) bring the formula in step (4) and step (2) into step (3), reconstruct is readjusting and simplifying energy function also, and proof procedure is as follows:

(51) at given rotary coding aperture K ₁, K ₂, K ₃, standard depth d ^*, and under the condition of noise σ (being the additive white Gaussian noise that N, N represent in frequency domain in frequency domain), the energy function E corresponding to estimating depth d can be written as:

$E\left(d\right|K_1^{d^{*}},K_2^{d^{*}},K_3^{d^{*}},\mathrm{\σ})=\underset{F_0,N_1,N_2,N_3}{E}\[\underset{i=1,2,3}{\Σ}\left|\right|{\hat{F}}_0{K}_i-F_i|{|}^2+|\left|C{\hat{F}}_0\right|{|}^2\]$

(52) by the F in step (2) _iwith in step (4) bring in the energy function of (51), can obtain:

$\begin{array}{c}E\left(d\right|K_1^{d^{*}},K_2^{d^{*}},K_3^{d^{*}},\mathrm{\σ})=\underset{F_0,N_1,N_2,N_3}{E}\[\underset{i=1,2,3}{\Σ}\left|\right|({\hat{F}}_0{K}_i^{d^{*}}+N_i)-\\ \frac{({\hat{F}}_0{K}_1^{d^{*}}+N_1)\stackrel{\‾}{K_1^d}+({\hat{F}}_0{K}_2^{d^{*}}+N_2)\stackrel{\‾}{K_2^d}+({\hat{F}}_0{K}_3^{d^{*}}+N_3)\stackrel{\‾}{K_3^d}}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}{K}_i|{|}^2\\ +\left|\right|C\frac{({\hat{F}}_0{K}_1^{d^{*}}+N_1)\stackrel{\‾}{K_1^d}+({\hat{F}}_0{K}_2^{d^{*}}+N_2)\stackrel{\‾}{K_2^d}+({\hat{F}}_0{K}_3^{d^{*}}+N_3)\stackrel{\‾}{K_3^d}}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}|{|}^2\]\end{array}$

(53) under thinking that picture signal and noise signal are separate condition, can be by the formula arrangement in (52):

$\begin{array}{c}E\left(d\right|K_1^{d^{*}},K_2^{d^{*}},K_3^{d^{*}},\mathrm{\σ})=\underset{F_0,N_1,N_2,N_3}{E}\[\underset{i=1,2,3}{\Σ}\left|\right|F_0\[\frac{K_1^{d^{*}}\stackrel{\‾}{K_1^d}+K_2^{d^{*}}\stackrel{\‾}{K_2^d}+K_3^{d^{*}}\stackrel{\‾}{K_3^d}}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}-K_i^{d^{*}}\]\\ +\frac{(N_1\stackrel{\‾}{K_1^d}+N_2\stackrel{\‾}{K_2^d}+N_3\stackrel{\‾}{K_3^d})K_i}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}-N_i|{|}^2\\ +\left|\right|C\[\frac{F_0(K_1^{d^{*}}\stackrel{\‾}{K_1^d}+K_2^{d^{*}}\stackrel{\‾}{K_2^d}+K_3^{d^{*}}\stackrel{\‾}{K_3^d})}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}+\frac{N_1\stackrel{\‾}{K_1^d}+N_2\stackrel{\‾}{K_2^d}+N_3\stackrel{\‾}{K_3^d}}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}\]|{|}^2\]\end{array}$

(54) for ease of differentiating, order due to N _iseparate white Gaussian noise signal, therefore, E (N _i)=0, var (N _i)=σ ², E (N _in _j)=0.Herein, B is an intermediate variable, only for calculating derivation.Then energy function can be expressed as:

$\begin{array}{c}E\left(d\right|K_1^{d^{*}},K_2^{d^{*}},K_3^{d^{*}},\mathrm{\σ})=\\ \underset{F_0}{E}\[\underset{i=1,2.3}{\Σ}\left|\right|F_0\frac{K_1^{d^{*}}\stackrel{\‾}{K_1^d}+K_2^{d^{*}}\stackrel{\‾}{K_2^d}+K_3^{d^{*}}\stackrel{\‾}{K_3^d}-K_i^{d^{*}}B}{B}|{|}^2\]\\ +\underset{N_1,N_2,N_3}{E}\[\underset{i=1,2.3}{\Σ}\left|\right|\frac{(N_1\stackrel{\‾}{K_1^d}+N_2\stackrel{\‾}{K_2^d}+N_3\stackrel{\‾}{K_3^d})K_i}{B}-N_i|{|}^2\]\\ +\left|\right|C\frac{F_0(K_1^{d^{*}}\stackrel{\‾}{K_1^d}+K_2^{d^{*}}\stackrel{\‾}{K_2^d}+K_3^{d^{*}}\stackrel{\‾}{K_3^d})}{B}|{|}^2+|\left|C\frac{N_1\stackrel{\‾}{K_1^d}+N_2\stackrel{\‾}{K_2^d}+N_3\stackrel{\‾}{K_3^d}}{B}\right|{|}^2\end{array}$

(55) F is defined ₀power spectrum expect for A, meanwhile, c=σ ²/ A.Therefore, energy function can arrange and be:

$\begin{array}{c}E\left(d\right|K_1^{d^{*}},K_2^{d^{*}},K_3^{d^{*}},\mathrm{\σ})=\\ \underset{\mathrm{\ξ}}{\Σ}A\[\left|\right|\frac{\stackrel{\‾}{K_2^d}(K_1^d{K}_2^{d^{*}}-K_2^d{K}_1^{d^{*}})+\stackrel{\‾}{K_3^d}(K_1^d{K}_3^{d^{*}}-K_3^d{K}_1^{d^{*}})}{B}|{|}^2\\ +\left|\right|\frac{\stackrel{\‾}{K_1^d}(K_1^d{K}_2^{d^{*}}-K_2^d{K}_1^{d^{*}})+\stackrel{\‾}{K_3^d}(K_2^d{K}_3^{d^{*}}-K_3^d{K}_2^{d^{*}})}{B}|{|}^2\\ +\left|\right|\frac{\stackrel{\‾}{K_1^d}(K_1^d{K}_3^{d^{*}}-K_3^d{K}_1^{d^{*}})+\stackrel{\‾}{K_2^d}(K_2^d{K}_3^{d^{*}}-K_3^d{K}_2^{d^{*}})}{B}|{|}^2\]\\ +{\mathrm{\σ}}^2\·\underset{\mathrm{\ξ}}{\Σ}\[\frac{C^2}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}+1\]\end{array}$

(56) through arranging, final energy formula identity result can be obtained:

$\begin{array}{c}E\left(\hat{d}\right|{K_1}^{d^{*}},{K_2}^{d^{*}},{K_3}^{d^{*}},\mathrm{\σ})\\ =\underset{\mathrm{\ξ}}{\Σ}A\frac{{\left|K_1^d{K}_2^{d^{*}}-K_2^d{K}_1^{d^{*}}\right|}^2+{\left|K_1^d{K}_3^{d^{*}}-K_3^d{K}_1^{d^{*}}\right|}^2+{\left|K_2^d{K}_3^{d^{*}}-K_3^d{K}_2^{d^{*}}\right|}^2}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+C^2}\\ +{\mathrm{\σ}}^2\·\underset{\mathrm{\ξ}}{\Σ}\[\frac{C^2}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}+1\]\end{array}$

Wherein, d ^*represent the standard depth be positioned at corresponding to imaging system, d represents the degree of depth of visible images at each level place.

(6) at noise parameter be σ condition under, d is positioned to rotary coding aperture ^*the estimation of Depth at place is carried out performance evaluation and can be obtained,

$\begin{array}{c}R(K_1,K_2,K_3|d^{*},\mathrm{\σ})\\ =\underset{d\∈D/d^{*}}{\min }E\left(d\right|{K_1}^{d^{*}},{K_2}^{d^{*}},{K_3}^{d^{*}},\mathrm{\σ})-E\left(d^{*}\right|{K_1}^{d^{*}},{K_2}^{d^{*}},{K_3}^{d^{*}},\mathrm{\σ})\end{array}$

Wherein, D={c ₁d ^*, c ₂d ^*..., c _ld ^*represent each degree of depth level, { c _ican be taken as 0.1,0.15 ..., 1.5}, step-length is 0.05.

(7) above result is normalized to the Performance evaluation criterion that can obtain rotary coding aperture:

$\left\{\begin{array}{c}M(K_1,K_2,K_3,d,d^{*})\\ ={\[\frac{1}{n}\underset{\mathrm{\ξ}}{\Σ}\frac{{\left|K_1^d{K}_2^{d^{*}}-K_2^d{K}_1^{d^{*}}\right|}^2+{\left|K_1^d{K}_3^{d^{*}}-K_3^d{K}_1^{d^{*}}\right|}^2+{\left|K_2^d{K}_3^{d^{*}}-K_3^d{K}_2^{d^{*}}\right|}^2}{|K_1^d{|}^2+|K_2^d{|}^2+|K_3^d{|}^2+C^2}\]}^{\frac{1}{2}}\\ R=\underset{d\∈D/d^{*}}{min}M(K_1,K_2,K_3,d,d^{*})\end{array}\right.$

The larger energy function evaluating estimation of Depth that characterizes of R value changes more precipitous, makes estimation of Depth performance stronger for the robustness of picture noise or color texture.

2) design and use genetic algorithm optimization to obtain the rotary coding aperture of low resolution.For the aperture that resolution is L × L, its possibility reaches 2 ^{l × L}kind, calculated amount is very large, therefore cannot directly use high resolving power to be in optimized selection process.

3) coordinate descent is used to improve the resolution of rotary coding aperture.In order to reduce the impact of optical diffraction, continue the performance of coding aperture, use the performance of coordinate descent to aperture to optimize, be optimized rotary coding aperture again.

4) estimation of Depth is recovered and total focus image restoration process.After using rotary coding aperture imaging system to obtain visible images, the present invention receives deconvolution and aberration method by dimension and carries out the work of depth recovery and total focus image restoration to it.

Below in conjunction with accompanying drawing and instantiation, further describe the present invention.

1) Performance evaluation criterion proposing rotary coding aperture imaging system is applied to estimation of Depth and recovers and total focus image restoration:

(1) fuzzy imaging model out of focus:

$f=f_0\⊗k^d+\mathrm{\η}$

Wherein, f represents imaging system and takes the blurred picture out of focus obtained, f ₀represent original picture rich in detail, k ^drepresent the point spread function of imaging system at different depth d place, white Gaussian noise is described.

The imaging model out of focus of (2) three rotary coding aperture system:

F _i＝F ₀·K _i ^d+N _i,i＝1,2,3

Wherein, this model is model expression in a frequency domain in (1).F _irepresent that rotary coding aperture imaging system takes the blurred picture out of focus obtained, F ₀represent original image, K _i ^drepresent the point spread function at different depth d place, N _irepresent that i-th rotary coding aperture obtains the white Gaussian noise in shooting image, i represents number of revolutions, and namely three rotations can produce three different blurred pictures out of focus.This model corresponds to the mathematical model of three blurred pictures out of focus of the shooting gained of rotary coding aperture imaging system.

(3) minimum value of energy function E is used to solve maximum a posteriori probability problem (MaximumAPosteriori) to realize the performance optimization to rotary coding aperture.

$E\left(\hat{d}\right|F_1,F_2,F_3,\mathrm{\σ})=\underset{F_0}{min}\underset{i=1,2,3}{\Σ}\left|\right|{\hat{F}}_0\·{K_i}^{\hat{d}}-F_i|{|}^2+||C\·{\hat{F}}_0|{|}^2$

Wherein, the jamtosignal (NoisetoSignalRatios) of representative image priori, can be expressed as σ ²/ A, wherein, A can make the average power spectra of multiple natural images to represent, ξ represents frequency, μ (F ₀) represent the natural image of integration.Min represents and minimizes, ∑ represent add and, || * || represent and ask mould.Use energy function minimum value to mean to make the difference between ambiguous estimation image and original blurred picture minimum, namely make estimation procedure precision the highest, depth recovery result is the most accurate.

(4) use dimension receive warp area method obtain estimate picture rich in detail.Namely to a set estimating depth energy function is used first to solve estimation picture rich in detail namely exist condition under, can obtain,

${\hat{F}}_0=\frac{\underset{i}{\Σ}{F}_i\·{{\stackrel{\‾}{K}}_i}^{\hat{d}}}{\underset{i}{\Σ}{\left|{{\stackrel{\‾}{K}}_i}^{\hat{d}}\right|}^2+{\left|C\right|}^2},i=1,2,3$

Wherein, the complex conjugate of K, refer at set estimating depth the frequency domain representation of the imaging system point spread function that place is corresponding.

(5) bring the formula in step (4) and step (2) into step (3), reconstruct is readjusting and simplifying energy function also, and proof procedure is as follows:

(51) at given rotary coding aperture K ₁, K ₂, K ₃, standard depth d ^*, and under the condition of noise σ (being N in frequency domain), the energy function E corresponding to estimating depth d can be written as:

$E\left(d\right|K_1^{d^{*}},K_2^{d^{*}},K_3^{d^{*}},\mathrm{\σ})=\underset{F_0,N_1,N_2,N_3}{E}\[\underset{i=1,2,3}{\Σ}\left|\right|{\hat{F}}_0{K}_i-F_i|{|}^2+|\left|C{\hat{F}}_0\right|{|}^2\]$

Wherein, N ₁, N ₂, N ₃refer to the white Gaussian noise in the frequency domain corresponding respectively to the visible images that three shootings of rotary coding aperture system obtain.

(52) by the F in step (2) _iwith in step (4) bring in the energy function of (51), can obtain:

$\begin{array}{c}E\left(d\right|K_1^{d^{*}},K_2^{d^{*}},K_3^{d^{*}},\mathrm{\σ})=\underset{F_0,N_1,N_2,N_3}{E}\[\underset{i=1,2,3}{\Σ}\left|\right|({\hat{F}}_0{K}_i^{d^{*}}+N_i)-\\ \frac{({\hat{F}}_0{K}_1^{d^{*}}+N_1)\stackrel{\‾}{K_1^d}+({\hat{F}}_0{K}_2^{d^{*}}+N_2)\stackrel{\‾}{K_2^d}+({\hat{F}}_0{K}_3^{d^{*}}+N_3)\stackrel{\‾}{K_3^d}}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}{K}_i|{|}^2\\ +\left|\right|C\frac{({\hat{F}}_0{K}_1^{d^{*}}+N_1)\stackrel{\‾}{K_1^d}+({\hat{F}}_0{K}_2^{d^{*}}+N_2)\stackrel{\‾}{K_2^d}+({\hat{F}}_0{K}_3^{d^{*}}+N_3)\stackrel{\‾}{K_3^d}}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}|{|}^2\]\end{array}$

(53) under thinking that picture signal and noise signal are separate condition, can be by the formula arrangement in (52):

$\begin{array}{c}E\left(d\right|K_1^{d^{*}},K_2^{d^{*}},K_3^{d^{*}},\mathrm{\σ})=\underset{F_0,N_1,N_2,N_3}{E}\[\underset{i=1,2,3}{\Σ}\left|\right|F_0\[\frac{K_1^{d^{*}}\stackrel{\‾}{K_1^d}+K_2^{d^{*}}\stackrel{\‾}{K_2^d}+K_3^{d^{*}}\stackrel{\‾}{K_3^d}}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}-K_i^{d^{*}}\]\\ +\frac{(N_1\stackrel{\‾}{K_1^d}+N_2\stackrel{\‾}{K_2^d}+N_3\stackrel{\‾}{K_3^d})K_i}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}-N_i|{|}^2\\ +\left|\right|C\[\frac{F_0(K_1^{d^{*}}\stackrel{\‾}{K_1^d}+K_2^{d^{*}}\stackrel{\‾}{K_2^d}+K_3^{d^{*}}\stackrel{\‾}{K_3^d})}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}+\frac{N_1\stackrel{\‾}{K_1^d}+N_2\stackrel{\‾}{K_2^d}+N_3\stackrel{\‾}{K_3^d}}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}\]|{|}^2\]\end{array}$

(54) for ease of differentiating, order due to N _iseparate white Gaussian noise signal, therefore, E (N _i)=0, var (N _i)=σ ², E (N _in _j)=0.Herein, var means to ask for variance, then energy function can be expressed as:

$\begin{array}{c}E\left(d\right|K_1^{d^{*}},K_2^{d^{*}},K_3^{d^{*}},\mathrm{\σ})=\\ \underset{F_0}{E}\[\underset{i=1,2.3}{\Σ}\left|\right|F_0\frac{K_1^{d^{*}}\stackrel{\‾}{K_1^d}+K_2^{d^{*}}\stackrel{\‾}{K_2^d}+K_3^{d^{*}}\stackrel{\‾}{K_3^d}-K_i^{d^{*}}B}{B}|{|}^2\]\\ +\underset{N_1,N_2,N_3}{E}\[\underset{i=1,2.3}{\Σ}\left|\right|\frac{(N_1\stackrel{\‾}{K_1^d}+N_2\stackrel{\‾}{K_2^d}+N_3\stackrel{\‾}{K_3^d})K_i}{B}-N_i|{|}^2\]\\ +\left|\right|C\frac{F_0(K_1^{d^{*}}\stackrel{\‾}{K_1^d}+K_2^{d^{*}}\stackrel{\‾}{K_2^d}+K_3^{d^{*}}\stackrel{\‾}{K_3^d})}{B}|{|}^2+|\left|C\frac{N_1\stackrel{\‾}{K_1^d}+N_2\stackrel{\‾}{K_2^d}+N_3\stackrel{\‾}{K_3^d}}{B}\right|{|}^2\end{array}$

(55) F is defined ₀power spectrum expect for A, meanwhile, c=σ ²/ A.Therefore, energy function can arrange and be:

$\begin{array}{c}E\left(d\right|K_1^{d^{*}},K_2^{d^{*}},K_3^{d^{*}},\mathrm{\σ})=\\ \underset{\mathrm{\ξ}}{\Σ}A\[\left|\right|\frac{\stackrel{\‾}{K_2^d}(K_1^d{K}_2^{d^{*}}-K_2^d{K}_1^{d^{*}})+\stackrel{\‾}{K_3^d}(K_1^d{K}_3^{d^{*}}-K_3^d{K}_1^{d^{*}})}{B}|{|}^2\\ +\left|\right|\frac{\stackrel{\‾}{K_1^d}(K_1^d{K}_2^{d^{*}}-K_2^d{K}_1^{d^{*}})+\stackrel{\‾}{K_3^d}(K_2^d{K}_3^{d^{*}}-K_3^d{K}_2^{d^{*}})}{B}|{|}^2\\ +\left|\right|\frac{\stackrel{\‾}{K_1^d}(K_1^d{K}_3^{d^{*}}-K_3^d{K}_1^{d^{*}})+\stackrel{\‾}{K_2^d}(K_2^d{K}_3^{d^{*}}-K_3^d{K}_2^{d^{*}})}{B}|{|}^2\]\\ +{\mathrm{\σ}}^2\·\underset{\mathrm{\ξ}}{\Σ}\[\frac{C^2}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}+1\]\end{array}$

(56) through arranging, final energy formula identity result can be obtained:

$\begin{array}{c}E\left(\hat{d}\right|{K_1}^{d^{*}},{K_2}^{d^{*}},{K_3}^{d^{*}},\mathrm{\σ})\\ =\underset{\mathrm{\ξ}}{\Σ}A\frac{{\left|K_1^d{K}_2^{d^{*}}-K_2^d{K}_1^{d^{*}}\right|}^2+{\left|K_1^d{K}_3^{d^{*}}-K_3^d{K}_1^{d^{*}}\right|}^2+{\left|K_2^d{K}_3^{d^{*}}-K_3^d{K}_2^{d^{*}}\right|}^2}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+C^2}\\ +{\mathrm{\σ}}^2\·\underset{\mathrm{\ξ}}{\Σ}\[\frac{C^2}{{\left|K_1^d\right|}^2+{\left|K_2^d\right|}^2+{\left|K_3^d\right|}^2+{\left|C\right|}^2}+1\]\end{array}$

Wherein, d ^*represent the standard depth be positioned at corresponding to imaging system, d represents the degree of depth at each level place of visible images, represent in a frequency domain, when rotating corresponding to the * time, be positioned at the point spread function of # depth.

(6) at noise parameter be σ condition under, d is positioned to rotary coding aperture ^*the estimation of Depth at place is carried out performance evaluation and can be obtained,

$\begin{array}{c}R(K_1,K_2,K_3|d^{*},\mathrm{\σ})\\ =\underset{d\∈D/d^{*}}{\min }E\left(d\right|{K_1}^{d^{*}},{K_2}^{d^{*}},{K_3}^{d^{*}},\mathrm{\σ})-E\left(d^{*}\right|{K_1}^{d^{*}},{K_2}^{d^{*}},{K_3}^{d^{*}},\mathrm{\σ})\end{array}$

Wherein, D={c ₁d ^*, c ₂d ^*..., c _ld ^*each degree of depth level corresponding to representative system, { c _ione group of degree of depth level parameter, can be taken as 0.1,0.15 ..., 1.5}, step-length is 0.05, d ∈ D/d ^*represent and do not comprise standard depth fuzzy core d ^*degree of depth set.

(7) above result be normalized and simplify the Performance evaluation criterion that can obtain rotary coding aperture:

$\left\{\begin{array}{c}M(K_1,K_2,K_3,d,d^{*})\\ ={\[\frac{1}{n}\underset{\mathrm{\ξ}}{\Σ}\frac{{\left|K_1^d{K}_2^{d^{*}}-K_2^d{K}_1^{d^{*}}\right|}^2+{\left|K_1^d{K}_3^{d^{*}}-K_3^d{K}_1^{d^{*}}\right|}^2+{\left|K_2^d{K}_3^{d^{*}}-K_3^d{K}_2^{d^{*}}\right|}^2}{|K_1^d{|}^2+|K_2^d{|}^2+|K_3^d{|}^2+C^2}\]}^{\frac{1}{2}}\\ R=\underset{d\∈D/d^{*}}{min}M(K_1,K_2,K_3,d,d^{*})\end{array}\right.$

The larger energy function evaluating estimation of Depth that characterizes of R value changes more precipitous, makes estimation of Depth performance stronger for the robustness of picture noise or color texture.

2) design and use genetic algorithm optimization to obtain the rotary coding aperture of low resolution.For the aperture that resolution is L × L, its possibility reaches 2 ^{l × L}kind, calculated amount is very large, and high-resolution therefore cannot be used to be in optimized selection process.But meanwhile, high resolving power ensures aperture not by the key of optical diffraction jamming performance.For addressing this problem, the present invention promotes resolution to the mode of higher standard and carries out whole optimizing process after taking first to use low resolution.Genetic algorithm optimization step is as follows:

(1) algorithm correlation parameter input initialization:

D={c ₁d ^*, c ₂d ^*..., c _ld ^*, represent the depth layer level resolving power performance parameter of rotary coding aperture system, d ^*=7;

i=1,2,3, represent the initial value of rotary coding aperture three anglecs of rotation;

G=0, g represent the iterations of genetic algorithm, and primary iteration number of times is 0;

Produce the set of S=4000 initial random binary sequence (namely only including 0,1), the length of each sequence is U=L*L, L=11;

(2) iterative process (carries out loop iteration process from g=1:G, G=60; )

(21) select (the optimum option process in genetic algorithm, the natural selection process of simulation biological heredity process):

For each binary sequence b, reformed and be combined into the aperture matrix k that a size is L × L; Then by matrix k according to anglec of rotation θ _ibe the aperture matrix k rotating and obtain corresponding each anglec of rotation _i(i=1,2,3).Then, use rotary coding aperture Performance evaluation criterion to calculate, and select Z=400 wherein optimum result as the optimum solution set of this time iteration result.Finally, use the anglec of rotation to optimize angle set H to optimum solution set Z and carry out calculating again of rotary coding aperture Performance evaluation criterion, choose the anglec of rotation input θ of optimum anglec of rotation set as next iteration _i.Wherein, the composition anglec of rotation element of H is composed as follows: fixing first anglec of rotation is θ ₁=0 °; From 0 ° to 355 °, every 5 ° of intervals are got an angle value and are formed an angle set J; θ ₂, θ ₃two angle values are got for unduplicated from angle set J; Then, by various θ ₁, θ ₂, θ ₃angle combinations value add the anglec of rotation as selected angle element and optimize in angle set H.

(22) the following computation process of this step is repeated, until the quantity of arrangement set is restored to S from Z:

Intersect:

From choosing arbitrarily and copying two sequences in the optimal set Z step (21), then, after two sequence step-by-step alignment, with Probability p ₁=0.2 exchange carrying out place value, and obtain two new subsequences.

Variation:

For two new sequences that previous step obtains, with Probability p ₂=0.05 overturns the value of each of each new sequence.

(3) calculate the performance evaluation value of all sequences remained, and obtain the optimal result of coding aperture.

Through the iteration optimization of genetic algorithm, rotary coding aperture optimum results is as shown in the aperture of 11 × 11 in Fig. 1, and the optimization anglec of rotation result of its correspondence is: θ _i=0 °, and 110 °, 235 ° }, i=1,2,3

3) coordinate descent is used to improve the resolution of rotary coding aperture.In order to reduce the impact of optical diffraction, and continuing the performance of coding aperture, using the performance of coordinate descent to aperture to optimize again, after to be optimized rotary coding aperture.Algorithm steps is as follows:

(1) bicubic interpolation mode is used first rotary coding aperture to be progressively successively increased to high resolving power (resolution is increased to 13 × 13 from 11 × 11 by such as first time optimization) from low resolution

(2) coordinate descent is used to optimize rotary coding aperture performance:

(21) two-value upset is carried out along each element of level (x) and vertical (y) two dimension directions to encoded light cycle matrix, and to newly-generated rotary coding aperture in-service evaluation criterion calculation aperture performance number.Then, constantly repeat this process, until each element of traversal rotary coding aperture matrix, retain the result of wherein aperture performance number optimum.Namely iteration result only changes the element value of an encoded light cycle matrix each time.

(22) iterative process in step (21) is repeated, until the evaluation criterion index convergence of coding aperture, till aperture no longer changes.

(3) repeat step (1) (2), until with the raising of resolution, the evaluation criterion index convergence of rotary coding aperture, till coding aperture no longer occurs obviously to change.The resolution of coding aperture finally converges to high-resolution 47 × 47 by 11 × 11 raisings of low resolution.

Carry high-resolution rotary coding aperture optimum results as shown in Figure 2, corresponding to the process that rotary coding aperture optimization resolution progressively improves.

4) estimation of Depth is recovered and total focus image restoration process.After using rotary coding aperture imaging system to obtain visible images, the present invention receives deconvolution and aberration method by dimension and carries out the work of depth recovery and total focus image restoration to it.

(1) first use dimension to receive warp area method and obtain the estimation picture rich in detail of different depth level then use aberration function e (x, y) to calculate original blurred picture out of focus and estimate that each degree of depth level of picture rich in detail and corresponding each degree of depth level point spread function convolution estimates the residual result of blurred picture out of focus to each degree of depth level.That is:

$e(x,y)=\underset{d\∈D}{\arg \min }\underset{i=1,2,3}{\Σ}|f_i-IFT({\hat{F}}_0^d\·K_i^d)|$

Wherein, arg represents the neighborhood window of local, and IFT represents inverse Fourier transform.

Within the scope of degree of depth level, degree of depth level corresponding to the minimum value of capture difference function in each local neighborhood is the estimation of Depth information in this region.Then, reconstruct depth results is averaged in a very little local window, the depth map of the visible images of rotary coding aperture system shooting can be obtained.

(2) acquired depth map is used to restore total focus picture rich in detail.Seek for the point spread function of the corresponding degree of depth level of blurred picture each several part out of focus corresponding to estimating depth, then using dimension to receive warp area method can restore picture rich in detail.I.e. total focus image recuperation as follows:

${\hat{f}}_0=IFT\left({\hat{F}}_0^{e(x,y)}\right(x,y\left)\right)$

Test the estimation of Depth recovery of the visible images of this clearly demarcated rotary coding aperture system and the total focus image restoration impact of performance in experiment, and contrast with the coding aperture performance of pertinent literature, concrete outcome as shown in Figure 3, Figure 4.

Claims (4)

1. the depth estimation method based on rotary coding aperture imaging system, it is characterized in that, comprise the steps: the Performance evaluation criterion proposing rotary coding aperture imaging system, and use genetic algorithm and coordinate descent to optimize rotary coding aperture according to this evaluation criterion, after realize estimation of Depth and total focus image restoration work by the rotary coding aperture pattern optimized.

2. as claimed in claim 1 based on the depth estimation method of rotary coding aperture imaging system, it is characterized in that, the Performance evaluation criterion concrete steps proposing rotary coding aperture imaging system are:

(1) fuzzy imaging model out of focus:

$f=f_0\⊗k^d+\mathrm{\η}$

Wherein, f represents imaging system and takes the blurred picture out of focus obtained, f ₀represent original picture rich in detail, k ^drepresent the point spread function of imaging system at different depth d place, describe white Gaussian noise, X represents Gaussian distribution, represent the standard deviation of Gaussian noise;

The imaging model out of focus of (2) three rotary coding aperture system:

F _i＝F ₀·K _i ^d+N _i,i＝1,2,3

Wherein, this model is model expression in a frequency domain: F in (1) _irepresent that rotary coding aperture imaging system takes the blurred picture out of focus obtained, F ₀represent original image, K _i ^drepresent the point spread function of different depth, N _irepresent that i-th rotary coding aperture obtains the white Gaussian noise in shooting image, i represents number of revolutions, and namely three rotations can produce three different blurred pictures out of focus; This model corresponds to the mathematical model of three blurred pictures out of focus of the shooting gained of rotary coding aperture imaging system;

(3) energy function minimum value is used to solve maximum a posteriori probability problem (MaximumAPosteriori) to realize the performance optimization to rotary coding aperture.

$E\left(\hat{d}\right|F_1,F_2,F_3,\mathrm{\σ})=\underset{F_0}{min}\underset{i=1,2,3}{\Σ}\left|\right|{\hat{F}}_0\·{K_i}^{\hat{d}}-F_i|{|}^2+||C\·{\hat{F}}_0|{|}^2$

Wherein, the jamtosignal (NoisetoSignalRatios) of representative image priori, can be expressed as σ ²/ A, wherein, A uses the average power spectra of multiple natural images to represent, ξ represents frequency, and use energy function minimum value to mean to make the difference between ambiguous estimation image and original blurred picture minimum, namely make estimation procedure precision the highest, depth recovery result is the most accurate;

(4) use dimension receive warp area method obtain estimate picture rich in detail.Namely to a set estimating depth energy function is used first to solve estimation picture rich in detail namely exist condition under, can obtain,

${\hat{F}}_0=\frac{\underset{i}{\Σ}{F}_r\·{{\stackrel{\‾}{K}}_i}^{\hat{d}}}{\underset{i}{\Σ}|{K_i}^{\hat{d}}{|}^2+|C{|}^2},i=1,2,3$

Wherein, the complex conjugate of K, refer at set estimating depth the frequency domain representation of the imaging system point spread function that place is corresponding, the ∧ of the letter top in the present invention all represents that the representative variable of this letter is one and calculates estimated value;

(5) bring the formula in step (4) and step (2) into step (3), reconstruct energy function, obtains one about estimating depth through readjusting and simplifying heat-supplied function,

$\begin{array}{c}E\left(\hat{d}|{K_1}^{d^{*}},{K_2}^{d^{*}},{K_3}^{d^{*}},\mathrm{\σ}\right)\\ =\underset{\mathrm{\ξ}}{\Σ}A\frac{|K_1^d{K}_2^{d^{*}}-K_2^d{K}_1^{d^{*}}{|}^2+|K_1^d{K}_3^{d^{*}}-K_3^d{K}_1^{d^{*}}{|}^2+|K_2^d{K}_3^{d^{*}}-K_3^d{K}_2^{d^{*}}{|}^2}{|K_1^d{|}^2+|K_2^d{|}^2+|K_3^d{|}^2+C^2}\\ +{\mathrm{\σ}}^2\·\underset{\mathrm{\ξ}}{\Σ}\[\frac{C^2}{|K_1^d{|}^2+|K_2^d{|}^2+|K_3^d{|}^2+C^2}+1\]\end{array}$

Wherein, d ^*represent the standard depth be positioned at corresponding to imaging system, d represents place's degree of depth of each level of visible images;

(6) the additive white Gaussian noise parameter in spatial domain is under the condition of σ, is positioned at d to rotary coding aperture ^*the estimation of Depth at place is carried out performance evaluation and can be obtained,

$\begin{array}{c}R(K_1,K_2,K_3|d^{*},\mathrm{\σ})\\ =\underset{d\∈D/d^{*}}{\min }E\left(d|{K_1}^{d^{*}},{K_2}^{d^{*}},{K_3}^{d^{*}},\mathrm{\σ}\right)-E\left(d^{*}|{K_1}^{d^{*}},{K_2}^{d^{*}},{K_3}^{d^{*}},\mathrm{\σ}\right)\end{array}$

Wherein, D={c ₁d ^*, c ₂d ^*..., c _ld ^*each degree of depth level corresponding to representative system, { c _ican be taken as 0.1,0.15 ..., 1.5}, step-length is 0.05.

(7) above result is normalized to the Performance evaluation criterion that can obtain rotary coding aperture:

$\left\{\begin{array}{c}\begin{array}{c}M(K_1,K_2,K_3,d,d^{*})\\ ={\[\frac{1}{n}\underset{\mathrm{\ξ}}{\Σ}\frac{|K_1^d{K}_2^{d^{*}}-K_2^d{K}_1^{d^{*}}{|}^2+|K_1^d{K}_3^{d^{*}}-K_3^d{K}_1^{d^{*}}{|}^2+|K_2^d{K}_3^{d^{*}}-K_3^d{K}_2^{d^{*}}{|}^2}{|K_1^d{|}^2+|K_2^d{|}^2+|K_3^d{|}^2+C^2}\]}^{\frac{1}{2}}\end{array}\\ R=\underset{d\∈D/d^{*}}{\min }M\left(K_1,K_2,K_3,d,d^{*}\right)\end{array}\right.$

The larger energy function evaluating estimation of Depth that characterizes of R value changes more precipitous, makes estimation of Depth performance stronger for the robustness of picture noise or color texture.

3. as claimed in claim 1 based on the depth estimation method of rotary coding aperture imaging system, it is characterized in that, design and use genetic algorithm optimization to obtain the rotary coding aperture of low resolution; Coordinate descent is used to improve the resolution of rotary coding aperture.

4. as claimed in claim 1 based on the depth estimation method of rotary coding aperture imaging system, it is characterized in that, estimation of Depth is recovered and total focus image restoration process, is to receive deconvolution and aberration method and carry out the work of depth recovery and total focus image restoration by tieing up to it.