JPH0587631A - Photometric device of camera - Google Patents

Abstract

PURPOSE:To obtain the exposure value optimum to an object by weighting the photometric output of a divided photometric region to extract some characteristics from the photometric output and calculating fidelity from the extracted characteristics and performing the weighting of coupling on the basis of fidelity to determine an exposure value. CONSTITUTION:The beam from an object passes through an imaging lens 2 to be reflected by a quick return mirror 3 and is formed into an image on a focusing screen 4 to be observed. The image of the object formed on the focusing lens 4 is re-formed on a photometric photodetector 7 by an image re-forming lens 6 and the photodetector 7 converts the detected beam photoelectrically to emit photometric output. An operator 8 combines the pattern of splitting photometry with the value obtained by extracting the characteristics of said pattern on the basis of the photometric output from the photodetector to operate an exposure value. At the time of photographing, a shutter driver 9 and an iris driver 10 are controlled on the basis of the operated exposure value to expose a film 5.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a photometric device for a camera which divides a photographic screen into a plurality of parts for photometry.

[0002]

2. Description of the Related Art Conventionally, as a photometric device for a camera of this type, the exposure value of a main subject is relatively high with a comparatively high probability by dividing photometry of a photographing screen with a plurality of light receiving elements and then comparing outputs of the respective photometric elements. It is known to be able to match. For example, one of the divided shooting screens
If the maximum brightness of the area is high and the brightness difference with other areas is large, it is considered that the scene includes the sun, etc., and the high brightness part is deleted from the photometric value, or the low brightness area is emphasized. It is possible to perform photometry.

Further, in order to obtain an appropriate exposure value for a subject having a special brightness distribution, "the maximum output of the light receiving elements is weighted, and the minimum output is weighted. An exposure calculation method (Japanese Patent Publication No. 2-7010) is proposed in which brightness information is calculated by adding the weighted value depending on the screen position, and the weighting coefficient is determined according to the scene. Has been done.

Further, although a neural network is known as one having excellent pattern recognition ability, the photometric output divided into the neural network is inputted,
A photometric device for a camera has been proposed in which the connection weight of the network is determined by learning in advance, and the output of this network is used as an exposure value while the connection weight is fixed in the state of being mounted on the camera. Kaihei 2-967
23). If this network is constructed appropriately, a proper exposure value can be obtained for a subject having a learned pattern.

[0005]

However, in the above-described conventional photometric calculation method, the exposure value of the subject at the boundary of whether or not it corresponds to the assumed shooting scene is not sufficient due to a slight change in the screen. There was a problem of becoming stable.

Further, in order to adapt to many shooting scenes, it becomes necessary to classify into many patterns,
The selection of the optimum exposure value becomes complicated, and it is impossible to predict an unexpected shooting scene at all, and there is a possibility that the exposure value will be far from an appropriate exposure value.

Further, by appropriately constructing a network for weighting shooting scenes, a proper exposure value can be obtained for a shooting scene that is assumed in advance. For that purpose, the network is large. The scale becomes so large that it cannot be handled by the microcomputer installed in the camera, and an enormous amount of data is required for pattern classification.

The object of the present invention is to solve these problems,
A photometric device for a camera that can give a smooth exposure value to changes in shooting scenes, predict exposure results, and give optimum exposure values to various subjects with a small amount of learning. Is to provide.

[0009]

In order to solve the above problems, a camera photometric device of the present invention is a camera photometric device comprising a photometric light receiving element for dividing a photographic screen into a plurality of photometric regions for photometry. A feature extraction means for extracting a plurality of subject information indicating a feature of the subject by outputs from a plurality of divided photometric regions of the photometric light receiving element; and weighted addition of the subject information according to a plurality of predetermined weighting factors. A first arithmetic means for performing arithmetic operations to generate a plurality of outputs; a fitness degree calculating means for computing a fitness degree for the plurality of outputs from the first arithmetic means according to the subject information; and a first arithmetic means It comprises a second arithmetic means for performing a weighted addition operation on a plurality of outputs in accordance with the conformity calculated by the conformity calculating means to generate an exposure control signal.

[0010]

According to the present invention, the photometric output of the divided photometric area is weighted, some features are extracted from the photometric output, the degree of conformity is calculated from the extracted features, and the combination is further combined according to the degree of conformity. The exposure value is determined by weighting.

[0011]

EXAMPLES Examples will be described below with reference to the drawings.
The present invention will be described in detail. 1 to 4 are views showing an embodiment of a photometric device for a camera according to the present invention, FIG. 1 is a schematic view showing the entire camera, and FIG. 2 is a plan view showing a photometric light receiving element. 3 is a conceptual diagram showing the network structure of the arithmetic unit, and FIG. 4 is a flow chart showing the operation of the arithmetic unit.

The photometric device of this embodiment is applied to a single-lens reflex camera 1, and as shown in FIG. 1, a photographing lens 2, a quick return mirror 3, a focusing screen 4, a re-imaging lens 6, a photometric device. The light receiving element 7, the arithmetic unit 8, the shutter drive unit 9, the diaphragm drive unit 10 and the like.

The light flux from the subject passes through the taking lens 2 and is reflected by the quick return mirror 3 to form an image on the focusing screen 4 for observation. The subject image formed on the focusing screen 4 is re-imaging lens 6
Is re-imaged on the light receiving element 7 for photometry.

The light-receiving element 7 is an element for photoelectrically converting the received light beam into a photometric output, and as shown in FIG. 2, the photographing screen is divided into five, and the respective photometric outputs (levels B1 to B5). Are connected to the arithmetic unit 8.

A well-known microcomputer is used as the arithmetic unit 8, and the divided photometry pattern and its characteristics are controlled based on the photometric output from the light receiving element 7 according to the flow chart of FIG. The exposure value is calculated by combining the extracted values. The output of the arithmetic unit 8 is connected to a shutter drive unit 9 and an aperture drive unit 10. During shooting, the shutter drive unit 9 and the aperture drive unit 10 are controlled based on the calculated exposure value, and the film 5 is displayed. Exposure is performed.

Next, referring to FIG. 3, the network structure of the arithmetic unit 8 will be described. The arithmetic unit 8 is
The input layer 81, the intermediate layer 82, and the output layer 83 are composed of three layers.

The input layer 81 includes levels I1 to I6 which are inputs to the network. Level I1 is
The brightness of the central portion of the screen is shown, and the photometric output level B1 is directly input. The level I2 indicates the brightness of the upper part of the screen, and the average value of the photometric output levels B2 and B3 is input. Level I3 indicates the brightness at the bottom of the screen, and the average value of the photometric output levels B4 and B5 is input. The level I4 indicates the brightness of the highlight portion of the screen, and the maximum value among the photometric output levels B1 to B5 is input. The level I5 indicates the brightness of the shadow portion of the screen, and the minimum value of the photometric output levels B1 to B5 is input. Level I6 is a constant term, and 1 is always input.

The input value of the input layer 81 is determined so as to show the characteristics of the subject, but is not limited to these characteristics in particular, and a value showing other characteristics, for example, by performing signal processing, is used. The performance is further improved by adding a squared value of the input value or a higher-order value higher than that.

The extraction elements F1 and F2 are input values to the fitness calculating section 84. The extraction element F1 receives the same value as the level I4, which is the maximum value among the photometric output levels B1 to B5, and here, this value is the brightness of the subject. The extraction element F2 has photometric output levels B1 to B
The maximum brightness difference of 5 is input.

Six neurons Nj are prepared in the intermediate layer 82, and they are connected to the respective inputs of the levels I1 to I6 by the learned weight coefficient Wij. At this time, the output of the neuron Nj is expressed by the following equation. Nj = .SIGMA.WijIi (i: 1 ... 6, j: 1 ... 6) (1) As shown in equation (1), the neuron Nj of this embodiment is
Is a simple linear form, the output for different inputs changes linearly, and the output does not greatly deviate from what the photographer expected.

Next, the outputs Nj of these neurons are
Fitness K1 to K6 determined by the fitness calculator 84
According to the following equation, weighted arithmetic mean is performed, and the output layer 83
Is output as a control value O. O = (ΣKj Nj) / ΣKj (j: 1 ... 6) (2)

The values of the goodnesses of fit K1 to K6 are obtained by so-called fuzzy processing according to the following six rules. Rule 1 is "the subject is dark and
The rule is that "the neuron N1 is adopted when the brightness difference is small", and the rule 2 is that "the neuron N2 is adopted when the subject is bright and the brightness difference is small", and the rule 3 is , "When the subject is very bright and the brightness difference is small, the neuron N3 is adopted", and the rule 4 is
The rule is "When the subject is dark and the brightness difference is large, the neuron N4 is adopted", and the rule 5
"If the subject is bright and the difference in brightness is large,
The rule 6 is "adopt a neuron N5" and the rule 6 is "adopt a neuron N6 when the subject is very bright and the brightness difference is large."

By the way, in the network according to this embodiment, in order to differentiate the function of each neuron Nj, the fitness Kj of each neuron obtained by fuzzy processing is added, and learning is performed by the following equation. Wnij = Woij + e (T-Nj) IiKj (3) However, e is a parameter that determines the magnitude of one correction, and takes a very small positive value. Also, T is a target value,
Woij is the previous weighting factor, and Wnij is the new weighting factor.

By performing the learning using the equation (3), each neuron Nj is learned by the data that conforms to the corresponding rules 1 to 6. For example, the neuron N1 is trained using data that fits into the region 1 in FIG. 7, and after learning, the optimal connection weight is obtained for the region 1.

Further, by looking at the size of each weighting coefficient Wnij after learning, the degree to which each input is emphasized in the areas 1 to 6 can be known, and the control of the input can be predicted. For example, among the weighting factors Wi1 for the neuron N1, the weighting factor W11
Is the largest value, it can be seen that the input I1, that is, the photometric value B1 in the central portion is most important.

The learning of these weighting factors Wij is usually carried out with data collected assuming various subjects. Furthermore, it is also possible to prepare some sets of weighting factors Wij that have been learned for each shooting subject such as a specific subject, for example, portrait, landscape, etc., so that the photographer can select them according to the subject. Several sets may be prepared in the camera, or another weighting factor Wij may be input from the outside.

Next, according to the flow chart of FIG.
The operation of this embodiment will be described with a focus on the arithmetic processing of. First, in step S1, the photometric values B1 to B5 from the light receiving element 7 are acquired through an AD converter or the like.

In step S2, the levels I1 to I6 of the input values to the network are calculated from the photometric values B1 to B6 acquired in step S1 based on the following equation. In addition,
MAX is a function showing the maximum value, and MIN is a function showing the minimum value. I1 = B1 (4) I2 = (B2 + B3) / 2 (5) I3 = (B4 + B5) / 2 (6) I4 = MAX (B1 to B5) (7) I5 = MIN (B1 to B5) )… (8)

In step S3, the extraction elements F1 and F2 are calculated from the level calculated in step S2 based on the following equation. F1 = I4 (9) F2 = I4-I5 (10)

In step S4, the conformances K1 to K6 are
It is calculated from the extraction elements F1 and F2 and the membership function based on the following formula. K1 = MIN (Bmax1 (F1), dB1 (F2)) (11) K2 = MIN (Bmax2 (F1), dB1 (F2)) (12) K3 = MIN (Bmax3 (F1), dB1 (F2)) (13) K4 = MIN (Bmax1 (F1), dB2 (F2)) (14) K5 = MIN (Bmax2 (F1), dB2 (F2)) (15) K6 = MIN (Bmax3 (F1), dB2 (F2)) (16)

As shown in FIG. 5, the membership function is a function that gives the fitness of the expression that the function Bmax1 is "dark", the function Bmax2 is "bright", and the function Bmax3 is "very bright". As shown, the function dB1 is made up of "small" and the function dB2 is made up of a function showing the fitness of the expression "large". That is, in Rules 1 to 6 described above, when there is an input of F1 with respect to the subject brightness (hereinafter referred to as Bmax), the matching degree with respect to the expressions “dark”, “bright”, and “very bright” is the membership function Bmax1 ( F
1), Bmax2 (F1), and Bmax3 (F1). Also, when there is an input of F2 with respect to the luminance difference (hereinafter referred to as dB), the conformity to the expressions of "small" and "large" is the membership function dB1.
(F2) and dB2 (F2).

These membership functions are stored in the memory of the microcomputer of the arithmetic unit 8 as compatible values for each input value as a table, or are realized on a program by combining a plurality of straight line expressions. It has become.

Next, in step S6, the levels I1 to I
6 and the learned connection weighting coefficient Wij, the neurons N1 to N6 are calculated based on the above-mentioned equation (1).
Here, with reference to FIG. 7, description will be given of how the neurons Nj of the intermediate layer 82 on the Bmax-dB coordinate are adapted. Each number (1 to 6) corresponds to the number of the neuron Nj in the intermediate layer 82, and the boundary of the region indicated by the dotted line is not clear, and the neurons of the adjacent intermediate layer 82 are near the boundary. Since it takes an intermediate value of the outputs N1 to N6, the output changes smoothly with respect to changes in Bmax (F1) and dB (F2).

Finally, in step S7, the goodness of fit K1.
From K6 and the neurons N1 to N6, the control value O is calculated according to the above-mentioned equation (2). That is, the goodnesses of fit K1 to K6 of the rules 1 to 6 calculated in step S4 are used as the calculation elements that become the weight Kj of the coupling between the intermediate layer 82 and the output layer 83 shown in FIG. 3, and the result of weighted addition averaging is the control value. It becomes O, and the target exposure value is obtained.

Next, specifically, the fitness K1 of each neuron
How to obtain K6 will be described with reference to FIGS. Figure 8
Indicates the case where F1 is input as the subject brightness, and at this time, the matching degree Bma for the expression “dark”
x1 (F1) is 0.3, the fitness Bmax2 (F1) for the expression "bright" is 0.7, and the fitness Bmax3 (F1) for the expression "very bright" is 0.
become.

FIG. 9 shows the case where the luminance difference F2 is input. At this time, the goodness of fit dB1 (F2) for the expression "small" is 0.9, and the goodness of fit dB2 (F2 for the expression "large". ) Becomes 0.1.

Therefore, the suitability K1 to K6 of the rules 1 to 6 when the inputs F1 and F2 are input are as follows in step S4 of FIG. K1 = MIN (0.3,0.9) = 0.3 K2 = MIN (0.7,0.9) = 0.7 K3 = MIN (0,0.9) = 0 K4 = MIN (0. 3, 0.1) = 0.1 K5 = MIN (0.7,0.1) = 0.1 K6 = MIN (0,0.1) = 0 Using these K1 to K6, the steps of FIG. The control value O is obtained in S7.

The present invention is not limited to the embodiments described above, and various modifications and changes can be made, which are also within the scope of the present invention. For example, in this embodiment, the fitness K1 of the neuron Nj is
The method of performing fuzzy processing on all inputs was used to determine K6. However, if the calculation time of the CPU of the arithmetic unit 8 is shortened or the program capacity is restricted, the value of the fitness K is calculated. Alternatively, the system may be one in which is 1 or 0 and does not take an intermediate value, that is, only one neuron Nj that fits the area shown in FIG. 9 is selected. In this way, although the smoothness near the boundary of the region becomes a little rough, the number of boundaries is much smaller than that of the conventional system, so the accuracy is higher than that of the conventional system, and the boundary The problem is unlikely to occur, and the calculation speed is increased because only the calculation of the selected neuron is required.

In this embodiment, the case where the number of rules (the number of neurons N) is 6 has been described, but the number of rules is not limited to this number, and the performance increases as the number of rules increases. This is the required accuracy and the installed C
It may be selected according to the capacity of PU or the like.

Although the smaller value is taken when calculating the goodness of fit, the product of both may be calculated.

[0041]

As described above, according to the present invention,
Since the degree of conformity is calculated and further weighted based on the extracted characteristics, a smooth exposure value can be obtained for a subject that is a boundary of an assumed shooting scene.

Further, since the output can be predicted by making the network structure linear and clarifying the function of each neuron, it is possible to select a few photometric shooting scenes without selecting many subject patterns. It is possible to give a more appropriate exposure value to the change of, and to give an optimal exposure value that can be predicted for various subjects with a small learning amount.

Further, the arithmetic unit is small and simple, and the arithmetic processing can be performed by the microcomputer installed in the conventional camera, which facilitates the commercialization.

[Brief description of drawings]

FIG. 1 is a schematic view showing an entire camera to which an embodiment of a photometric device for a camera according to the present invention is applied.

FIG. 2 is a plan view showing a photometric light receiving element used in an embodiment of a photometric device for a camera according to the present invention.

FIG. 3 is a conceptual diagram showing a network structure included in an arithmetic unit of an embodiment of a photometric device for a camera according to the present invention.

FIG. 4 is a flowchart for explaining the operation of the arithmetic unit of the embodiment of the photometric device for the camera according to the present invention.

FIG. 5 is a diagram showing a membership function representing a degree of conformity with respect to subject brightness in an embodiment of a photometric device for a camera according to the present invention.

FIG. 6 is a diagram showing a membership function representing a degree of conformance with respect to a luminance difference in an embodiment of a photometric device for a camera according to the present invention.

FIG. 7 is a diagram showing a combination of a luminance difference and a matching degree of subject luminance.

FIG. 8 is a diagram showing a state in which the degree of suitability for the expression “dark” is determined in the embodiment of the photometric device for a camera of the present invention.

FIG. 9 is a diagram showing a state in which the degree of conformity to the expression “small” is determined in the embodiment of the photometric device for a camera of the present invention.

[Explanation of symbols]

 1 1-lens reflex camera 2 Shooting lens 3 Quick return mirror 4 Focusing screen 5 Film 6 Re-imaging lens 7 Light receiving element for photometry 8 Arithmetic device 9 Shutter drive device 10 Aperture drive device

Claims (1)

[Claims] 1. A photometric device for a camera, comprising a photometric light receiving element for dividing a photographic screen into a plurality of photometric areas for photometry, wherein the output of the photometric light receiving element from the plurality of divided photometric areas is used to measure the subject. Feature extraction means for extracting a plurality of subject information indicating features, first computing means for performing a weighted addition operation on the subject information according to a plurality of predetermined weighting factors, and generating a plurality of outputs, according to the subject information A fitness calculation means for computing a fitness for a plurality of outputs from the first computing means, and a weighted addition operation for a plurality of outputs of the first computing means according to the fitness calculated by the fitness computing means, A photometric device for a camera, comprising a second arithmetic means for generating an exposure control signal.