JPH04348306A - Focus detector - Google Patents

Abstract

PURPOSE:To perform a deviation arithmetic operation in different conditions and to obtain superior responsiveness without prolonging the arithmetic time. CONSTITUTION:This focus detector is equipped with a microcomputer 8 which has a prefilter means 5, a memory 9, a postfilter 10, deviation arithmetic operation 11, etc., and the deviation amount between a couple of object images is calculated by shifting the image outputs of the object images within a 1st specific shift quantity range in a 1st condition to detect the quantity of relative deviation between the couple of object images, and the deviation between the couple of object images is calculated by shifting the image outputs of the object images within a 2nd specific shift quantity range in a 2nd condition so as to detect the quantity of relative deviation between the couple of object images. The detection range of the relative images which is determined by the 1st specific shift quantity in the 1st condition is set wider than the detection range of the relative deviation between the object images which is determined by the 2nd specific shift quantity in the 2nd condition, and a defocusing quantity is calculated from the processing result of at least one of 1st and 2nd processing means.

Description

[Detailed description of the invention]

0001

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a focus detection device for detecting the focus adjustment state of an imaging optical system, and more particularly to a focus detection device for a camera.

0002

2. Description of the Related Art The focusing state of an imaging optical system or the distance to the target object is detected by forming optical images of the target object on respective integrated photoelectric element arrays and calculating the photoelectric outputs of those arrays. Various focus detection devices have been proposed. For example, light fluxes that have passed through different areas of the pupil of the imaging lens are guided onto a single image sensor, and the relative shift between the images of the target object on both image sensors is photoelectrically detected to determine the focusing state of the imaging lens. Focus detection devices for detecting focus have been widely put into practical use.

Regarding such a focus detection device that photoelectrically detects the relative shift amount of a pair of object images on an image sensor to detect the focus adjustment state of an imaging optical system, the applicant has proposed a method disclosed in Japanese Patent Application Laid-Open No. In Publication No. 37513, the image output of the object image is relatively shifted within a predetermined pixel number range to calculate the amount of correlation between the two images, and the relative amount of the two images is calculated based on the multiple correlation amounts thus calculated. We have proposed a shift calculation method that calculates the shift amount with an accuracy of one shift amount or less.

0004

However, different performance is required depending on the focusing state of the imaging optical system. For example, if the photographic lens is far away from the in-focus position, the image of the pair of objects on the image sensor will be relatively distorted.
Many low-order spatial frequency components are included, and it is necessary to detect a large amount of deviation from such a pair of object images. On the other hand, it has been found that when the photographic lens is near the in-focus position, it is preferable to detect the amount of deviation with high precision from a pair of object images containing relatively many high-order spatial frequency components.

[0005] Furthermore, it has been found that it is difficult to cope with different performance requirements depending on the focus adjustment state using a single shift calculation. The present invention was made in view of the fact that it is difficult to respond to the different performances required depending on the focus adjustment state with such a single shift calculation, and it is possible to perform shift calculations under different conditions. It is an object of the present invention to provide a focus detection device with excellent responsiveness that does not require an increase in calculation time.

[0006]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be explained below with reference to the drawings. In FIG. 1, the photoelectric conversion device 1 is a CCD type or MO
a pair of photoelectric element arrays 2 consisting of an S-type image sensor;
3, and on these arrays 2 and 3, an optical image of the object is formed by a pair of light beams that have passed through different areas of the pupil of the photographing lens of the camera. The amount of relative deviation between the light image on array 2 and the light image on array 3 represents the focus adjustment state of the lens. The above arrays 2 and 3 each have a spatial pitch p0
Photoelectric elements a1 '...an' arranged on a straight line with
and b1'...bn', and each photoelectric element a1'...an', b1'...bn' has primary data a1...an, b with a spatial pitch p0, respectively.
1 ...bn is output in time series. The data related to array 2 and the data related to array 3 are shown below as series a and series b, respectively.
This is called series data.

The MTF characteristics of each photoelectric element array 2 and 3 are as follows:
As shown in FIG. 2(a), D. It is sufficiently large near the C (DC) component, then decreases as the frequency increases, and becomes zero near the frequency of 1/p0. Therefore, the primary data a1...an, b1...bn are D. of the spatial frequency components of the optical image. It contains enough frequency components near the C component and almost no frequency components near 1/p0.

Primary data ai, b from the photoelectric conversion device 1
i is, for example, logarithmically amplified in the nonlinear processing section 4 and then sent to the prefilter means 5. This prefilter means 5 is implemented by hardware, specifically as shown in FIG. 3(a) or (b).
It consists of a transpersal filter shown in the figure below. The filters in FIGS. 3(a) and 3(b) include series-connected -data delay circuits D1, D2, D3... and predetermined weighting coefficients w1, w2, w3, w4, w5, respectively.
Multipliers w1, w2,...w5 that multiply input data by
and an adder T1 that adds the outputs of these multipliers, and in FIG. 3(a), the primary data is a1, b1.
, a2, b2...an, bn are input in the order shown in FIG. 3(b).
1, b2...bn are input in this order. In this way, the transpersal filter uses the primary data a1, a2...an of the a series input sequentially.
Predetermined weighting coefficients w1 = 0.28, w2 = 0.76 as shown in Fig. 2 (b-2) are applied to the five consecutive pieces of data.
, w3 = 1, w4 = 0.76, and w5 = 0.28, respectively, and after adding these, the sum is output, and the primary data b1, b2 . . . bn of the b series is processed in exactly the same way. A filtering process and a filter that multiply a plurality of data by a predetermined weighting coefficient and add them are respectively referred to as a weighted addition and a weighted addition filter. The above weighting coefficient sequence (0.28, 0.76, 1, 0.76, 0.
The MTF characteristics of the weighted addition filter with 28) are shown in Figure 2.
As shown in (b-1), D. Extract C component sufficiently and 1/
This characteristic removes frequency components around 4p0 or above, in other words, spatial frequency components above Nyquist frequency fN=1/2p determined by sampling pitch p=2p0. Of course, a weighted addition filter having such an MTF characteristic can be achieved not only by the above-mentioned weighting coefficient sequence but also by other weighting coefficient sequences.

The data filtered by the prefilter means 5 is sampled and held by the sample hold means 6 at a sampling pitch p=2p0. In this way, the a-series and b-series primary data ai, bi with a spatial pitch p0 are converted into a-series secondary data Ai and b-series secondary data Bi with a spatial pitch 2p0 through the prefilter means and sample hold means. Since the secondary data Ai and Bi have a pitch of 2p0, their number is about half of the primary data ai and bi. The secondary data Ai, Bi can be expressed by the following formula.

0010

[Math 1]

After these secondary data Ai and Bi are A/D converted by an A/D converter 7, they are sent to a microcomputer (
Hereinafter, it will be referred to as a microcomputer. ) 8 and its memory section 9
be accommodated in. The CPU of the microcomputer 8 performs functions such as filtering processing and shift calculation. In order to visualize these functions, here, for convenience, inside block 8,
Post-filter unit 10 that performs a weighted addition filtering function
and a shift calculation section 11 that performs calculation functions are each made into blocks. The post-filter unit 10 inputs the secondary data Ai, Bi in the memory unit 9 with a spatial pitch 4p shown in FIG. 2(c-3).
0 weighting coefficient sequence (-0.25, 1, -0.25) or the spatial pitch 4p0 weighting coefficient sequence (-0.5, 1, -0.5) shown in Fig. 2 (d-3). Perform weighted addition processing to create tertiary data. Specifically, secondary data A
Since i and Bi have a spatial pitch of p=2p0, the post-filter section filters the continuous five secondary data as shown in Fig. 2(c-
3) or (d-3) as (-0.25, 0, 1, 0,
-0.25), or (-0.5, 0, 1, 0, -0.5
) are multiplied by the weighting coefficient sequence and added.

[0012] This tertiary data has been subjected to weighted addition filtering processing by both the pre-filter means 5 and the post-filter section 10. Figure 2 (b-2) and Figure 2 (
c-3) Synthesis MTF of filters each having a weighting coefficient sequence
is shown in Fig. 2 (c-1), and similarly Fig. 2 (b-2) and (d
The composite MTF of the filter of -3) is shown in FIG. 2(d-1).

The deviation calculation unit 11 calculates the relative deviation amount of the optical images on the arrays 2 and 3 from the secondary data of the a series and the b sequence, or from the tertiary data of the a series and the b series. Next, the relationship between spatial frequency components and the accuracy of calculation results will be described. In general, when calculating the amount of deviation by calculating the data of the sampling pitch p, spatial frequency components higher than the Nyquist frequency fN = 1/2p determined by the sampling pitch p are harmful and cause false detection; Very effective and highly accurate calculation results can be expected for spatial frequency components around fN /2, and the frequency is around this frequency fN.
The accuracy decreases as the value becomes smaller than /2. On the other hand, if the photographing lens is greatly deviated from the in-focus position, the optical image on the photoelectric element array will be greatly blurred, and this optical image will have significantly reduced high-order spatial frequency components. Components near the C component increase.

The operation of the device of FIG. 1 is as follows. Spatial pitch p0 chronologically output from photoelectric conversion means 1
The primary data ai, bi sequentially pass through the nonlinearization means 4 and the prefilter means 5, and then are sampled by the sample and hold means 6 at a pitch p=2p0 to become secondary data Ai, Bi with a spatial pitch of 2p0. After the secondary data Ai and Bi are A/D converted, the microcomputer 8
is input.

As shown in FIG. 4, the microcomputer 8 processes the input secondary data Ai, Bi in step [1].
The secondary data Ai and Bi stored in the memory area (1) and stored in step [2] are processed to calculate a value Lm related to the amount of deviation of the optical image. Secondary data Ai, Bi
has already undergone filtering processing by the prefilter means 5 of the MTF characteristic shown in FIG.
All harmful spatial frequency components of N = 1/4p0 or more have been removed, and D. Since there is a possibility that the C component is sufficiently contained, even if the amount of deviation is large and the optical image is blurred, the deviation can be detected with a certain degree of accuracy.

In step [4], it is determined whether the calculation result Lm indicates that the photographing lens is near the in-focus position, that is, whether Lm is less than or equal to a predetermined value. In the case of N0, immediately proceed to step [8] and calculate the above calculation result L.
The amount of defocus, that is, the amount of deviation between the imaging position of the photographic lens and the planned focal point plane (film plane) is converted from m. On the other hand, if Yes in step [4], fN is present in the optical image.
There is a high possibility that a high-order spatial frequency component around Step [5] to perform deviation calculation by component
Move to.

In this step [5], the memory area (1
) to the secondary data Ai, Bi in Figure 2 (c-3) or (d-
The weighted addition process using the weighted coefficient sequence of 3) is performed to create tertiary data and store it in the memory area (2). Since this tertiary data has been subjected to filtering processing using the MTF characteristics shown in FIG. 2(c-1) or (d-1), D. The C component is considerably or completely removed, fN /2 (=1/8p0)
is expected to contain sufficient high-order spatial frequency components. In step [6], memory area (2) is specified as the memory area to be used, and in step [7], memory area (2) is specified as the memory area to be used.
2) Performs deviation calculation processing based on the tertiary data from 2). In step [8], the calculation result Lm is converted into a defocus amount.

In this way, when the photographing lens is largely deviated from the in-focus position, the deviation is detected using secondary data that contains a relatively large number of relatively low-order spatial frequency components, and In this case, highly accurate deviation detection is performed using tertiary data containing relatively many high-order spatial frequency components. According to the defocus amount thus determined, the display means 12 and the focus adjustment means 13 display the focus adjustment state and perform focus adjustment, respectively.

Furthermore, the memory area (2) is the memory area (1).
) can also be used. In addition, if there is sufficient memory capacity and separate areas can be used as memory areas (1) and (2), and if A/D conversion takes time and there is excess CPU capacity, the filter in step [5] The processing can also be performed simultaneously with the storage in step [1]. The properties of the pre-filter means and the post-filter means will be explained.

The combined MTF characteristics of the pre-filter having the weighting coefficient sequence shown in FIG. 2(b-2) and the post-filter having the weighting coefficient sequence shown in FIG. 2(c-3) or (d-3) are shown in FIG. 2(c-1
) or (d-1), and this Figure 2 (c-1
) or (d-1) from a single weighted addition filter, the weighting coefficient sequence of this single weighted addition filter will be as shown in FIG. 2 (c-2) or (d-2). Here, the total number of weighting coefficients of the pre-filter and post-filter is 8, and the number of weighting coefficients of the single filter is 13. Similarly, for example, FIGS. 2(b-2), (e-
A combination filter in which three weighted addition filters having weighted coefficient sequences of 3) and (f-3) are combined in series is shown in FIG.
It exhibits the composite MTF characteristic shown in f-1), and this MTF characteristic can also be obtained from a single weighted addition filter having the weighting coefficient sequence shown in FIG. The total number of weighting coefficients of the single filter is 11, and the number of weighting coefficients of the single filter is 25. Furthermore, Figure 2 (e-3
) and (f-3), respectively, and the MTF characteristics of a single filter having the weighting coefficient sequence of FIG. 2(f-4) are the same, and the combined filter The total number of weighting coefficients is 6, and that of the single filter is 9.

In this way, in general, the same MTF characteristic as that of a single weighted addition filter can be obtained from a combination filter in which a plurality of weighted addition filters are combined in series. The sum of the weighting coefficients can be significantly reduced than the number of weighting coefficients of a single weighted addition filter. This reduction in the number can be expected to simplify the structure if the weighted addition filter is configured by hardware, and can be expected to shorten the processing time if the weighted addition is performed by a microcomputer.

Furthermore, as will be described later, by preparing a plurality of relatively simple weighting coefficient sequences as post-filters, selecting them appropriately and using them in combination with a pre-filter, it is possible to easily create a plurality of combined filters. Obtainable. As described above, providing the prefilter with an MTF characteristic that removes spatial frequency components higher than the Nyquist frequency determined from the spatial pitch of data used for shift calculation has the following advantages. That is, since the output of the pre-filter has been removed from components that would be detrimental to the displacement calculation, it is advantageous that the output of the pre-filter can be used in addition to the output of the post-filter for the displacement calculation.

[0023] In addition, when the pre-filter is configured with hardware and a plurality of filters are prepared as post-filter and their filtering is processed by software, the output of the pre-filter is stored in memory and the There is an advantage that the data can be sequentially used for multiple post-filtering processes. Next, the deviation calculation section will be explained.

[0024] Each memory stores N pieces of a-series data A1.
...AN, b-series data B1...BN are stored. While shifting the a series data to the b series data by a predetermined number of data L (integer), the following correlation amount C (
Calculate L).

[0025]

[Math 2]

Here, q and r are given by the following equations, for example.

[0027]

[Math 3]

S is a constant integer representing the number of addition terms [X
] represents the largest integer not exceeding X. Figure 5(a) is a plot of the correlation amount C(L) when the shift amount L is sequentially changed. The cyst amount Lm corresponds to the amount of structural deviation.

The shape of the curve F near the minimum point Fm is shown in FIG.
) As a result of experiments, it was found that the shape is approximately symmetrical V-shaped around the maximum value Fm. A method of interpolating the shift amount Lm that gives the minimum value Fm based on this will be described below. Shift amount L-1 increased by one data,
Correlation amounts C(L-1), C(L), C for L, L+1
Let (L+1) be C-1, C0, and C1, respectively. As shown in FIG. 5(b), C-1≧C0 and C1 >C0
When is satisfied, a local minimum exists between C-1 and C1. In order to find this minimum value by interpolation, the maximum C1 and minimum C0 of the above three correlation amounts are drawn by a straight line l
1, and draw a straight line l-1 having a slope equal in absolute value but opposite in sign to the slope of this straight line l1 so as to pass through the intermediate value C-1. The intersection of both straight lines l1 and l-1 becomes the minimum value Cext of the function F. C0 in the coordinate axis C(L) direction
The distance DL between and Cext is as follows.

DL=0.5×(C-1-C1) Therefore, the minimum value Cext is C0-|DL|. The interpolation shift amount Lm that gives this minimum value is as follows. Lm=L+DL /E where E=MA×{C1 -C0, C-1-C0}
MA×{Ca, Cb} means selecting the greater of Ca and Cb. Note that the magnitude of the correlation amount C(L) varies depending largely on the pattern of the subject image, but it is standardized so that the minimum value does not depend on the pattern of the subject image. For this purpose, using the above value E as the normalization factor, the value obtained by dividing C0 - |DL | by E is calculated as the normalized minimum value C
ext'. When the minimum value is normalized so that it does not depend on the subject image, the function F
The minimum value Fm is a value close to zero that is almost unrelated to the subject image, and the other minimum values Fe are considerably large values in comparison.

Therefore, the normalized minimum value Cext' and the reference value C
ref and when Cext'<Cref,
The interpolated shift amount that gives the minimum value represents the amount of deviation of the optical image. Note that the reference value Cref is selected to be an intermediate value between the above correlation amount Fm and another minimum value Fe. An algorithm for achieving the shift calculation described above will be explained using the flowchart of FIG. 6.

In FIG. 6, the microcomputer sets the shift amount L to zero in step [11], and sets the information amount parameter Inform, which will be described later, to zero. Step [1
2], when the shift amounts are L-1, L, and L+1, the correlation amounts C(L-1), C(L), and C(L+1) are calculated based on formula (1), and the memory C-1, Store in C0 and C1 respectively. In step [13] MA×{C1, C0
, C-1} > Cth, after setting Info to 1, in step [14] the conditions C1 > C0 and C-1
It is determined whether ≧C0 is satisfied, and when it is satisfied, it is determined that there is an extreme value between the content of C-1 and the content of C1. When the above conditions are met, in step [15] memory C-1,
Values DL, E, and Cext' are calculated from the contents of C0 and C1. In step [16], the normalized minimum value Cext'
and the reference value Cref, and the condition Cext'<Cre
It is determined whether f is satisfied. When this condition is satisfied, it is determined that the minimum value Cext at this time is the minimum value of the correlation function F in FIG. 5, and the interpolated cyst amount Lm=(L+DL/E) is calculated and stored in step [17]. At the same time, a parameter Corr is set to indicate that the shift amount Lm is within a predetermined maximum amount to be shifted.
Set el to 1.

In step [18], the parameter E is compared with a predetermined threshold value Eth. This comparison is made for the following reasons. Generally, when the data Ai and Bi do not contain enough effective spatial frequency components, the above step [
The reliability of the calculation results in [15] and [17] decreases. As the effective spatial frequency component decreases, the parameter E
will also become smaller. Therefore, the reliability of the calculation result can be determined from the magnitude of this parameter E.

When E>Eth, in step [19], the parameter Inform is set to 1, assuming that the calculation result is reliable. Conversely, when E>Eth is not the case, step [
20], set the parameter Info to 0. On the other hand, if step [14] or [16] is not satisfied, the sign of L is inverted in step [21]. Since the shift amount L is now set to zero, L is also zero.

Next, in step [22], it is determined whether the cyst amount L satisfies the condition L≧0. If the conditions are met, L is increased by 1 in step [23]. In this case, since L=0 and the condition of step [22] is satisfied, L=1 is set after step [23]. In step [24], this newly set shift amount L is compared with a predetermined shift amount lf. This predetermined shift amount lf is an integer value that determines the maximum shift amount. Step [22
] or when the conditions in step [24] are not satisfied, the process returns to step [12]. After returning from this step [12], when you come to step [21] again, since L=1, the sign is reversed and L=-1, and step [22]
The process returns to step [12] again via . In this way, for example, when lf = 5, the shift amount L is sequentially changed from 0 to 1, -1, 2, -2, 3, - until the maximum correlation is obtained.
The shift amount is gradually increased to 3, 4, and -4 to find the shift amount that gives the maximum correlation. When the shift amount L reaches the predetermined shift amount lf without obtaining the maximum correlation, the parameter Correl is set to indicate that the maximum correlation cannot be obtained even if the data is shifted to the maximum shift amount in step [25]. Set to 0 to end the algorithm routine. Further, step [13] is to determine whether or not the determination in the case of ending with Correl=0 has been made based on a sufficient amount of information.

In the above correlation calculation, for example, the correlation amounts C(L) and C(L+1) when L=0 are equal to the correlation amounts C(L-1) and C(L) when L=1, respectively. The correlation amounts C(L-1) and C(L) when L=0 are equal to the correlation amounts C(L) and C(L+1) when L=-1, respectively. Similarly, C(L) and C(L+1) when L=1 are respectively C(L+1) when L=2.
-1), equal to C(L), and C(L-
1), C(L) are C(L), C(L+1) when L=-2
)be equivalent to. Despite this relationship, in the flowchart of FIG. 6, when L changes, the three correlation quantities C(L-1)
, C(L), and C(L+1) were all newly calculated. Since the computation of C(L) takes a considerable amount of time, this means an increase in computation time. Therefore, in an actual program, it is appropriate to store the previous calculated value except when L=0 and calculate only the newly required correlation amount. In that case, step [13] is the correlation amount C [
C(L')>Ct every time L'] is newly calculated.
It can be replaced by determining whether or not h. The algorithm for this shift calculation starts from L=0, L=1, -1, 2,
-2..., and if a minimum value is found that satisfies the predetermined condition, it can be determined that this is the maximum correlation even if the calculation of the entire range of -lf ≦L≦lf has not been completed, and the actual autofocus drive During this process, the true focal point is often near L = 0, so it is only necessary to calculate equation (1) for several cyst amounts, which can significantly shorten the calculation time. . Also, if the predetermined shift amount lf is selected to be about 3, the number of calculations of equation (1) will be only 7 times, and when the number of data N = 50 and the number of addition terms to calculate equation (1) is S = 30. In this case, only 210 calculations of Ai-Bi+L are required.

Embodiments of the present invention will be explained based on the above background explanation. A process that calculates the deviation of two images in a large shift range (first process) under conditions that exhibit performance in areas with large defocus (first condition), and a condition that exhibits performance near in-focus (second condition). The process of calculating the deviation of the two images within a small shift range (second process) can be performed first, but if the second process is performed first and the first process is performed only when detection is not possible, the photographing lens will Since it is important that the object is near the in-focus state, in most cases only the second processing is required, which has the advantage of requiring less calculation time.

In such a case, the first
An example will be explained. Spatial pitch p0 of a series and b series
From the primary data ai, bi, the respective spatial pitches 2p0
The steps up to creating the secondary data Ai, Bi are the same as in the first embodiment. A/D converted secondary data Ai, Bi
is stored in the memory area (1) in step [31]. In step [32], the data in the memory area (1) is subjected to weighted addition filtering using the weighting coefficient sequence shown in FIG. 2 (c-3) or (d-3). Create tertiary data with the C component suppressed and store it in the memory area (2). In step [33], specify the memory area (2) and set the predetermined cyst amount lf2 (lf2 is 2 or 3).
Set to .

In the flowchart of the drawing, the predetermined shift amount is expressed as a designated correlation calculation area. In step [34], within this predetermined shift amount, the shift calculation algorithm process of FIG. 6 is performed for high-order spatial frequency components around fN/2. As a determination means for determining the validity of the deviation calculation result, it is determined in step [35] whether the maximum correlation exists within a predetermined shift amount lf2 (Correl=1) and the reliability of the calculation is sufficient (Inform=1). Determine Y
If es, the shift amount Lm is converted into a defocus amount in step [36].

[0040] If No in step [35],
In step [37], specify the memory area (1) and set the predetermined shift amount lf to lf2 (if lf1>lf2, lf
1≒N/2). In step [38], within the predetermined shift amount lf1, D. The shift calculation algorithm process shown in FIG. 6 is performed based on the spatial frequency component that sufficiently includes the C component. Step [36] is executed based on this result.

As described above, in this embodiment, first, D. A shift calculation is performed based on the data with the C component suppressed. When the determining means determines that the result of the second process is not valid, a shift calculation is further performed in a wide shift range based on image data that sufficiently includes a DC component (first process). When automatic focus adjustment of the photographic lens is performed continuously, the photographic lens is generally located near the in-focus position, and therefore, in most cases, it is expected that the result in step [35] will be Yes, and the calculation process will be performed. Time can be significantly reduced.

Next, a second embodiment will be described with reference to FIG. 8, in which the process (first process) of performing shift calculation in a wide shift range is performed first. Steps [41], [42], [
43] are steps [31], [37], and [38] in FIG.
is the same as D. It is determined in step [44] whether the interpolated shift amount Lm calculated from data sufficiently containing the C component indicates that the photographing lens is near the in-focus position, and if No, step [50] ) to convert the shift amount into a defocus amount. If the above determination is Yes, in step [45], step [43]
The calculation results Lm(1) and E(3) are stored. Steps [46], [47] and [48] are the same as steps [32], [33] and [34] in FIG. The shift calculation results Lm(2) and E(2) are calculated based on data from which the C component has been sufficiently or considerably removed. Step [4
9], one of the interpolation shift amounts Lm(1) and Lm(2) is selected from the parameters E(1) and E(2). The subject himself is D. When there is only a spatial frequency component near the C component, the calculation result Lm(2
) becomes very low in reliability, and at this time the parameter E also becomes very small. Therefore, when E(2) is less than a predetermined threshold Eth and E(1) is sufficiently large, Lm(1) is selected; otherwise, Lm(2) is selected, and in step [50] Convert the focus amount.

In the above, the microcomputer only performs a single post-filter process, but a case where the microcomputer performs a plurality of post-filter processes will be described below. In Figure 1,
In order to clearly show that the microcomputer 8 performs a plurality of post-filter processes, the j-th post-filter unit 10j is shown outside the post-filter unit 10 by a dotted line block.

Next, a third embodiment, which is an improvement on the second embodiment, will be described. In FIG. 9, steps [51] to [5
5] are the same as steps [41] to [45] in FIG. 8, respectively. In step [55], D. After the calculation results Lm(1) and E(1) based on the data containing the C component are stored in memory, the parameter j is set to 2 in step [56]. Memory area (1) in step [57]
Figure 2 (d-
3) Weighted addition filter processing of the weighted coefficient sequence, generally speaking, as shown in Figure 2 (d-1), the peak frequency f2 is determined by the pitch p, Nyquist frequency fN = 1/2p.
/2 and D. A weighted addition filter process is performed on the MTF characteristic that sufficiently removes the C component, and the resulting data is stored in the memory area (2). In step [58], the memory area (2) is specified as the memory to be used, and the predetermined shift amount lf is
For example, specify 2 as =lf2 and set the threshold value Eth=Eth
Specify 2. Under these conditions, the algorithm processing shown in FIG. 6 is performed in step [59] below, and as a result, step [6
0], and if Inform=1 and Correl=1, the defocus amount is detected from the interpolated shift amount Lm at this time in step [65].

If the condition in step [60] is not satisfied, step [61] calculates the calculation result Lm(2
), E(2) is stored. Next step [62]
Then, parameter j is increased by 1 to make j=3. In step [63], it is determined whether j=5. Since j=3 now, the process moves to step [57]. step [
57], when j=3, the peak frequency f3 exists near 1/2 of the peak frequency f2 as shown in FIG. 2(f-1), and D. The data is stored in the memory area (3) after being filtered to sufficiently remove the C component. Note that this filtering process is performed by performing the weighted addition process in FIG. 2(e-3) and the weighted addition process in FIG. 2(f-3) on the data in memory area (1) that has undergone the filtering process in FIG. 2(b-2). This is achieved by sequentially performing weighted addition processing in series. Subsequent steps [58]
The processing from to [62] is the same as when j=2. The parameter j in step [62] is 4. The weighted addition filter in step [57] when j=4 has a peak frequency f4 near 1/4 of the peak frequency f2 and D. This results in MTF characteristics in which the C component is sufficiently removed. This is the same as the subsequent step i=3. In this way, the filter characteristics in step [57] are sequentially changed as j increases, but the value of the predetermined shift amount lfj may be constant regardless of j. When the parameter j becomes 5, the process moves to step [64] and the data E(1), E(2), E(3), E(4) stored in steps [55] and [61] are
), the interpolation shift amounts Lm(1) and Lm(
2) Select the most reliable one from , Lm(3), and Lm(4), or if none of them is highly reliable, proceed to step [65] without selecting anything.

In this example, the filtering process is performed a maximum of three times in step [57], but the number of times can be selected as appropriate. Also, as j increases, as the filter in step [57], the MTF peak sequentially shifts to the lower frequency side and D. Since a filter that removes the C component is selected, even if the spatial frequency component of the object contains a large amount of only a certain specific component, highly accurate detection can be expected by any of the above filter processes.

In the embodiments described so far, the sampling pitch p=2p stored in the memory area (1) is used to calculate the defocus amount when the defocus amount is not near the in-focus area.
0 data was used as is. If it is not near the focus, set the predetermined shift amount lf to a wider range, for example, lf ≒ N/2
It is desirable to take it in moderation. However, if the number N of a-series and b-series data is 50 each, the calculation time in this case becomes enormous. In addition, when calculating the defocus amount when the focus is not near the focus, the calculation accuracy of the defocus amount does not need to be very high, and since the optical image itself is blurred due to defocus, data with a relatively rough sample pitch is sufficient. It is possible to achieve the objective in this way, and by doing so, the calculation time is shortened.

Such a case will be explained with reference to FIG. First, secondary data with sample pitch p during A/D conversion is stored in memory area (1) (step [71].) When the sampling interval is multiplied by n for the data in memory area (1), Filter processing is performed to remove spatial frequency components higher than the Nyquist frequency, and data sampled at a sampling pitch of n×p is stored in the memory area (1').
Put it in. (Step [72]) In general, it is desirable to perform filter processing to remove components above the Nyquist frequency in conjunction with changing the sample pitch, but when the amount of defocus is large, the image is greatly blurred. In this case, the optical image is blurred and does not contain components higher than the Nyquist frequency of the sample pitch in question, so the above filtering process is omitted and a new sample pitch n is created.
You may sample by ×p. Next, the used memory area (1
) and specify the predetermined shift amount lf = lf1≒N'
/2 is specified. Here, the number of samples N' is approximately 1/n of the number of samples N in memory area (1). (Step [73]) Next, algorithm processing for calculating the amount of image shift is performed (Step [74]), and the obtained image shift amount Lm is corrected by a factor of n due to the fact that the sample pitch is multiplied by n. Let Lm×n → Lm (step [
75]). Next, the defocus amount is calculated from this (step [76]).

Here, we have shown an example in which the processing time is shortened by increasing the sample pitch in areas where the amount of defocus is large, but even near the in-focus area, the filtering process in the third embodiment can further reduce the MTF peak. When moving to the lower spatial frequency side, the processing time can be shortened by increasing the sample pitch accordingly. The embodiment of FIG. 11 shows such a case. Here, a fourth embodiment, which is an improvement of the first embodiment, will be described with reference to FIG. 11, in which processing near focus (second processing) with a narrow shift range is performed first.

FIG. 2(b-1)
) is stored in the memory area (1
) (step [81]), and further processed with the filter of FIG. 2 (d-3) in step [82], that is, f2 = 1/(8p0) as shown in FIG. 2 (d-1).
Data containing information regarding a spatial frequency band centered at is stored in a memory area (2). Next, use memory (
2), and set the predetermined shift amount lf to, for example, lf
=lf2=3, and the threshold value Eth regarding the amount of information
is designated as Eth2 (step [83]). Based on this, the algorithm processing (step [84]) for calculating the shift amount in FIG. 6 is performed, and as a result, the amount of information is sufficient (In
fom=1) and within the predetermined detection shift amount (Correl
= 1) (step [85]), the resulting shift amount L
The defocus amount is calculated from m (step [86]) and the process ends. If the condition in step [85] is not satisfied, this means that the system is not in the vicinity of focus, or even if it is in the vicinity of focus, the optical image only contains a spatial frequency lower than the spatial frequency f2. The calculation is performed with the sample pitch doubled to 2p in order to shorten the processing time. Therefore, if the future sample pitch is p', then p' =
2p=4p0.

Specifically, the filtering process shown in FIG. ), and the results are stored in the memory area (3) for each pitch p' = 2p. The filter characteristics at this time are as shown in Figure 2 (e-1), and the sample pitch p
' Nyquist frequency fN = 1/2p' =
Spatial frequencies above 1/8p0 have almost disappeared. Note that in step [88], the amount of information as a result of the algorithm processing in step [84] is sufficient (Inform=1).
If not, a decision is made to proceed to a loop that switches filters near focus. First, in step [89], parameter j=4 is set as the initial value of the loop. step [
FIG. 2 (
Filtering is performed on the weighting coefficient sequence of f-3) with a spatial pitch of 8p0, and the data at the sample pitch p' is also stored in the memory area (4). The data stored in area (4) is as shown in FIG. 2(f-1). Remove C component, spatial frequency f4 = 1/4p' = 1/16p0
It has been subjected to filter processing with characteristics that mainly extract the surrounding area.

Next, the memory area to be used [4] is designated, the predetermined shift amount lf = lf4 = 3, and the information amount threshold value Eth = Eth4 (step [91]). Based on this, the algorithm processing shown in FIG. 6 is performed (step [92]). In this case, steps [84] and [9]
In the algorithm processing of step [2], the predetermined shift amount lf is the same as 3, but the sample pitch is twice as different, so step [92] has a wider range for detecting the relative shift amount of the object image, and the detection area is also wider. .

[0053] If the amount of information is sufficient (Inform=1) (
step [93]) and a predetermined shift amount (Correl=
1) (Step [96]) Then, considering the sample pitch 2p at this time from Lm at this time, the image shift amount X is set to X=
Calculate by Lm×2p (step [100]). If the condition in step [93] is not satisfied, the parameter j is incremented by 1 to make j=5. In this case step [9
5] Step again through

Return to [90]. step

In [90], D is again applied to the data in memory area (3).
．． C component is removed and the spatial frequency f5 (f5 < f4
) is performed and stored in the data memory area (5). For example, f5 = 1/
Filter processing is performed such that 2f4 = 1/8p' = 1/32p0. In this case, it is possible to make the sample pitch even rougher depending on the filter at that time, but as the number of data decreases, the accuracy gradually deteriorates, so it is not preferable to set the number N of data to less than about 20 to 25.

[0054] In this way, step [91
] and [92], and if the amount of information is sufficient (step [9
3]) and within the correlation detection area (step [96]), the amount of image shift is calculated in step [100]. Step [9
If the condition is not satisfied in step [3], j = 6 in step [94].
Next, the conditions of step [95] are satisfied and step [
97]. Also, step [88] and step [96]
If the condition is also satisfied, the process moves to step [97].

At this point, the system is not near focus, or the subject is D. It has been found that it contains only spatial frequency components very close to the C component. Therefore, D with pitch p' = 2p stored in memory area (3) as memory to be used.
．． Specify the image data from which the C component has not been removed, and set the maximum data amount lf as lf3=N'/2, (N'=
N/2) and Eth4 as the threshold value (step [97]). Perform the algorithm processing of FIG. 6 (step [98]), Inform=1, Corr
If el=1 (step

[99]) From this result Lm and the pitch p' = 2p, the image shift amount X = Lm x 2p
is calculated (step [100]). The results of these steps [100] are converted into defocus amounts in step [101]. step

When the condition [99] is not satisfied, the detection is terminated as undetectable.

In this embodiment, primary data ai, bi of a series and b series with a spatial pitch p0 are sampled at a sample pitch p=2p0 after passing through the prefilter means 5 of FIG. (1), and further this stored data is sample pitch p'
=2p=4p0 sampled memory area (3)
is stored in Assuming that the number of a-series and b-series primary data is 100 each, the a-series and b-series data stored in memory areas (1) and (3), respectively, are about 50 and about 25, respectively. In this way, since the number of data to be processed is changed depending on the situation, it is possible to perform speedy processing with a limited calculation scale while effectively using information.

In all of the embodiments described above, the weighted addition process and the shift calculation process in the microcomputer are performed completely separately. However, it is also possible to partially or completely combine these two processes. For example, Ai - bi +L may be calculated in memory for a large number of i's and a plurality of L's, and this memory result may be subjected to weighted addition filter processing, and then the remaining shift calculation may be performed. The process and the deviation calculation process may be completely matched as shown in the following equation.

[0058]

[Math 4]

Here, the weighting coefficient sequence of the weighted addition filter is (-0.5, 1, -0.5).

[0060]

[Effects of the Invention] As is clear from the above explanation, in the past, it was difficult to cope with the different performance required depending on the focus adjustment state with a single shift calculation, but according to the present invention, it is difficult to cope with the different performance required depending on the focus adjustment state. It is possible to perform the shift calculation at the same time, and the calculation time does not increase accordingly, resulting in excellent responsiveness.

[Brief explanation of the drawing]

FIG. 1 is a block diagram showing an embodiment of the present invention.

FIG. 2 shows MTF characteristics and weighting coefficient sequences; FIG. 2 (a) shows the MTF characteristics of a photoelectric element array; FIG. 2 (b-1)
, (c-1), (d-1), (e-1), and (f-1) are graphs showing the MTF characteristics of the weighted addition filter, respectively; ), (c-3), (d-2
), (d-3), (e-2), (e-3), (f-2)
, (f-3) and (f-4) are diagrams showing weighting coefficient sequences, respectively.

FIGS. 3(a) and 3(b) are block diagrams each showing a transversal filter.

FIG. 4 is a flowchart regarding focus detection.

FIG. 5(a) is a graph showing the relationship between the correlation amount and the shift amount, and FIG. 5(b) is a diagram for explaining an interpolation method.

FIG. 6 is a flowchart showing an algorithm for calculating a deviation.

FIG. 7 is a flowchart diagram regarding the first embodiment.

FIG. 8 is a flowchart diagram regarding a second embodiment.

FIG. 9 is a flowchart diagram regarding a third embodiment.

FIG. 10 is a flowchart diagram regarding a third embodiment.

FIG. 11 is a flowchart diagram regarding a fourth embodiment.

[Explanation of symbols of main parts]

2, 3...Photoelectric element array 5...Pre-filter 8...Microcomputer 10, 10j...Post filter section 11...Difference calculation section

Claims (9)

[Claims] (1) In a focus detection device that photoelectrically detects the relative shift amount of a pair of object images on a photoelectric conversion means to detect the focus adjustment state of an imaging lens, the relative shift amount of the pair of object images is detected. a first processing means for calculating a shift between the pair of object images by shifting the image output of the object image within a first predetermined shift amount range under a first condition; In order to detect the relative displacement amount of the object images, the image output of the object images is shifted within a second predetermined shift amount range under a second condition different from the first condition to detect the displacement between the pair of object images. a second processing means for performing calculation, which is determined by the first predetermined cyst amount in the first condition from the detection range of the relative shift of the object image determined by the second predetermined shift amount in the second condition. The defocus amount is calculated based on a processing result of at least one of the first processing means and the second processing means. A focus detection device featuring: (2) The first condition and the second condition are
The focus detection device according to claim 1, wherein the sampling pitch is a sampling pitch for sampling the image output of the processing means and the second processing means, and each sampling pitch is different. (3) When calculating the defocus amount based on the processing result of at least one of the first processing means or the second processing means, the shift calculation is terminated with a partial shift within the range of the predetermined shift amount. The focus detection device according to claim 1, wherein the remaining shift is not performed in the case where the focus detection device is shifted. (4) The second processing means performs the shift calculation, and if the result is determined to be valid, the defocus amount is calculated; otherwise, the first processing means performs the shift calculation. A focus detection device according to claim (1), characterized in that: (5) The focus detection device according to claim 1 or 2, wherein the second condition of the second processing means is to remove a DC component from the image output. (6) The focus detection device according to claim 1 or 2, wherein the first condition of the first processing means is not to remove a DC component from the image output. (7) The focus according to claim 1 or 2, wherein the first condition is a condition that mainly extracts lower-order spatial frequency components than the second condition. Detection device. (8) The focus detection device according to claim (2), wherein the first condition is such that the sampling pitch is coarser than the second condition. (9) The first processing means performs the deviation calculation, and when it is determined as a result that the focus is near, the second processing means performs the deviation calculation, and in other cases, the first processing means performs the deviation calculation. A focus detection device according to claim (1), characterized in that the focus detection device performs the following.