CN1560697A - Automatic quick focusing method on high exponent moment - Google Patents

Abstract

The invention provides an automatic quick focusing method for high order moment, the relation between image f(x, y) in the camera, the object g(x, y) and the point expanding function of the imaging system of the camera h(x, y) are expressed into formula 1, thus it can shoot three frames of image from three different positions, thus the focusing position of the digital camera can be determined according to the three frames of images.

Description

Automatically quick focus method based on High Order Moment

Technical field

The invention belongs to technical field of imaging, particularly a kind of automatically quick focus method based on High Order Moment.

Background technology

The principle of camera auto-focus is: clap when defocusing image clap when focusing on the contained high fdrequency component of image lack, in other words be that the high fdrequency component that focusedimage comprised is more than out-of-focus image, and it is big more to defocus degree, and the high fdrequency component that image comprised is few more so.So can judge whether image focuses on according to the contained high fdrequency component of image, in the middle of a series of images that is obtained in the focusing process, the image that high fdrequency component is maximum is a focusedimage, at this moment digital camera correctly focuses on.

The auto focusing method of traditional digital camera generally adopts mountain-climbing servo-actuated method, i.e. HCS (Hill-climbing search) algorithm.The HCS algorithm comprises three patterns: (1) search pattern; (2) calibration mode; (3) look around pattern.Search pattern determines the moving direction of initial camera lens and finds the maximal value of focusing operator; Calibration mode and the pattern of looking around are used for judging whether the focusing operator maximum of points that finds at search pattern is focus point, whether the focusing operator maximal value of promptly judging in search pattern to be found is global maximum, that is to say a process of eliminating local maximum.In these three patterns, can adopt some focusing operators to calculate optimal value in this pattern.Focusing operator commonly used has: Tenengrad operator, correction Laplce summation SML (Sum-Modified-Laplacian) operator, modulus deviation summation SMD (Sum-Modulus-Difference) operator, brightness change VAR (Variance) operator and FSWM (Frequency-Selective-Weighted-Median) filter operator.

The thought of Tenenbaum operator is: the gradient of each point of computed image, then simply with all gradient magnitude additions, that width of cloth image of gradient magnitude maximum is exactly a focusedimage.

The thought of SML operator is: the absolute value of the secondary deviation of each pixel of computed image is exactly a focusedimage with these absolute value summations and that maximum width of cloth image then.

The thought of SMD operator is: calculate the summation of the single order gray-scale deviation of the horizontal direction of two adjacent pixels and vertical direction, then with the summation addition of this both direction, that width of cloth image that obtain at last and maximum is exactly a focusedimage.

The thought of VAR operator is: the summation of the difference of each pixel gray scale of computed image and image averaging gray-scale value, with these and addition, that width of cloth image that obtain at last and maximum is exactly a focusedimage then.

The thought of FSWM filter operator is: to piece image along continuous straight runs and vertical direction filtering, obtain filtering result's quadratic sum, the quadratic sum maximum be exactly focusedimage.

The auto focusing method of this present practical application has several shortcomings: the one, because the optical imaging system of digital camera is a very complicated system, be subjected to light, camera lens is not normal, the influence of diffraction etc., the point spread function of each digital camera all is different in different practical applications, can't express with a very clear and definite expression formula, can only know digital camera by inference when being in defocus condition according to the principle of its optical imagery, the imaging system of digital camera is a low-pass filter, but the concrete parameter of this low-pass filter is unknowable, can't express with the expression formula of a clear and definite low-pass filter, can only suppose that in actual applications it is the 2-d gaussian filters device of a low pass.So in actual applications, can not find a kind of focusing operator can express the state of image completely, therefore utilize focusing operator to analyze in the self-focusing process that captured image carries out digital camera, the focusing operator value not only has a global maximum point, also has a lot of local maximums, its curve is not dull a rising or the dull smooth curve that descends, and the existence of local maximum has increased very big difficulty for the automatic focus of digital camera.Require than higher process because the automatic focus process of digital camera is real-time, general using HCS algorithm finds a maximum of points promptly to think focus point, thereby and can be in each position of camera all photographic images carry out global search.If do not carry out global search, can not find a kind of algorithm from a series of data, to find out global maximum thereby fully local maximum is filtered out with a lot of local maximums.Even so adopt new HCS optimizing searching algorithm can not overcome the problem of local maximum fully, therefore adopt existing auto focusing method to carry out in the automatic focus process of digital camera, inevitably mistake can be introduced, assurance can not be reached under any circumstance, can both correctly focus at every turn.

Second shortcoming of existing digital camera auto focusing method is: the method that the employing of digital camera auto focusing method is adjusted lens location according to the comparative result of focusing operator value realizes focusing on, if the position that focuses in the initial position of camera lens is distant, just need so on the one hand to take multiple image, can increase the time of Flame Image Process like this; On the other hand, when focal position is distant in lens location, it is bigger to defocus degree, the local maximum point of focusing operator value is a lot of so, when utilizing the calibration mode of HCS algorithm and looking around the process of pattern elimination local maximum, camera lens need move forward and backward photographic images, so also can increase Flame Image Process time, increased the probability of wrong focusing.Having increased the Flame Image Process time just is equivalent to increase the self-focusing time of digital camera, can realize self-focusing words in a short period of time if require digital camera, because existing auto focusing method will utilize a lot of two field pictures, so just increased requirement, increased logarithmic code camera processing chip and carry out the requirement of image-capable image capturing system.

Summary of the invention

The objective of the invention is to overcome the deficiency of existing focus method, provide a kind of new, be different from the higher accuracy of having of existing auto focusing method, shorter focal time fully, and the focus method fast automatically that has great practical value based on High Order Moment.

In order to reach above-mentioned goal of the invention, the technical solution used in the present invention is as follows:

The present invention adopts the 2-d gaussian filters device $h(x,y)=\frac{1}{{2\mathrm{\π\σ}}^2}\exp (-\frac{x^2+y^2}{{2\mathrm{\σ}}^2})$ The point spread function of representing the imaging system of camera, so according to the characteristic of High Order Moment, can be with magazine image f (x, y) with the own g (x of object that is taken, y) and the point spread function h of the imaging system of digital camera (x, between y) concern f (x, y)=g (x, y) (x, y) derivation draws second order centre distance to * h

$u_{\mathrm{pq}}^{\left(f\right)}=\underset{k=0}{\overset{p}{\mathrm{\Σ}}}\underset{l=0}{\overset{q}{\mathrm{\Σ}}}\left[\begin{array}{c}p\\ k\end{array}\right]\left[\begin{array}{c}q\\ l\end{array}\right]u_{k,l}^{\left(h\right)}{u}_{p-k,q-l}^{\left(g\right)}$

So, by take three secondary different images at diverse location, just can determine the position of focusing,

$v_0=\frac{(v_3+v_2)u_{20}^{\left(f_1\right)}+(v_1+v_3)u_{20}^{\left(f_2\right)}+(v_2+v_1)u_{20}^{\left(f_3\right)}}{u_{20}^{\left(f_1\right)}+u_{20}^{\left(f_2\right)}+u_{20}^{\left(f_3\right)}}$

Wherein,

v ₀For digital camera lens is in when focusing on camera lens apart from the distance of imaging plane,

v ₁Camera lens is apart from the distance of imaging plane when being in position 1 for digital camera lens,

v ₂Camera lens is apart from the distance of imaging plane when being in position 2 for digital camera lens,

v ₃Camera lens is apart from the distance of imaging plane when being in position 3 for digital camera lens,

u ₂₀ ^(f1)For digital camera lens is in the second order centre distance of position 1 captured image,

u ₂₀ ^(f2)For digital camera lens is in the second order centre distance of position 2 captured images,

u ₂₀ ^(f3)Be in the second order centre distance of position 3 captured images for digital camera lens.

The invention has the beneficial effects as follows: have higher focusing accuracy than conventional auto focusing method, and focus on required time also still less.

Embodiment

The present invention is described further below by specific embodiment.

Respectively five kinds of different material objects are focused on by five kinds of different focus methods and to contrast, thereby a kind of superiority of focus method fast automatically based on High Order Moment provided by the invention is described.First kind is conventional HCS algorithm; Calculate the position of focus point by provided by the invention based on the automatically fast focus method of High Order Moment for second kind, use the search pattern of FSWM operator, HCS algorithm then; The third calculates the position of focus point by provided by the invention based on the automatically fast focus method of High Order Moment, uses three patterns of FSWM operator, HCS algorithm then; Calculate the position of focus point by provided by the invention based on the automatically fast focus method of High Order Moment for the 4th kind, use the search pattern of VAR operator, HCS algorithm then; Calculate the position of focus point by provided by the invention based on the automatically fast focus method of High Order Moment for the 5th kind, use three patterns of VAR operator, HCS algorithm then.

The result of five kinds of focus methods contrasts as shown in the table:

The analysis of the accuracy of the different auto focusing methods of table 1

Image sequence	The focal position	Revise standard deviation MSD (modified standard deviations)
							Algorithm 1	Algorithm 2	Algorithm 3	Algorithm 4	Algorithm 5
		??Books	??28	?2.02	?0.78	?0.78						?1.20	?1.12
??Bag	??34						?2.56	?0.76	?0.57	?1.08	?0.98
		??Flowers	??36	?2.03	?0.54	?0.46						?1.14	?0.85
??People	??41						?2.38	?1.09	?0.81	?1.46	?1.23
		??Lamp	??21	?1.89	?0.82	?0.69						?0.98	?0.76

The analysis of the different auto focusing method required times of table 2

Image sequence	The focal position	Focus on required time (unit: second)
							Algorithm 1	Algorithm 2	Algorithm 3	Algorithm 4	Algorithm 5
		??Books	?28	?0.43	?0.29	?0.33						?0.18	?0.25
??Bag	?34						?0.65	?0.31	?0.38	?0.26	?0.34
		??Flowers	?30	?0.58	?0.36	?0.42						?0.35	?0.40
??People	?41						?0.51	?0.32	?0.41	?0.27	?0.41
		??Lamp	?22	?0.62	?0.36	?0.39						?0.31	?0.38

Revise being calculated as follows of standard deviation MSD in the table 1:

$\mathrm{MSD}=\sqrt{\frac{{\mathrm{\Σ}}_{i=1}^N(m_p-m_a)}{N}}$

Claims (3)

1, a kind of automatically quick focus method based on High Order Moment, image f (x, y) with the g (x of object own, y) and the point spread function h of the imaging system of camera (x, y) pass between be f (x, y)=g (x, y) * h (x, y), it is characterized in that described focus method by the different image of three frames at different position shot object itself, described three two field pictures have been determined the position that focuses on; Concrete focusing formula is:

$v_0=\frac{(v_3+v_2)u_{20}^{\left(f_1\right)}+(v_1+v_3)u_{20}^{\left(f_2\right)}+(v_2+v_1)u_{20}^{\left(f_3\right)}}{u_{20}^{\left(f_1\right)}+u_{20}^{\left(f_2\right)}+u_{20}^{\left(f_3\right)}}$

Wherein, v ₀Camera lens is apart from the distance of imaging plane, v when being in focusing for camera lens ₁Camera lens is apart from the distance of imaging plane, v when being in position 1 for camera lens ₂Camera lens is apart from the distance of imaging plane, v when being in position 2 for camera lens ₃Camera lens is apart from the distance of imaging plane when being in position 3 for camera lens, For camera lens is in the second order centre distance of position 1 captured image,

For camera lens is in the second order centre distance of position 2 captured images, Be in the second order centre distance of position 3 captured images for camera lens.

2, the automatically fast focus method based on High Order Moment according to claim 1 is characterized in that (x y) adopts the 2-d gaussian filters device for the point spread function h of the imaging system of described camera $h(x,y)=\frac{1}{2{\mathrm{\π\σ}}^2}\exp (-\frac{x^2+y^2}{{2\mathrm{\σ}}^2})$ Represent that wherein σ is a coefficient of diffusion.

3, the automatically quick focus method based on High Order Moment according to claim 1 and 2 is characterized in that described second order centre distance is

$u_{\mathrm{pq}}^{\left(f\right)}=\underset{k=0}{\overset{p}{\mathrm{\Σ}}}\underset{l=0}{\overset{q}{\mathrm{\Σ}}}\left[\begin{array}{c}p\\ k\end{array}\right]\left[\begin{array}{c}q\\ l\end{array}\right]u_{k,l}^{\left(h\right)}{u}_{p-k,q-l}^{\left(g\right)}.$