JPH03167535A - Exposure arithmetic unit for camera - Google Patents

Abstract

PURPOSE:To automatically obtain an exposure correcting value with respect to various screens at high speed by allowing a neutral network to learn each brightness data vector in numerous model patterns. CONSTITUTION:A subject unit is provided with a single-layered neutral network 5 consisting of plural neuron units where the brightness data vector is inputted through a prescribed synapse combining strength. The neutral network 5 is allowed to learn so as to output the exposure correcting value information on a screen when the brightness data vector as a the set of each brightness data of plural points in the arbitrary screen is inputted, and exposure is determined based on this output. Thus, most preferable exposure can be obtained in a short time in any condition.

Description

[Detailed Description of the Invention] [Industrial Application Field] The present invention relates to an exposure calculation device for a camera [Prior Art] Conventionally, exposure control for a camera has been disclosed in Japanese Patent Laid-Open No. 57-42.
As described in Publication No. 026, multiple brightness patterns are set based on actual photographic data, each brightness value measured by dividing the screen into multiple areas is compared with the brightness pattern, and center-weighted, average, There are methods that control exposure by selecting the optimal one from a range of exposure determining factors, such as emphasizing high brightness and emphasizing low brightness. Furthermore, as described in Japanese Patent Application Laid-Open No. 173226/1982, there is a device that detects a backlight condition from the spot brightness value at the center of the screen and the average brightness value of the entire screen and performs exposure correction.

[Problem to be solved by the invention]

However, although conventional exposure control can handle a limited number of pre-programmed scenes, it may not be able to handle other scenes well. Also,
Exposure control that focuses on the center of the screen will give you an appropriate exposure if the subject you want to photograph is in the center of the screen, but
In other cases, proper exposure cannot be obtained. Even in such cases, if you position the subject you want to photograph in the center of the screen and then use AE lock, you can always obtain the correct exposure regardless of the subject's position, but AE lock is cumbersome to operate. , not suitable for quick shooting.

This invention was made to address the above-mentioned circumstances,
The purpose is to provide a camera exposure calculation device with a simple configuration that can obtain optimal exposure in a short time under any circumstances.

[Means and effects for solving the problem] The camera focus detection device according to the present invention uses a single-layer neural network consisting of a plurality of neuron units into which luminance data vectors are input via predetermined synaptic connection strengths. Be equipped.

This neural network is trained to output exposure compensation value information for a given screen when a brightness data vector, which is a collection of brightness data for each point of a given screen, is input, and based on that output. to determine the exposure. Here, the range of the exposure correction value is approximately ±5EV. During learning, it is assumed that screens with similar brightness data vectors have similar exposure compensation values, and the neural network is made to learn each brightness data vector using a large number of model patterns. .

FIG. 1 schematically shows a means for determining an exposure compensation value using a neural network according to the present invention.

The input screen shown on the left side is the photographic screen shown in the finder, and the input screen shown on the left is the shooting screen shown in the viewfinder.
~n (n is a positive integer, shown in FIG. 1)), and brightness data Di is obtained for each area Pi. The neural network shown on the right inputs the brightness data vector Di for each area of the input screen and determines the exposure correction value Ex for that screen. The 221 ral network used in this invention has a plurality of luminance data vectors (in this case, 2
It is a single-layer network consisting of 5) neuron ● units, and a total brightness data vector is input to each neuron ● unit shown as a square.

Neural network training consists of the following two steps. First, brightness data vectors for a large number of model patterns are input into a neural network, and a predetermined synaptic connection strength is obtained by unsupervised learning. Unsupervised learning is the process of modifying the synaptic connection strength of each neuron unit to selectively respond to a given input vector without any training data. Automatically classified on the network. Then, the output neuron unit of the neural network when the luminance data vector of the model pattern is input again to the neural network is associated with the exposure correction value in that model pattern. That is, the values written in each neuron unit in FIG. 1 are exposure correction values.

When photographing, as shown in Figure 1, when the luminance data vector Di for each area Pi in the screen to be photographed is input to the neural network trained in this way, any one neuron unit Firing strongly (in Figure 1, firing neuron units are shaded)
, the exposure compensation value Ex corresponding to that neuron unit
(E7 in FIG. 1) can be determined to be the appropriate exposure correction value, and if the exposure is corrected accordingly, photographs can always be taken with the appropriate exposure.

The neural network used in this invention will be explained with reference to FIGS. 2 to 4. This neural network model is based at Helsinki University of Technology (Finland)
It was proposed by κohonen and is called a self-organizing network. The learning method is known as self-organizing feature mapping.

Figure 2 shows the model of each 221 Ron◆ unit. The neuron unit is connected to each input value of the input vector via a synaptic connection, compares the input value with each synaptic connection strength, and generates an output according to the degree of matching between the two.

The output y1 of the i-th neuron unit is expressed by the following equation.

yl=f(ΣwljXxkD −= (1)
j=t Here, f O is the output function of the neuron unit (usually a sigmoid monotonically increasing function), Wlj is the synaptic connection strength of the i-th neuron unit to the j-th input xkj, and xkj is the The jth input value of the kth input vector, where n is the number of dimensions of the input vector.

Neural networks are constructed by arranging each neuron unit two-dimensionally, but here we will focus on
Figure 3 shows a one-dimensional model for TTT conversion.

As shown in FIG. 3, this neural network performs self-organizing feature mapping that introduces the effect of feedback coupling of signals from the output of the neuron ◆ unit to the input. Here, wjk is the feedback coupling strength from the kth neuron unit to the ith neuron unit.

FIG. 4 shows the degree of influence of such feedback connections on the synaptic connection strength. In other words, due to the effect of feedback connections, the synaptic connection strength exhibits a characteristic change in the shape of a Mexican hat, and when the output value of a particular neuron unit increases, neuron units located in a positional relationship close to it will also be affected by it. The output value of neuron units further outside the neuron unit increases, but the output value of neuron units further outside the neuron unit decreases.

If we simplify this so that the learning of the synaptic connection strength considering the effect of this feedback connection is performed only in the vicinity of the optimally matched neuron unit, the learning process (self-organizing feature mapping) of this neural network is as follows. It can be done by following these steps.

#1: Initialize Wl(0) with a random number and set the number of learning times t
-0. Here, Wi(L)=(wll(L).
wl2(L). ...win(t)).

Hereinafter, step #2 for each input vector Xk. #3
Repeat the process.

#2: Set t-t+1, and find the optimal neuron unit C that satisfies the following equation.

II Xk −we(t) If = wln ( If
Xk −W1(t)If l...(2) Here, Xk − (xkf, xk2. =-xk
n), we(t) is the synaptic connection strength of the optimally fitting neuron unit C.

#3: In the case of iεN c(t), W i(t+1) −
W l(t) +α(t) x (Xk−wl(t
)), and in cases other than iεN c(t), Wl(t+1
)−Wl(t).

Here, α(1) is the learning coefficient, N e(t) is the set of 22-Ron● units in the vicinity of the optimal fitting neuron unit C, and usually α(t) +11 N c(t
) are both monotonically decreasing functions.

By performing such self-organizing feature mapping, the synaptic connection strengths of neighboring neuron units become similar. Furthermore, since synaptic connection strengths reflecting statistical properties such as similarity and correlation between model patterns are obtained, vector quantization of each model pattern is performed and classification of model patterns becomes possible. Furthermore, the process of determining the position of the main subject in any photographic screen can be considered as the process of determining the optimal fit neuron unit C, but as can be inferred from step #2, this is a fast decision process with a small amount of calculation. Processing is possible.

〔Example〕

Hereinafter, an embodiment of an exposure calculation device for a camera according to the present invention will be described with reference to FIGS. 5 to 7. FIG. 5 is a block diagram of an automatic exposure control camera as an example. A plurality of light-receiving elements 2i (L-1 to L-n) are provided to detect the luminance of each of a plurality of regions of a subject on one screen that is incident through the lens 1, and the luminance data D of each region obtained by these elements is provided.
i is supplied to the normalization calculation circuit 3. Normalization calculation circuit 3
normalizes the luminance data Di to a real value from 0 to 1 so that it can be input to the neural network 5, and obtains the above-mentioned input data pattern xkj.

The output of the normalization calculation circuit 3 has the synaptic connection strength already acquired through learning, and is input to a single-layer neural network 5 consisting of a plurality of neuron units.

The neural network 5 receives the luminance data xkj and calculates the synaptic connection strength w of each neuron unit.
ilt) and the vector distance Nl-Σ(xkj-wij
(t) ) 2 is calculated, the minimum value Dc-sin (Nl) of the vector j:\ vector distance is searched, and the exposure correction value Ex corresponding to the neuron unit that provides the minimum value is output. Furthermore, in general,
Neural network 5 is 5 to 20 of the total number of photometric areas
Consists of about twice as many neuron◆ units. Is this exposure compensation value information exposure? The signal is input to the fIr calculation circuit 4. The lens's open F-number, aperture value, shutter speed, film sensitivity, and mode information (shutter priority, aperture priority, etc.) are also supplied from the photographing information input section 8 to the exposure calculation circuit 4. The exposure calculation circuit 4 performs abex calculation (brightness value - open F value + film sensitivity - shutter speed 10 aperture lii!) according to the mode information and exposure correction calculation, and sends the results to the shutter control circuit 6 and the aperture control circuit 7. supply to.

FIG. 6 shows a specific circuit configuration of the neural network 5. The neural network 5 consists of memories 9, 10, 11, 12 for storing various data, and various arithmetic circuits 13, 14. The memories include a memory 9 that stores the normalized luminance data xkj output from the normalization calculation circuit 3, and a memory 9 that stores the synaptic connection strength wi acquired through learning.
lt), a memory 10 for storing normalized luminance data xkj
and the synaptic connection strength wlj(t) vector distance MN
A memory 11 for storing i, and a memory 12 for storing exposure correction value information indicating the correspondence between neuron units on the learned neural network and exposure correction values of input patterns are provided. As the calculation circuits, a circuit 13 that calculates the inter-vector distance Nl from the normalized luminance data xkj and the synaptic connection strength wlj(t), and a circuit 14 that detects the minimum value of the inter-vector distance 11ftN1 are provided.

Normalized luminance data xk output from the normalization calculation circuit 3
j is stored in the memory 9 in the neural ◆ network 5. The synaptic connection strength w1j (
t) is stored in the memory 10 in advance. Memory 9, 1
The data of 0 is supplied to the vector distance calculation circuit 13,
Therefore, the distance between vectors N1-Σ(xkj-wlHt))
' is performed, and the result of the calculation is stored in the memory 11. Since this inter-vector distance calculation is independent for each neuron unit, parallel calculation is possible. Therefore, the inter-vector distance calculation circuit 13 is composed of a plurality of processors, each of which performs processing for each neuron unit, and is capable of high-speed processing. In addition, in a neural network dedicated arithmetic element, that is, a neurochip, each neuron unit is responsible for this arithmetic processing.

The minimum value detection circuit 14 detects the optimal combination competition unit C having the minimum value from the obtained inter-vector distance Nl. The exposure calculation circuit 4 reads out the exposure correction value information Ex corresponding to the optimum neuron unit from the correspondence relationship determined in advance and stored in the memory 12, and reads out the exposure correction value information Ex corresponding to the obtained optimum neuron unit. E
x and the aperture value, aperture value, shutter speed, film sensitivity, and mode information supplied from the shooting information input unit 8, perform avex calculation and exposure correction calculation, and send the results to the shutter control circuit 6 and the aperture control device. Supply to 7.

As described above, it becomes possible to perform automatic exposure control appropriate to the scene of the subject within the input screen.

In addition, as exposure correction value information, equation (1) defining the output of the neuron unit: yi-f(Σw 1jX
Using an exposure compensation value that is a weighted average of multiple exposure compensation values corresponding to multiple neurons selected in order of p+ from the maximum value obtained by x kj) in proportion to the output value of the neural network. is also possible. In addition, a neural network that applies only the learned results is possible with only the above configuration, but
In order to perform initial learning and additional learning when adapting the neural network to a camera operator, the following learning circuit configuration is required.

FIG. 7 explains the circuit configuration for neural network learning. This structure includes a synaptic connection strength calculation circuit 16 and an exposure correction value information setting circuit 1 in the circuit shown in FIG.
This can be achieved by simply adding 8.

In this operation, first, the synaptic connection strength wlj of each neuron unit of the neural network is initialized with a small random number. Luminance data for each region regarding a large number of learning model patterns is stored in the memory 15.
Data Ex regarding exposure correction values is stored in the memory 17. After determining the optimal neuron unit by detecting the minimum value of the distance between vectors using the luminance data in the memory 15 as in the case of FIG. 6, the synaptic connection strength calculation circuit 16 determines the optimal neuron unit. Synaptic connection strength W of a group of neuron units near the center
1 is modified and learned using the following equation.

However, iεNc(t).

Wl(t+1) −Wt(t)+α(t) X (Xk −Wi(t))
...(3) Here, α(1) is the learning coefficient O~1
Nc(t) is the set of neuron units near the optimal neuron unit C, its initial value is not very small compared to the size of the neural network, and α(1 ) and Nc(L) are both monotonically decreasing functions. t is the number of learning times, and learning is repeated until the maximum number of learning times t wax.

After completing the learning for all model patterns, the exposure compensation value information setting circuit 18 determines with which neuron unit of the neural network the exposure compensation value of each model pattern stored in the memory 17 is associated. The result is set in the memory 12 as exposure correction value information.

yi (=f (Σwijx x kj) )≧θk
(threshold value in the k-th input pattern), if the neuron
The exposure correction value of the k-th input pattern is associated with the unit i.

When performing additional learning to adapt a trained neural network to a camera operator, when an arbitrary input pattern is given, the optimal neuron unit is determined by detecting the minimum value of the distance between vectors. , the synaptic connection strength Wl of the neuron unit group Nc' which is in the vicinity of the neuron unit and has a common corresponding exposure correction value is corrected and learned by the synaptic connection strength calculation circuit 16 using the following equation.

Wl = WI + (20
- (4) However, in LεNc', αO is a learning coefficient.

This invention is not limited to the embodiments described above, and can be modified in various ways. The learning method explained is unsupervised learning,
When an input pattern is given, the output of the neural network is compared with the teaching data (exposure correction value of the input pattern), and if the neural network fires correctly, the synaptic connection strength approaches the input pattern, and if it fires incorrectly, it If the signal is firing, supervised learning that moves the synaptic connection strength away from the input pattern is also possible.

It is also possible to learn by including teaching data as input values. Note that in order to be able to return to the initial learning state before additional learning at any time, it is necessary to store the synaptic connection strength in the initial learning state in memory.

Furthermore, the input data that the neural network self-organizes is not only the brightness data of the entire screen mentioned above, but also the brightness data of the focus position and its vicinity in conjunction with the automatic focusing mechanism, as well as the main subject and its surroundings. Nearby luminance data may also be used.

In addition, there is a main subject spot brightness control that performs exposure control using the brightness data of the main subject as a spot brightness value without using the self-organization of a neural network. It is also possible to perform main subject weight point brightness control.

Furthermore, although the embodiments have been described with respect to an automatic exposure control camera, the present invention may be applied to a manual exposure control camera that only displays determined exposure information in a viewfinder or the like.

〔Effect of the invention〕

As explained above, according to the present invention, in processes that were conventionally difficult to formulate and program, such as determining exposure compensation values according to the scene of the subject, the neural network It is possible to provide an exposure calculation device for a camera that can automatically and quickly obtain exposure correction values for various screens simply by learning luminance data vectors. Then, if we use the parallel processing of neural networks,
Even higher speeds are possible. In addition, by having a neural network learn the camera operator's preferred compositions and compositional habits, and adapting the neural network to the camera operator, exposure can be automatically determined in accordance with the photographer's intentions.

[Brief explanation of the drawing]

Fig. 1 is a diagram showing an outline of the camera exposure calculation device according to the present invention, Fig. 2 is a diagram showing a model of a neuron unit, Fig. 3 is a diagram showing a model of a neural network, and Fig. 4 is a diagram showing a feedback connection. FIG. 5 is a block diagram of an embodiment of an automatic exposure control camera as an embodiment of the present invention. FIG. 6 is a diagram showing the circuit configuration of a neural network. FIG. 2 is a diagram showing a circuit configuration for learning a neural network. DESCRIPTION OF SYMBOLS 1... Lens, 2... Light receiving element, 3... Normalization calculation circuit, 4... Exposure calculation circuit, 5... Neural...
Network, 6...Shutter! IJOl1 circuit, 7
. . . Aperture control circuit, 8 . . . Photographing information input unit.

Claims (1)

[Claims] Consisting of means for measuring the brightness of a plurality of points within a photographic screen and a plurality of neuron units, each neuron unit having a set of brightness data for each point output from the measuring means. , i.e., a single layer neural network in which the luminance data vector is input via a predetermined synaptic connection strength, and a neural network that receives said predetermined synaptic connection strength by modifying the synaptic connection strength to selectively respond to the luminance data vector of the model pattern. learning means for acquiring synaptic connection strengths and associating an output of the neural network with an exposure correction value of the model pattern when the luminance data vector of the model pattern is inputted again to the neural network; 1. An exposure calculation device for a camera, comprising: means for determining exposure according to an output of the neural network when a brightness data vector of a screen is input to the neural network.