JPH0296723A - Exposure controller - Google Patents

Abstract

PURPOSE:To take pictures of any object patterns at intended exposure by provid ing a network composed of an input layer, an intermediate layer and an output layer. CONSTITUTION:Through the use of a neuro-computer 25, exposure control and focal point detection are performed. The neuro-computer 25 is composed of many units, which are classified to be the input layer, the intermediate layer, and the output layer. Each unit is connected in this order; input layer intermediate layer output layer, and the network is formed. The strength of the connection is determined by learning. Namely, the layers are coupled in order with the strength of the connection that they learn so that an exposure correcting amount of average photometry is outputted according to an object pattern when it is inputted. In such a manner an output from a photoelectric conversion element is inputted to the network, and exposure is controlled according to an output from the network and those from the photoelectric conversion elements. Thus, any object patterns can be photographed at intended exposure.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to an exposure control device that can photograph any subject that satisfies certain conditions.

[Prior Art] As a conventional example of such an exposure control device, Japanese Patent Application Laid-Open No. 1986-
There is an apparatus described in Japanese Patent No. 42026.

In a multi-Ill light device that divides the subject into multiple areas and generates multiple photometric outputs corresponding to each area, depending on the maximum brightness, the maximum brightness, the intermediate brightness between the maximum brightness and the average brightness, etc. The exposure is controlled by selecting one of the brightness, the average brightness, the intermediate brightness between the average brightness and the minimum brightness, and the minimum brightness. Furthermore, Japanese Patent Application Laid-Open No. 173226/1983 discloses a device that detects a backlight condition from a spot photometry value at the center of the screen and an average photometry value for the entire screen, and displays a warning or corrects the exposure value. A technology has been disclosed that uses subject distance information for backlight detection to eliminate detection errors.

[Problems to be Solved by the Invention] These conventional examples recognize subject patterns using a microcomputer, so they may be applicable only to a limited number of subject patterns, or may make incorrect detections, resulting in unintended photographs. There were drawbacks such as. Furthermore, when attempting to perform intended exposure control for many subject patterns, the subject patterns or brightness cannot be formulated, resulting in drawbacks such as an enormous program and a long computation time. Furthermore, even if a photograph is produced as intended by the manufacturer, the evaluation of the photograph largely depends on the individual's sensibilities, so good results may not necessarily be obtained for all users.

The purpose of this invention is to solve these drawbacks by detecting the pattern of the subject using a network that generates the desired output for various types of human power, and controlling the exposure based on the result. To provide an exposure control device capable of taking a photograph with an intended exposure.

[Means and effects for solving the problem] The exposure control device according to the present invention has a coupling strength that has been learned to output an exposure correction amount for average M1 light according to the subject pattern when the subject pattern is input. A network is provided that consists of an input layer, a middle layer, and an output layer that are sequentially connected. The outputs of the plurality of charging conversion elements are input to the network, and exposure is controlled based on the output of the network and the output of the plurality of light 78 conversion elements.

[Embodiments] Hereinafter, embodiments of the exposure control device according to the present invention will be described with reference to the drawings. FIG. 1 shows its block diagram. As can be seen from FIG. 1, this embodiment uses a neurocomputer to perform exposure control and focus detection. First, the neurocomputer will be explained with reference to FIGS. 2 to 12.

Figure 2 shows a model of a neurocomputer.

This model was proposed by Rumelhart et al. and uses backpropagation (
Back propagation ton) model (hereinafter referred to as
It is called the BP model (abbreviated as BP model). A neurocomputer consists of many units (neurons), and the units are classified into a human layer, an intermediate layer, and an output layer. Each unit is connected in the direction of input layer → middle layer → output layer, forming a network (neural net). The strength of the connections between each unit is determined by learning. However, there is no mutual connection between units within each layer. Figure 2 shows the model of each unit.

Next, the principle of this BP model learning algorithm will be explained. When a certain pattern P is given to the input layer, the actual output value that appears in the output layer is 0pj0, and the desired output value at that time (hereinafter referred to as the teacher signal) is tpj, then the difference between the two Epj is as follows. expressed.

Epj-1/2 (tpj 0pj0) 2 ・
(1) To perform learning, it is sufficient to change the strength of connections of all units so as to reduce this error Epj.

When pattern P is given, the strength of coupling W from the i-th unit of the (K-1) layer to the j-th unit of the layer
The amount of change in jl is defined as follows.

Here, K is set to 0 at the output layer and increases toward the input layer.

Δp WjlKo: -a E p / 9 WjlK
-(2) aEp/aWjlK = (f3Ep/f3netpjK) (ane t pjK/ a WjlK)
−(3) Here, net p3 −ΣWjkK−
Opk"

Also, if f is expressed as a sigmoid (SigIold) function, then 0pkK=f (ne tpkK),
Equation (3) is transformed as follows. The sigmoid function is shown in FIG.

a E pj/ a WJIK -1 δpjK・opiK"l...
(4) Here, δpj is the backward propagation of the error in the 11th layer, and δpjK--F) Epj/a n
e tpjK. Therefore, equation (2) is transformed as follows. Here, η is a constant.

ΔpWjiK −η・δpjK−OpIK+1 ・・・
(5) In the case of crab knit, Epj-1/2 (t
pj −OpJ ), 0pjO−f (netp
jO), so 1.0 2 The backward propagation amount δpj0 of the output layer is as follows.

δpjO −(t pj−0pjO)・f′ (ΣW j k O
・Opk') (6) In the case of intermediate units, there is no coupling of units within each layer, so the backward propagation of error m is as follows.

δpjK −−a Ep / ane tpjK (aOpiK
/a ne epjK) )−Σδpk” WkjK
-'-f' (net pj) & ΣδpkK-1wkjK-1 夷・f' (ΣWjk -Opk"')
...(7) Kashi.

Equation (7) is a recursive function of δ.

A general formulation of Δp WjlK is as follows.

ΔpWjlK(n+1) - ηδI)J ・Opl +aΔpWH (n
) - (8), K K+, where ΔpWjlK (0) - 0, where n represents the number of times of learning. The second term on the right side of equation (8) is added to reduce error vibration and speed up convergence.

From equation (8), the bond strength is updated as follows.

WjIK(n+1) −Wj) (n)+ΔpWjIK(n)(K −0
,1,2,...) ...(9) Here,
When the sigmoid function fl is defined as fl-1/(1+e-0e")-(10), fl'-fi (1-fl), so the amount of backward propagation can be simplified as shown in the following equation.

In the case of output crab knit: δpj' = OpJ (1-Opjo) (tpj-Opjo) ・
...(11), 0 In case of intermediate unit; δpjK - OpjK (1 - OpjK) As can be seen from the above, the calculation of ΔW starts from the output layer unit and moves to the intermediate layer unit.

In this way, learning proceeds in the opposite direction to the processing of input data.

Therefore, learning using the BP model is performed as follows.

First, the training data is manually processed and the results are output. Next, the strength of the coupling is changed to reduce the resulting error (the difference between the actual output and the teacher signal).

Then, input the learning data again. This action
Repeat until ΔW converges.

FIG. 5 shows the basic circuit configuration of the BP model.

Random access memory (hereinafter referred to as RAM) 1 stores the strength of connection Wjl, and consists of N pages of 1 (-1 to N) for each layer.RAM2 stores the strength of connection Wjl when pattern P is given. Store the amount of assimilation ΔWjl, and set N of k-1 to N
Consists of pages. RAM3 is the backward transmission of error f! lδp
j, and consists of (N+1) pages from k-0 to N.

RAM4 stores the output value Opj of each unit, k-0
~N (N+1) pages. 5 is an operational circuit for Opj, 6 is an operational circuit for δpj, and 7 is an operational circuit for Δp Wjl. 9 is a sequence controller that controls the entire sequence.

The learning process using the BP model shown in FIG. 5 will be explained.

Here, the operation when the BP model is simulated by a Neumann type computer will be explained with reference to the flowcharts of FIGS. 6 to 9. Figure 6 shows Opj
7 is a flowchart of the δpj calculation, FIG. 8 is a flowchart of the Wpj calculation, and FIG. 9 is a flowchart of the learning level determination.

In step 1 (Sl), the coupling strength Wjl in RAMI is initialized to a random value. Input value 0pj in step 2
N+1 is set in the RAM 4, and in steps 3 to 9, the unit output value 0pjK is calculated by the arithmetic circuit 5 in order from the input layer to the output layer.

Next, in steps 11 to 20 of FIG. 7, the backward propagation amount δpj0 of the error in the output layer is determined by the mW circuit 6 from the output value 0pj0 and the teacher signal tpj indicating a desired output according to equation (11).

Next, in steps 21 to 24 in FIG.
10 Find (1). Note that the initial values ΔpWjlK (0) of ΔpWjlO are all 0. Step 25
Then, the arithmetic circuit 8 calculates the bond strength Wjl according to equation (9).
Find O(1). As a result of the above, the output layer op, +,
δpj', ΔpWj10.0 (1), and WjJO(]) are found. What about these after this?
The JJ period data is stored in RAMI to RAM4 in an updated form.

Next, learn the middle layer. Returning to the flowchart in FIG. 7, the 6pj0. Wj
Using lo (1) and 0pjO stored in RAM1;:t'6, the backward propagation amount δpjK of the error is determined. Next, in the flow chart shown in FIG.
jlK (1) is determined according to equation (9). Similar to the output layer, the data obtained above is stored in RAMI to RAM4 in an updated form. The above flow is the human labor layer (K-
N+1), and the first learning is completed.

By performing the above learning a plurality of times (n), the strength of coupling Wjl between each unit is determined, and when an input value Opj indicating a certain input pattern P is given, a desired output value P is determined.
A network for obtaining pj will be automatically formed.

FIG. 9 is a flowchart for determining the mean square error Ep between the actual output value Opj and the teacher signal tpj. The smaller this value is, the closer the actual output value will be to the desired output value. If the current Ep is smaller than a certain threshold value ε, the learning is finished, and if it is larger than ε, the learning is repeated.

Although learning for one input cover turn P has been described above, it is also possible to perform learning with a plurality of input cover turns and a plurality of output cover turns corresponding to each pattern. It is also possible to learn to output a specific cover turn for a plurality of cover turns.

The BP model described above can be realized with the Neumann-type microcomputers that are currently widely used in consumer devices, but as it is, one of the major strengths of neurocomputers, which is the ability to speed up through parallel processing, will not be utilized. . Therefore, it is preferable that the processes shown in FIGS. 6 to 9 be performed in parallel by a plurality of computers.

FIG. 10 shows the configuration of a parallel processing system for this purpose.

A plurality of microprocessors Pt-Pn are connected to host processor 11. The neural network shown in FIG. 2 is divided into n partial networks, and each is assigned to a microprocessor P1 to Pn. The host processor 11 controls mutual timing between the microprocessors PL and Pn, and controls the timing between the microprocessors PI to Pn.
The data distributed in n are integrated to perform processing such as pattern recognition. Each microprocessor P1 to Pn
is the output ff1O shown in FIG. 5 according to the calculation procedure described above.
Perform operations on multiple consecutive columns of pj. Therefore, the microprocessors Pl to Pn have δpj necessary to increase the output value for which they are responsible.

It is equipped with a RAM and an arithmetic circuit for storing ΔWjJ and WjJ, respectively. When the calculation of the output values of all the units in charge is completed, each processor P1 to P n E
Communication for updating data is performed while synchronizing with the The host processor 11 determines the learning achievement level and controls the timing of the microprocessors P1 to PnH.

When performing processing such as pattern recognition based on the learned results, the necessary output value Ppj0 is finally determined by performing output calculations from the input layer shown in FIG. In this case as well, by executing distributed processing using a plurality of microprocessors as shown in FIG. 11, speeding up can be achieved due to the parallelism of the neural network.

Although the circuit shown in FIG. 5 is basically required in the learning process, the configuration can be greatly simplified if only the learning results are applied.

FIG. 11 shows the basic circuit configuration in this case.

The human input data is set to 1 by sequentially calculating opj -f (J, WjkK.K.Opk") through the input section 12 (for example, an A/D converter or the like). The coefficient memory 14 in which the coupling strength WjlK is stored is a ROM or a rewritable RO.
It may be M.

FIG. 12 is a schematic block diagram of a learning system during manufacturing for products to which learning results are applied.

Product 16 is R that stores the bond strength WjlK.
Built-in OM17. 18 is a learning device, ROM1
．． 7 and the learning device 18 are basically the same as the device shown in FIG.
8 is separated. Note that it is not necessary to perform learning each time for each product of the same type, so it is also possible to copy and use the ROM 17.

In the above explanation, the learning of the BP model and the application of the results have been realized through simulations using currently used Neumann type computers.

This is mainly because learning requires complex algorithms, and it is extremely difficult for hardware to automatically self-organize the weights of connections between neurons. However, if the strength of the connection WIj is known, the BP model shown in Figure 1 can be configured in hardware if the learning results are applied to a machine.

If you want to increase speed through parallel processing or apply it to inexpensive consumer products, there is no point unless you use this method.

This can be achieved by configuring each unit in FIG. 2 with an inverter and replacing the coupling strength Wlj with a resistor network R1j, which can be easily achieved using recent LSI technology.

Next, an exposure control device to which the neurocomputer described above is applied will be explained with reference to FIG.

There is a diaphragm 19 on the front side of the photographing camera 20, and a subject image through the diaphragm is incident on a light receiving section 21 formed by arranging photoelectric conversion elements PIIn in a matrix as shown in FIG. Therefore, the light receiving unit 21 outputs the brightness information of the subject in the narrowed down state for each photoelectric conversion element, and converts it into a digital value via the amplifier 22 and A/D converter 23 and sends it to the arithmetic circuit as a BV' value. (ALU)24
supplied to The arithmetic circuit 24 is a circuit for calculating the brightness BV (-BV'-AVo) of the subject from the light that has passed through the aperture 19, and for this purpose, the open aperture (ilAVo) of the aperture 19 is input.

The BV value for each photoelectric conversion element output from the arithmetic circuit 24 is supplied to the arithmetic circuit 27 and also to the neurocomputer 25. Neurocomputer 25
The brightness distribution pattern of the subject outputted from the arithmetic circuit 24 is used as human power Opj', and finally the correction 1=CV is obtained. The strength of this connection Wjl is stored in the coefficient memory 26.

The arithmetic circuit 27 calculates the average value of brightness BV, film sensitivity SV,
Aperture value AV, shutter speed TV.

Mode signal (MOD) such as shutter priority or aperture priority
Apex calculation (BV+5V-TV+AV+
CV) and determine the shutter speed or aperture value.

The output of the arithmetic circuit 27 is sent to the shutter control device 28 and the aperture control device i! 1j29, and exposure is controlled by these. In this way, in this embodiment, exposure is controlled in consideration of the correction value calculated by the neurocomputer 25 for the average photometric value.

30 is a sequence controller.

The basic block configuration of the neurocomputer 25 may be as shown in FIG. 5, but here, in order to increase the speed, learning is performed using a parallel computer as shown in FIG. 10. Here, the connection strength Wjl of each unit is learned in advance so that an exposure correction signal is output when a subject pattern is input to the network. The strength of connection Wjl obtained as a result of learning is stored in the coefficient memory 26.

This learning system is shown in FIG. model pattern 3
An input unit 2 inputs various object patterns 0pjN+1 to the neurocomputer 33, and includes an A/D converter and the like. The teacher signal 35 includes various object patterns opjN.
The exposure compensation signal corresponding to ``l'' is input to the neurocomputer 33 as eye 1fitpj.The neurocomputer 33 uses learning to find the strength of coupling Wjl that makes the actual output 0pj0 match the teacher signal tpj. The resulting coupling strength Wjl is stored in a coefficient memory 34. This coefficient memory 34 is incorporated into the camera as a coefficient memory 26.

In order to proceed with learning efficiently, the neural network is configured such that, as shown in FIG. 15, learning is performed in independent neural networks Sll, . . . for each row, and integrated at the output layer SO. Neural network is human power layer 3
7, an intermediate layer 38, and an output layer 39, and the learning principle is as described above.

FIG. 16 shows a specific example of the model pattern.

(a) is an example of backlight photography; in this case, the correction amount CV is set to +IEV, and a person is photographed brightly. (b) is an example of the sea, and in this case, the sky is too bright and the sea appears dark, so the correction amount is set to +〇, 5EV.

(c) is an example of a neon street at night. If taken using the conventional method, the image would appear as bright as daytime, and the mood would not be expressed, so the correction amount is set to 15 EV. Note that this correction value is a correction value for the average photometric value.

The above are just three examples, but in reality, hundreds of patterns can be learned.

It should be noted that neurocomputers have the excellent property of producing correct output even for patterns that were not input during the learning process after a certain amount of learning. Very effective in solving difficult problems. In addition, the neurocomputer's learning allows it to self-organize the relationship between a huge variety of subject patterns and the main parts of the subject, which until now could not be programmed using a Neumann-type computer, making it possible to control exposure as intended. can. Furthermore, parallel control of the neurocomputer enables high-speed calculations, making it suitable for use in silver-halide cameras that require speed.

This invention is not limited to the embodiments described above, and can be modified in various ways. The above explanation deals with the case where the neurocomputer learns the main part of the subject and performs exposure control, but it is also possible to give the position of the main part of the subject as a teacher signal to the neurocomputer and perform learning such as spot metering. be.

Furthermore, by inputting not only the brightness of the subject but also temperature, humidity, etc. as manual parameters of the neurocomputer, it is possible to learn to create subtle seasonal sensations that are difficult to formalize.

〔Effect of the invention〕

As explained above, according to the present invention, a neurocomputer is trained in advance to output a desired amount of exposure compensation for various subject patterns manually, and exposure control is performed based on the output. Accordingly, it is possible to provide an exposure control device that can take a photograph with an intended exposure for any subject pattern.

[Brief explanation of the drawing]

Fig. 1 is a block diagram of an embodiment of an exposure control device according to the present invention, Fig. 2 is a diagram showing a model of a neurocomputer used in the embodiment, and Fig. 3 is a diagram showing a model of each unit constituting the network. , FIG. 4 is a diagram showing a sigmoid function, FIG. 5 is a block diagram of a neurocomputer, FIGS. 6 to 9 are flowcharts when the neurocomputer in FIG. The figure is a flowchart for determining the output Opj of each unit, Figure TS7 is a flowchart for determining the backward propagation amount of error δpj, Figure 8 is a flowchart for determining the coupling strength coefficient Wj+, and Figure 9 is a flowchart for determining the level of learning. , Fig. 10 is a block diagram of a parallel processing system, Fig. 11 is a block diagram of a device that applies learning results, Fig. 12 is a block diagram of a system that trains a device that applies learning results, and Fig. 13 is an example. A block diagram of a device for learning a neurocomputer,
FIG. 14 is a diagram showing an example of the arrangement of photoelectric conversion elements in the example, FIG. 15 is a diagram showing a network in the example, and TSL 6 (a) to (C) are diagrams showing examples of subjects to be trained. . 21... Light receiving section, 24.27... Arithmetic circuit, 25...
・Neurocomputer, 26...Coefficient memory, 28
...Shutter control device, 29...Aperture control device.

Claims (1)

[Claims] a light receiving section including a plurality of photoelectric conversion elements; an optical system that forms a subject image on the light receiving section; a network connected to the output of the light receiving section and outputting an exposure correction signal; means for controlling the exposure of the camera based on the output of the network; It comprises a single layer or a plurality of intermediate layers consisting of a plurality of units connected with each other with a strength, and an output layer consisting of a plurality of units connected with each unit of the intermediate layer with a predetermined connection strength. Exposure control device.