JPH05260370A - Automatic iris circuit - Google Patents

Abstract

PURPOSE:To attain sufficient iris control with respect to various luminance patterns of an object. CONSTITUTION:An image pickup area is divided into mXn sets of blocks by a block integration circuit 12 and each luminance of the blocks is obtained. Each luminance is standardized by an averaging circuit 14 and a standardizing circuit 16, and the result is given to a neural network 18. The learning is finished in the neural network 18 to obtain a proper output with respect to the inputted luminance pattern in advance, the output of the neural network 18 is normalized by a mean value and the result is compared with a reference luminance voltage at a comparator circuit 22. As a result, an iris 24 or an AGC circuit 11 or the both are controlled.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION The present invention relates to a VTR with a built-in camera.
The present invention relates to an image pickup apparatus such as a (video tape recorder), and more particularly, to an iris that regulates the amount of incident light from a subject and an auto iris circuit that controls an AGC (automatic gain control) circuit for obtaining a luminance signal of a desired level.

[0002]

2. Description of the Related Art FIG. 14 shows a conventional auto iris circuit 9
0 is a block diagram of a part of a camera-integrated VTR in which 0 and this auto iris circuit 90 are used. Referring to FIG. 14, this camera-integrated VTR has a lens 26 for condensing incident light from a subject and forming an image of the subject on a predetermined image forming surface, and an image forming surface of the lens 26. The light receiving surface is arranged so that it is arranged between the CCD (charge coupled device) 10 for converting the optical image of the subject into an electric signal, the lens 26 and the CCD 10, and passes through the lens 26 to C.
The iris 24 for controlling the amount of light rays incident on the CD 10 and the iris 24 so that the brightness signal has a desired level based on the brightness signal output from the CCD 10.
And a conventional auto iris circuit 90 for controlling

A conventional auto iris circuit 90 is a CCD
When high-intensity light is incident on 10, the incident light amount is iris 2
4, a high-brightness clipping circuit 92 for clipping the high-brightness signal in order to prevent it from being extremely narrowed down and thereby causing an extreme decrease in brightness, and a large weighting to the brightness signal from the bottom of the screen captured by the CCD 10. Lower-weighted photometry circuit 94 for outputting a signal indicating the brightness level, a center-weighted photometry circuit 96 for heavily weighting the center of the screen and outputting the signal as a brightness level, and a lower-weighted photometry circuit. 94 and a detection circuit 98 for taking a weighted average of the outputs of the center-weighted photometry circuit 96, a reference luminance generation circuit 38 for outputting a constant reference luminance signal, and a connection to the detection circuit 98 and the reference luminance generation circuit 38. The signal output from the detection circuit 98 is compared with the reference luminance signal, and the signal output from the detection circuit 98 is output based on the comparison result. There comprising a comparator circuit 22 for controlling the iris 24 to vary equal direction as the reference luminance signal.

Downward-weighted photometry is a photometric system in which the lower part of the screen is heavily weighted. The reason why this photometric method is often adopted is as follows. Camera integrated V
An image captured by TR generally includes the sky above. The amount of light incident from the sky is large. On the other hand, the subject is often on the ground, and in this case, the subject is present in the lower half of the screen. The amount of light incident from the ground is small. Therefore, if the iris control is performed based on the brightness of the entire screen, the iris 24 may be excessively narrowed down, and the subject on the ground may not be photographed sufficiently bright. The downward-weighted photometry method is to control the iris 24 so that a subject on the ground can be photographed with sufficient brightness by giving a large weight to a luminance signal generated from the bottom of the screen. ..

On the other hand, the center-weighted photometry system is a system for performing photometry by heavily weighting the center of the screen. This is the method adopted because the subject often exists in the center of the screen during normal shooting. By giving a large weight to the central portion, the iris 24 is provided so that the subject existing in the central portion is photographed with appropriate brightness.
Is controlled.

The conventional auto iris circuit 90 and camera-integrated VTR shown in FIG. 14 operate as follows. The lens 26 collects incident light from the subject and forms an optical image of the subject on the image plane. The CCD 10 photoelectrically converts an optical image of the subject formed on the light receiving surface into an electric signal and outputs a luminance signal. This signal is applied to a video signal circuit (not shown) and subjected to predetermined video signal processing. At this time, the amount of light entering the CCD 10 from the lens 26 is regulated by the iris 24.

The high-intensity clip circuit 92 includes a CCD 10
The luminance signal output by is given. The high-brightness clipping circuit 92 clips high-brightness light incident on the CCD 10 when the brightness becomes a high level, and clips the high-brightness light so that the level of the brightness signal given to the lower-weighted photometry circuit 94 and the central-weighted photometry circuit 96 is increased. Keep below a certain value. By suppressing the brightness level to a certain value or less in this way, it is possible to prevent the iris 24 from being excessively narrowed down when the high brightness light is incident, and the brightness level from dropping extremely.

The lower-weighted photometry circuit 94 and the central-weighted photometry circuit 96 output signals representing the brightness level according to their respective photometric methods, and apply them to the detection circuit 98. The detection circuit 98 includes a lower weighted photometry circuit 94 and a center weighted photometry circuit 9
The signal output from 6 is weighted averaged, detected, and provided to the comparator 22. The comparator 22 compares the signal level output from the detection circuit 98 with the reference luminance signal supplied from the reference luminance generation circuit 38, and changes the output of the detection circuit 98 in the direction of approaching the same level as the reference luminance signal. Control the iris 24.

By using a so-called dual photometry system using two circuits, that is, the lower-weighted photometry circuit 94 and the center-weighted photometry circuit 96, high-luminance light enters the upper part of the screen and high-luminance light surrounds the screen. It is possible to prevent the brightness of the screen from being extremely lowered when is included.

Also, when the iris 28 is fully opened, that is, when the amount of light incident on the CCD 8 is not suppressed by the iris 28 and the luminance signal level output from the CCD 8 has not reached the desired level. is there. In such a case, the auto iris circuit 91 as shown in FIG. 15 is used.

The auto iris circuit 91 shown in FIG.
Is different from the auto iris circuit 90 shown in FIG. 14 in that when the input of the high-intensity clip circuit 92 is below the predetermined level of the output of the CCD 10, automatic gain control is performed so as to amplify the brightness signal to a desired level. A to do
That the second reference luminance generation circuit 3 is connected to the output of the GC circuit 11 and generates the second reference luminance signal.
9, the detection circuit 98 and the reference luminance generation circuit 39 are connected, the signal output from the detection circuit 98 is compared with the reference luminance signal, and the signal output from the detection circuit 98 is equal to the reference luminance signal based on the comparison result. AGC to change
And a comparator circuit 23 for controlling the circuit 11.

The auto iris circuit 91 shown in FIG.
14 is almost the same as the operation of the auto iris circuit 90 shown in FIG. The difference is that the operation of the AGC circuit 11 is controlled by the comparison circuit 23. For this reason, even when the iris 28 is fully opened,
It is possible to bring the level of the luminance signal output from the AGC circuit 11 close to a desired level, and it is possible to obtain a luminance signal of a desired level for a subject with low illuminance as compared with the auto iris circuit 90 shown in FIG. .. However, C
The AGC circuit 11 does not operate when the luminance signal output from the CD 10 is at a predetermined level or higher. Ie CCD
The output of 10 is given to the auto iris circuit 91 as it is.

[0013]

However, the above-mentioned conventional auto iris circuits 90 and 91 have the following problems. Consider, for example, the case where shooting is performed at the sea or at a ski resort. In this case, from the bottom of the screen CCD
The light incident on 10 is larger than that in the case of normal ground imaging. On the other hand, the subject is often a human, and when the downward weighted metering method is used, the incident light from the background (sea, snow, etc.) at the bottom of the screen has a high brightness, and the subject becomes extremely dark. There is a problem that it will end up. Also, if the subject you want to shoot is in the peripheral part instead of the center of the screen, or even if the subject is in the center part, there is backlight in the background part of the center part, or a subject with a high brightness level such as spot light enters. In such a case, the iris is intensively controlled by the high brightness light. As a result, there arises a problem that the subject is not photographed with sufficient brightness. This is a problem that cannot be overcome even by the gain control by the AGC circuit.

Therefore, an object of the present invention is to perform iris control so that a desired level of luminance signal can be obtained even for various luminance patterns on a screen to be photographed. It is to provide an auto iris circuit capable of

An object of the present invention is to provide iris control and AGC control so that a desired level of luminance signal can be obtained even for various luminance patterns on the screen to be photographed. Is to provide an auto iris circuit capable of performing.

[0016]

According to a first aspect of the present invention, an auto iris circuit adjusts an output of an image pickup device to a desired level in response to a brightness signal output from the image pickup device whose incident light amount is regulated by an iris. An auto iris circuit for controlling the iris, and obtains a luminance value standardized according to a predetermined standard for each of a plurality of predetermined partial regions of the image pickup region of the image pickup device based on the luminance signal. Standardized brightness value acquisition means for, and a neural network that inputs the standardized brightness value,
Iris control means for controlling the iris in response to the output of the neural network.

The auto iris circuit according to claim 2 is
The auto iris circuit according to claim 1, wherein the normalized brightness value acquisition means outputs the brightness value acquisition means for acquiring the brightness value of each partial region based on the brightness signal, and the brightness value acquisition means. It includes a brightness value averaging means for obtaining an average value of the brightness values of the partial areas, and a brightness value normalizing means for standardizing the brightness values of the respective partial areas based on the output of the brightness value averaging means. The iris control means compares the output of the neural network with the output normalizing means for normalizing the output of the brightness value averaging means and the output of the output normalizing means with a predetermined reference signal, and outputs the comparison result. Comparing means for controlling the iris on the basis.

The auto iris circuit according to claim 3 is
The auto iris circuit according to claim 1, wherein each partial region is set to one of a plurality of predetermined luminance sections based on the normalized luminance value of the partial region output from the normalized luminance value acquisition means. It further includes a classifying unit for classifying and outputting the frequency distribution information. The neural network receives the output of the classification means as an input.

The auto iris circuit according to claim 4 is:
The auto iris circuit according to claim 3, wherein the imaging device has an imaging region divided into one or a plurality of partial region groups each including one or a plurality of partial regions. This auto iris circuit classifies the brightness values of the partial areas output by the standardized brightness value acquisition means into each partial area group and outputs one integrated brightness value for each partial area group. Further includes integration means. The neural network receives the output of the luminance value integrating means in addition to the output of the classifying means. The iris control means divides the image pickup area into a plurality of predetermined photometric areas, weights the luminance signal for each photometric area with a predetermined coefficient, and outputs a predetermined important photometric signal output means for outputting a predetermined important photometric signal. And comparing the output of the neural network with the output normalizing means for normalizing the output by the weighted photometry signal and the output of the output normalizing means with a predetermined reference signal, and comparing for controlling the iris based on the comparison result. And means.

The auto iris circuit according to claim 5 is
The iris and the automatic gain control circuit adjust the output of the imaging device to a desired level in response to a luminance signal output from the imaging device whose incident light amount is regulated by the iris and whose level is adjusted by the automatic gain control circuit. An auto iris circuit for controlling, for each of a plurality of predetermined partial areas of an image pickup area of an image pickup device, based on a luminance signal, for obtaining a luminance value standardized according to a predetermined standard. Normalized luminance value acquisition means, a neural network that inputs the normalized luminance value,
Control means for controlling the iris and the automatic gain control circuit in response to the output of the neural network.

The auto iris circuit according to claim 6 is
The auto iris circuit according to claim 5, wherein the normalized brightness value acquisition means outputs the brightness value acquisition means for acquiring the brightness value of each partial area based on the brightness signal and the brightness value acquisition means. It includes a brightness value averaging means for obtaining an average value of the brightness values of the partial areas, and a brightness value normalizing means for standardizing the brightness values of the respective partial areas based on the output of the brightness value averaging means. The control means compares the output of the neural network with an output normalizing means for normalizing based on the output of the luminance value averaging means, and the output of the output normalizing means with a predetermined first reference signal. A second comparing signal for controlling the iris on the basis of the first comparison means and a second comparing signal for comparing the output of the output normalizing means with a predetermined second reference signal and controlling the automatic gain control circuit based on the comparison result. And the comparison means of.

The auto iris circuit according to claim 7 is
The auto iris circuit according to claim 5, wherein each partial area is set to one of a plurality of predetermined luminance sections based on the normalized luminance value of the partial area output from the normalized luminance value acquisition means. It further includes a classifying unit for classifying and outputting the frequency distribution information. The neural network receives the output of the classification means as an input.

The auto iris circuit according to claim 8 is
The auto iris circuit according to claim 7, wherein the imaging device has an imaging region divided into one or a plurality of partial region groups each including one or a plurality of partial regions. This auto iris circuit classifies the brightness values of the partial areas output by the standardized brightness value acquisition means into each partial area group and outputs one integrated brightness value for each partial area group. Further includes integration means. The neural network receives the output of the luminance value integrating means in addition to the output of the classifying means. The control means divides the image pickup area into a plurality of predetermined photometric areas, weights the luminance signal for each photometric area with a predetermined coefficient, and outputs a predetermined important photometric signal output means for outputting a predetermined important photometric signal. , For comparing the output of the neural network with an output normalizing means for normalizing the output by the weighted photometry signal and the output of the output normalizing means with a predetermined first reference signal, and controlling the iris based on the comparison result. And a second comparing means for comparing the output of the output normalizing means with a predetermined second reference signal and controlling the automatic gain control circuit based on the comparison result.

[0024]

In the auto iris circuit according to the first aspect of the invention, based on the luminance signal output from the image pickup device whose incident light amount is regulated by the iris, each of the partial regions of the image pickup region of the image pickup device is in accordance with a predetermined standard. Normalized brightness values are obtained and given to the neural network independently of the position of each subregion. In response to the output of the neural network, the iris control means controls the iris. Since the input of the neural network is performed regardless of the position of each partial area, the influence of the luminance distribution of the subject on the iris control is reduced. Further, since the brightness value is standardized, the pattern change of the input data to the neural network is reduced and the load on the neural network is reduced.

In the standardized brightness value acquisition means of the auto iris circuit according to the second aspect, the brightness value acquisition means acquires the brightness value of each partial area, and the brightness value averaging means calculates the average value of the brightness values of these partial areas. Desired. And
The average value normalizes the brightness value of each partial area. Further, the output of the neural network is standardized based on the output of the luminance value averaging means. The standardized output is compared with a predetermined reference signal by the comparison means, and the iris is controlled based on the comparison result. Since the brightness value is standardized by the average value, the pattern change of the input data to the neural network due to the change of the brightness level is reduced.

In the classifying means of the auto iris circuit according to the third aspect, any one of a plurality of brightness sections predetermined for each partial area based on the normalized brightness value of the partial area output from the normalized brightness value acquiring means. It is classified into one. Then, the frequency distribution information is input to the neural network. That is, the characteristics of the luminance distribution are given to the neural network as frequency distribution information.

The brightness value integrating means of the auto iris circuit according to claim 4 classifies the brightness values of the respective partial areas outputted by the normalized brightness value acquiring means into the respective partial area groups, and then into the respective partial area groups. To output one integrated luminance value. This integrated luminance value is input to the neural network together with the output of the classifying means. The weighted photometric signal output means outputs a predetermined weighted photometric signal by weighting the luminance signal for each photometric area of the image pickup area with a predetermined coefficient. The weighted photometric signal is input to the neural network, and the output of the neural network is standardized by the output standardization means with the weighted photometric signal as a reference. The comparing means compares the output of the output normalizing means with a predetermined reference signal,
Control the iris based on the comparison result. This allows
It can be corrected by the output of the neural network according to the pattern of the brightness distribution, in addition to the control by the conventional weighted metering method.

According to a fifth aspect of the present invention, the standardized luminance value acquisition means of the auto iris circuit is based on the luminance signal output from the image pickup device whose incident light amount is regulated by the iris and whose level is adjusted by the automatic gain control circuit. A brightness value standardized according to a predetermined standard for each of the partial areas of the image pickup device is acquired and given to the neural network. The control means controls the iris and the automatic gain control circuit in response to the output of the neural network. Since the input of the neural network is performed regardless of the position of each partial area, the influence of the luminance distribution of the subject on the control of the iris control and the automatic gain control circuit is reduced. Further, since the brightness value is standardized, the pattern change of the input data to the neural network is reduced and the load on the neural network is reduced.

In the normalized brightness value acquisition means of the auto iris circuit according to the sixth aspect, the brightness value of each partial area is acquired by the brightness value acquisition means. By the brightness value averaging means,
The average value of the brightness values of the partial areas output by the brightness value acquiring means is obtained, and the brightness value normalizing means standardizes the brightness values of the respective partial areas based on the output of the brightness value averaging means. The output of the neural network is standardized by the output normalizing means based on the output of the luminance value averaging means. The first comparing means compares the output of the output normalizing means with a predetermined first reference signal, and controls the iris based on the comparison result. The second comparing means compares the output of the output normalizing means with a predetermined second reference signal, and controls the automatic gain control circuit based on the comparison result.
Since the brightness value is standardized by the average value, the pattern change of the input data to the neural network due to the change of the brightness level is reduced.

According to a seventh aspect of the present invention, in the auto-iris circuit classifying means, each of the plurality of brightness sections is defined in advance for each partial area based on the normalized brightness value of the partial area output from the normalized brightness value acquiring means. Classify into one. Further, the frequency distribution information is given to the neural network. That is, the characteristics of the luminance distribution are given to the neural network as frequency distribution information.

The brightness value integrating means of the auto iris circuit according to claim 8 classifies the brightness values of the respective partial areas outputted by the normalized brightness value acquiring means into the respective partial area groups, and then into the respective partial area groups. To output one integrated luminance value. This integrated luminance value is given to the neural network together with the output of the classification means. The weighted photometry signal output means outputs a predetermined weighted photometry signal by weighting the luminance signal for each photometry area with a predetermined coefficient. The output standardizing means standardizes the output of the neural network by the weighted photometric signal. The first comparing means compares the output of the output normalizing means with a predetermined first reference signal, and controls the iris based on the comparison result. The second comparing means compares the output of the output normalizing means with a predetermined second reference signal,
The automatic gain control circuit is controlled based on the comparison result. As a result, the iris control by the conventional weighted photometry can be performed and the iris control can be corrected by the output of the neural network according to the pattern of the luminance distribution.

[0032]

1 is a block diagram of a part of an auto-iris circuit and a camera-integrated VTR including the iris auto circuit according to a first embodiment of the present invention. With reference to FIG. 1, this camera-integrated VTR includes a lens 26, an iris 24, a CCD 10, and an auto iris circuit 2 for controlling the iris 24 based on a luminance signal output from the CCD 10.

The auto iris circuit 2 includes an A / D (analog-digital) conversion circuit 28 for converting an analog luminance signal output from the CCD 10 into a digital signal, and an A / D conversion circuit 28.
Based on the digital brightness signal output from the / D conversion circuit 28, the image pickup screen of the CCD 10 is divided into a plurality of blocks (partial areas), each of which is divided into m parts vertically and n parts horizontally and the brightness value of each block is calculated. M × n block integration circuit 12 for obtaining by integration, an average value circuit 14 for obtaining an average value of luminance values of a plurality of blocks output from the block integration circuit 12, the block integration circuit 12, and the average value The normalization circuit 16 connected to the circuit 14 and for normalizing the brightness value of each block output from the block integration circuit 12 using the average value given from the average value circuit 14, and the normalization circuit 16 It is connected to the neural network 18 that receives the normalized luminance values of m × n blocks, the average value circuit 14 and the neural network 18, and Network 18, a standardization circuit 20 for standardizing the output of the network 18 using the average value given from the average value circuit 14, a reference brightness generation circuit 38 for generating a reference brightness signal, a reference brightness signal and a standardization circuit. 20 and a comparator 22 for outputting a signal for controlling the iris 24.

Referring to FIG. 4, the block integration circuit 12 divides the image pickup screen 82 of the CCD 10 (FIG. 1) into m × n blocks of m in the vertical direction and n in the horizontal direction as described above. The brightness of the block is obtained by integrating the brightness signal for each block.

Referring to FIG. 2, the neural network 18 is a hierarchical neural network and has m ×
The output from the n-block integration circuit 12 and the normalization circuit 16
The input layer 40 which receives m × n luminance data standardized by the input layer, the intermediate layer 42 which receives the output of the input layer 40 as an input, and the output of the intermediate layer 42 which is an input and whose output is the normalization circuit 20. Output layer 44 (see FIG. 1).

The input layer 40 includes S + 1 neurons 46.
including. Among the neurons 46, the standardized luminance data of each block output from the normalization circuit 16 is input to each of the 1st to Sth neurons.
That is, S = m × n holds. The S + 1th neuron 46 has no input and always outputs -1. S +
This first neuron is called an inhibitory neuron.

The intermediate layer 42 includes H + 1 (H ≧ 1) neurons 48. 1 to H of these neurons 48
The input of the th neuron 48 is connected to each of the S + 1 neurons of the input layer 40. The H + 1-th neuron has no input like the S + 1-th neuron of the input layer 40, is an inhibitory neuron, and always outputs -1.

The output layer 44 includes one neuron 50. The outputs of the H + 1 neurons 48 of the intermediate layer 42 are connected to the inputs of the neurons 50, respectively. The neuron 50 outputs one signal based on the output of the neuron 48 of the intermediate layer 42.

Referring to FIG. 3, for example, 1 of the intermediate layer 42
Each of the nth to Hth neurons 48 has N inputs. If the i-th input value is y _i , each input value is multiplied by the weighting coefficient w _i , and the neuron 48 outputs the value z according to the following equation.

[0040]

[Equation 1]

In the equation (1), ω _i represents the coupling coefficient (weighting coefficient) of this neuron 48 with the neuron of the previous layer that supplies the i-th input (i = 1, ..., N). Further, the function sig (x) (sigmoid function) is 0 ≦ sig
It is an increasing function that satisfies (x) ≦ 1. An example of sig (x) is given by the following formula (2).

[0042]

[Equation 2]

If the equation (2) is expressed in a graph, the curve shown in FIG. 10 is obtained. It is assumed that the neural network 18 of the iris control circuit 2 of the present embodiment has finished learning by previously giving some luminance patterns (input patterns) and desired iris control signals (teacher signals) corresponding thereto. ． In the learning process, each neuron
Data output from the output layer 44 and a desired output value are given to the given luminance pattern, and the error is used to change the strength of the coupling between the neurons.
By repeating this for a plurality of brightness patterns, each brightness pattern is separated, and after learning is completed, the neural network 18 can perform desirable iris control even for brightness patterns that have not been learned. The strength of the connection between each neuron is adjusted so that it is possible. That is, when the learning is completed, the neural network 18 classifies the input brightness pattern into several categories, and the input pattern is present even when the brightness pattern that has not been learned is input. It becomes possible to obtain desired output values by classifying them into categories.

With reference to FIGS. 1 to 4, the auto iris circuit 2 of the first embodiment operates as follows. The optical image of the subject formed on the image plane by the lens 26 is
It is converted into a luminance signal by the CCD 10. This signal is given to a video signal processing circuit (not shown) and subjected to predetermined video signal processing. This luminance signal is also digitized by the A / D conversion circuit 28 of the auto iris circuit 2 and given to the block integration circuit 12. Block integration circuit 12
4 divides the screen of the CCD 10 into m × n blocks as shown in FIG. 4, obtains the luminance value of each block, and supplies it to the average value circuit 14 and the normalization circuit 16. The average value circuit 14 calculates the average value of the brightness values for each of the m × n blocks given from the block integration circuit 12,
It is given to the standardization circuits 16 and 20. The normalization circuit 16 normalizes the brightness value of each block given from the block integration circuit 12 with the average value given from the average value circuit 14,
It is given to the neural network 18 as m × n inputs.

Input layer 40 of neural network 18
1 to S (= m × n) -th neuron 46 is given a predetermined one from the m × n normalized luminance values from the normalizing circuit 16. The neuron 46 gives an output to each neuron 48 of the intermediate layer 42 according to the pattern of the change of the output with respect to the input, which is obtained by learning in advance. Each neuron 48 of the intermediate layer 42 is connected to the input layer 40.
Inputs from the S + 1 neurons are obtained, outputs are obtained according to the coupling strength between the neurons which has been given by learning in advance, and the outputs are given to the neurons 50 of the output layer 44. The neuron 50 of the output layer 44 outputs a voltage that changes according to the output change pattern for the input, which is obtained by learning in advance, with respect to the data given from the H + 1 neurons 48 of the intermediate layer 42.

In this way, the neural network 18
Classifies the input luminance pattern into several categories according to the learning result that has been completed in advance, and also classifies the luminance pattern that has not been learned into a certain category to obtain a desired output value. Obtainable.

Referring to FIG. 1, the output of the neural network 18 is standardized by the normalizing circuit 20 by the average value given by the average value circuit 14, and is given to the comparator 22. The comparator 22 compares the output of the normalization circuit 20 with the reference luminance signal, and controls the iris 24 so that the voltage from the normalization circuit 20 changes in the direction to be equal to the reference luminance signal.

As described above, in the auto iris circuit 2 of the first embodiment, the position of the screen is not weighted, and the screen is divided into a plurality of blocks to perform the iris control by the neural network. Therefore, it is possible to correct the iris control signal at the time of backlighting or when spot light is incident, which is difficult with the conventional technique. Further, in this embodiment, the input value and the output value of the neural network are standardized by the average value.
Even if the luminance level of the luminance signal output from the CCD 10 changes, the change of the pattern of the input data given to the neural network 18 is small. Therefore, the burden of pattern separation of the neural network can be reduced even when the brightness level changes.

FIG. 5 is a block diagram of the auto iris circuit 4 according to the second embodiment of the present invention. Like the auto iris circuit 2 shown in FIG. 1, the auto iris circuit 4 of the second embodiment also has a CCD 10 of a camera-integrated VTR.
And is for controlling the iris 24 according to the luminance signal given from the CCD 10. Lens 2
6, the iris 24, and the CCD 10 are the same as those shown in FIG.

The auto iris circuit 4 is connected to the output of the CCD 10, and an A / D conversion circuit 28 for A / D (analog-digital) converting an analog luminance signal output from the CCD 10 and an A / D conversion circuit. Based on the luminance signal converted into a digital signal by 28, the image pickup screen 82 of the CCD 10 is divided into m × n blocks as shown in FIG. 4, and the luminance signal is integrated for each block to obtain an integrated value. And a mean value circuit 14 connected to the output of the block integrating circuit 12 for calculating the mean value of the luminance values of m × n blocks,
Connected to the outputs of the block integration circuit 12 and the average value circuit 14, the brightness values of m × n blocks output from the block integration circuit 12 are standardized using the average value given from the average value circuit 14. Standardization circuit 16 for
Based on the standardized luminance value of each block output from the normalization circuit 16, m × n blocks are classified into h stages, and the classification result is the number of blocks belonging to each stage (frequency distribution information). Circuit 3 for outputting as
0 and the output of the histogram circuit 30 are connected to the hierarchical neural network 32, the output of the neural network 32 and the average value circuit 14, and the output of the neural network 32 is the average value from the average value circuit 14. The standardization circuit 20 for normalization, the reference luminance generation circuit 38 for outputting the reference luminance signal, and the inputs of the standard luminance generation circuit 38 and the outputs of the standard luminance generation circuit 38 and the standardization circuit 20 are respectively connected to the standardization circuit 20. Comparing the output voltage and the voltage of the reference luminance signal, a comparator 22 for controlling the iris 24 so that the output voltage level from the normalization circuit 20 changes in the direction in which it becomes equal to the voltage level of the reference luminance signal. Including.

Referring to FIG. 6, the neural network 32 has an input layer 58 which receives as input the h histogram classified data provided from the histogram circuit 30,
The intermediate layer 60 that receives the output of the input layer 58 as input, and the intermediate layer 6
And an output layer 62 having an output of 0 as an input.

The input layer 58 includes h (h ≧ 2) neurons 52. The arbitrary i-th neuron 52 receives the i-th stage histogram classification data given from the histogram circuit 30 as an input. That is, the histogram circuit 30 provides the i-th neuron 52 with information indicating how many blocks the normalized brightness belongs to the i-th stage section.

The intermediate layer 60 includes H neurons 54. Each neuron 54 receives all the outputs of the h neurons 52 included in the input layer 58 as inputs, and gives one output to the output layer 62. The output layer 62 includes one neuron 56. The neuron 56 receives the outputs of all the neurons 54 included in the intermediate layer 60 as input, and outputs the output determined according to the pattern of the input signal by the normalization circuit 20.
Give to.

Referring to FIG. 7, for example, neuron 54 included in intermediate layer 60 has N (= h) inputs y _{1 to} y.
_{Has N.} Each input y ₁ ~y _N is weighted by the weighting coefficients ω ₁ ~ω _N defining the degree of coupling between the neurons of the preceding layer, respectively. The weighting factors ω _{1 to} ω _N are changed based on the error between the output pattern of the neural network 32 for various input patterns and the desired output pattern in the learning process described above. The output z of the neuron 54 is determined according to the following equation (3).

[0055]

[Equation 3]

In equation (3), y _i is the neuron 54
Is the i-th input value, ω _i is a coupling coefficient (weighting coefficient) (where i = 1, ..., N = h) that determines the degree of coupling of this neuron 54 with the i-th neuron in the previous layer, and θ is Each represents the threshold value of the neuron 54. An example of the sigmoid function sig (x) is given by the following equation (4).

[0057]

[Equation 4]

The values of the weighting factors ω _{1 to} ω _N in each neuron of the neural network 32 are determined in the learning process of the neural network 32. This value is applied to the neural network 32 as described above, and is changed based on the error between the output value of the neural network 32 and the desired output value for each input pattern. By repeating this, learning is performed and each luminance pattern is separated. After the learning is completed, the neural network 32
The connection strength between neurons is such that desired iris control can be performed even for a luminance pattern that has not been learned.

The auto iris circuit of the second embodiment shown in FIGS. 5 to 7 operates as follows. The optical image of the subject formed by the lens 26 is converted into a luminance signal by the CCD 10. This luminance signal is given to a video signal processing circuit (not shown). The luminance signal is also given to the A / D conversion circuit 28 and converted into a digital signal. The block integration circuit 12 obtains the brightness value of each block by integrating the input digital signal for each of m × n blocks as shown in FIG. 4, and acquires the average value circuit 14 and the normalization circuit. Give to 16. Average value circuit 14
Is an average value of m × n luminance values, and the normalization circuit 1
Give to 6, 20. The normalization circuit 16 calculates the luminance values of the m × n blocks output from the block integration circuit 12,
The value is normalized by the average value given from the average value circuit 14 and given to the histogram circuit 30.

The histogram circuit 30 includes a normalization circuit 16
The luminance values of m × n blocks given by the above are classified into h levels according to their levels. The h frequency distribution information is input to the neural network 32 by the histogram circuit 30.

Input layer 58 of neural network 32
In the i-th neuron 52, for example, the frequency of the i-th stage is input from the histogram circuit 30 as histogram classified data. The neuron 52 processes the input value according to the weighting learned in advance and gives it to each neuron 54 included in the intermediate layer 60. Each neuron 54 included in the intermediate layer 60 corresponds to each neuron 5 in the input layer 58.
Preliminarily learned weighting is applied to the input from 2 to perform an operation, and one output is given to the neuron 56 of the output layer 62. The neuron 56 of the output layer 62 responds to the signal input from each neuron 54 of the intermediate layer 60 by
Pre-learned weighting is performed and calculation is performed to obtain one output, which is given to the normalization circuit 20.

The output of the neural network 32 is standardized according to the average value given from the average value circuit 14 and given to the comparator 22 in order to keep the value proper. The comparator 22 compares the voltage level of the reference brightness signal from the reference brightness generation circuit 38 with the voltage level of the signal supplied from the normalization circuit 20, and the voltage level of the signal output from the normalization circuit 20 is the reference brightness signal. The iris 24 is controlled so that it changes in the same direction as the voltage level of the iris 24.

In the auto iris circuit according to the second embodiment, the image pickup screen 82 of the CCD 10 is divided into a plurality of blocks which are not weighted by the position, and the brightness of each block is obtained. The iris is controlled based on this. Therefore, unlike the conventional case, there is no fear that the iris 24 is inappropriately controlled according to the position of the high-intensity part in the subject. Further, since the luminance value of each block obtained by the block integration circuit 12 is once standardized by using its average value, in the classification by the histogram circuit 30,
A histogram having a feature can be obtained by the distribution pattern of the brightness irrespective of the brightness level. Therefore, a histogram is obtained for various input luminance patterns, and the classification result based on the histogram is obtained by the neural network 32.
If the neural network 32 is learned in advance by an error from the desired output, the neural network 32 controls the iris 24 in an appropriate manner according to the characteristics appearing in the output of the histogram circuit 30. can do. Moreover, by using the histogram, the pattern of the data given to the neural network 32 is reduced as compared with the case where the histogram is not used, and there is an effect that the burden on the neural network 32 in discriminating the luminance pattern can be reduced.

FIG. 8 is a block diagram of the auto iris circuit 6 according to the third embodiment of the present invention. This auto iris circuit 6 is provided with the lens 2 as in the first and second embodiments.
Camera integrated V with 6, iris 24 and CCD 10
It is used in TR to control the iris 24 based on the luminance signal output from the CCD 10.

Referring to FIG. 8, auto iris circuit 6
Is an A / D conversion circuit 28 connected to the output of the CCD 10 for A / D converting the luminance signal output from the CCD 10.
And the luminance signal converted into a digital signal by the A / D conversion circuit 28, m × n of the imaging region 82 shown in FIG.
A block integration circuit 12 for calculating the luminance value of each block by integrating each of the equally divided blocks.
And an input is connected to the output of the block integration circuit 12, and based on the brightness value of each block, a large weight is given to the brightness value of each block near the center, as in the conventional center-weighted photometry method. And a center-weighted photometry circuit 34 for outputting a photometry signal for controlling the iris 24.

The auto iris circuit 6 further includes an average value circuit 18 for calculating the average value of the luminance values of the m × n blocks output from the block integration circuit 12, and m × n output from the block integration circuit 12. Standardization circuit 16 for standardizing each luminance value of each block by the average value given from the average value circuit 18, and standardization of each m × n block output from the normalization circuit 16 Luminance values are classified according to their levels, blocks belonging to each brightness level are classified using a histogram, and m × n output from the histogram circuit 30 for outputting frequency distribution information and the normalization circuit 16 Intensity value of each block
The blocks are grouped into several areas according to their positions, and the area division circuit 34 for outputting each area in the form of an integrated luminance value and the outputs of the histogram circuit 30 and the area division circuit 34 are input. A neural network 36, a normalization circuit 20 for normalizing the output of the neural network 36 with the output of the center-weighted photometry circuit 34, a reference luminance generation circuit 38 for generating a reference luminance signal, and a reference luminance generation circuit 38. The voltage level of the reference luminance signal is compared with the voltage level of the signal supplied from the normalization circuit 20, and the voltage level of the signal output from the normalization circuit 20 is equal to the voltage level of the reference luminance signal based on the comparison result. A comparator 22 for controlling the iris 24 to change in direction.

The histogram circuit 30 includes a normalization circuit 16
The standard value of each block standardized by is classified into p stages (p ≧ 1) according to the luminance level, and the frequency distribution information of each stage is given to the neural network 36. The area dividing circuit 34 is for classifying each of the m × n blocks so as to belong to one of the q areas, and collectively providing luminance data for each area to the neural network 36. q is 1 ≦ q
It is selected so as to satisfy ≦ m × n.

Referring to FIG. 9, the neural network 36 includes a histogram circuit 30 and an area dividing circuit 3.
4 includes the output of the input layer 64 as an input, an intermediate layer 66 that receives the output of the input layer 64 as an input, and an output layer 68 that receives the output of the intermediate layer 66 as an input.

The input layer 64 has p neurons 70 (H ₁ ...
H _p ), q neurons 72 (A _{1 to} A _q ) each receiving the output of the area dividing circuit 34, and an inhibitory neuron 74 having no input and always outputting −1.
Including and That is, the input layer 64 includes p + q + 1 neurons, and the number of inputs is p + q. this house,
A signal representing the frequency of the block belonging to the i-th luminance level is input from the histogram circuit 30 to the i-th neuron 70 (Hi). Further, the j-th neuron 72 (A _j ) receives the integrated signal of the luminance values of the blocks classified into the j-th area from the area dividing circuit 34.

Each of the intermediate layers 66 has a p of the input layer 64.
It includes R neurons 76 connected to + q + 1 neurons and inhibitory neurons 78 having no inputs. That is, the middle layer 66 includes R + 1 neurons. Of these, each of the neurons 76 is connected to all of the neurons 70, 72, 74 of the input layer 64. Then, each neuron 76 performs a predetermined calculation after weighting the input values obtained by learning in advance, and gives one output to the output layer 68.

The output layer 68 includes one neuron 80. The neuron 80 receives the outputs of all the R + 1 neurons 76 and 78 included in the hidden layer 66 as its inputs. The neuron 80 outputs a single signal by weighting the R + 1 inputs in advance by learning and performing a predetermined calculation.

Neural network 3 shown in FIG.
The neuron used in No. 6 is of the same kind as the neuron 48 used in the first embodiment shown in FIG. The first non-linear transformation performed by each neuron is
The sigmoid function as shown in FIG. 10 is used as in the embodiment of FIG.

The auto iris circuit of the third embodiment operates as follows. The neural network 36 changes the weighting coefficient in each neuron based on the difference between the output value of the neural network 36 and the desired output value when the histogram classified data and the data obtained by the area division are input in advance. This learning process is repeated for various input patterns. Upon completion of this learning, the neural network 36 outputs desired output values to various patterns of the inputted histogram classified data and the data obtained by the area division, even if the patterns are other than the learned patterns. Each weighting coefficient is adjusted so that it can be obtained, and as a result, it is possible to obtain a desired output value for various input patterns.

The optical image of the subject formed by the lens 26 is C
It is converted into a luminance signal by the CD 10 and given to a video signal processing circuit (not shown). The luminance signal is also converted into a digital signal by the A / D conversion circuit 28 and given to the block integration circuit 12. The block integration circuit 12 is
The image pickup screen of the CCD 10 is divided into m × n blocks as shown in FIG. 4, and the input digital luminance signal is integrated for each block to obtain and output the luminance value of each block. .. The brightness value of each block is given to the center-weighted photometry circuit 34,
Outputs the photometric signal to the normalization circuit 20 by performing the same processing as the conventional center-weighted photometry method such that the central block is heavily weighted with respect to the luminance value of each block. ..

On the other hand, the average value circuit 18 obtains the average value of the brightness values of the m × n blocks output from the block integration circuit 12, and supplies the average value to the normalization circuit 16. Normalization circuit 16
Is used to normalize the brightness values of m × n blocks given from the block integration circuit 12 using the average value given from the average value circuit 18, and give them to the histogram circuit 30 and the area division circuit 34.

The histogram circuit 30 includes a normalization circuit 16
The standardized luminance values of m × n blocks, which are given by, are classified into p levels based on the luminance level,
A signal (frequency distribution information) representing the number of blocks belonging to each stage is given to the neural network 36 using the histogram. The area dividing circuit 34 has a standardized m
The brightness values of the × n blocks are represented by q
The luminance values are integrated for each area in accordance with which of the areas, and q data is given to the neural network 36.

The neural network 36 performs an operation on the pattern of the histogram-classified data given by the histogram circuit 30 and the data obtained by the area division given by the area division circuit 34 according to the weighting obtained in advance by learning. Then, one output is obtained and this is given to the normalization circuit 20. Normalization circuit 2
In the case of 0, the output of the neural network 36 is standardized by the output from the center-weighted photometry circuit 34 and output.

The comparator 22 compares the voltage level of the reference luminance signal supplied from the reference luminance generation circuit 38 with the normalization circuit 20.
The iris 24 is controlled so that the voltage level of the signal output from the normalization circuit 20 changes in the same direction as the voltage level of the reference luminance signal based on the comparison result.

In the auto iris circuit shown in the third embodiment, the main iris 24 for controlling the iris 24 by the A / D conversion circuit 28, the block integration circuit 12, the center-weighted photometry circuit 34 and the comparator 22. A loop is formed. The control of the iris 24 by this main loop is the same as that of the conventional center-weighted photometry method.
However, as described above, depending on the brightness distribution of the object, the conventional method alone may not be able to properly control the iris 24 with only the main loop.
The average value circuit 18, the normalization circuit 16, the histogram circuit 30, the area division circuit 34, and the neural network 36 are for correcting the iris control when the main loop alone cannot control the iris control. Configure an auxiliary loop.

Since the brightness value of each block is standardized by the average value circuit 18 and the normalization circuit 16, the input pattern given to the neural network 36 is unlikely to change with a change in the brightness level, and the neural network 36 is less likely to change. The burden on In addition, the normalized brightness value is used as it is in the neural network 3.
Instead of inputting 6, the luminance value data classified and integrated for each area by the frequency distribution by histogram classification and area division is given to the neural network 36.
Therefore, the information of the distribution of the brightness level and the position of the brightness level is given to the neural network as the characteristic information of the input pattern, so that the input pattern can be easily separated by the neural network 36.

As described above, according to the third embodiment,
Even for an object for which sufficient iris control is difficult only with the main loop, correction can be performed with an auxiliary loop using a neural network, and proper iris control can be performed.

In the third embodiment, the center-weighted photometry circuit is used as the circuit forming the main loop. However, not only the center-weighted photometry circuit but also the lower-weighted photometry system as described in the conventional technique is used. A metering method may be used, or another method of weighted metering method may be used.

FIG. 11 is a block diagram showing a part of an auto-iris circuit 3 according to the fourth embodiment of the present invention and a camera-integrated VTR including the auto-iris circuit 3. The auto iris circuit 3 of the fourth embodiment is the first embodiment of the present invention.
The auto iris circuit 2 of the embodiment is applied to the conventional auto iris circuit 91 shown in FIG.

Referring to FIG. 11, the auto iris circuit 3 is for controlling the iris 24 and the AGC circuit 11 based on the brightness signal output from the CCD 10 and amplified by the AGC circuit 11 to a predetermined brightness level. Is. This auto iris circuit 3 has the same structure as the auto iris circuit 2 of the first embodiment shown in FIG.
The / D conversion circuit 28, the m × n block integration circuit 12,
It includes an average value circuit 14, a brightness value normalizing circuit 16, a neural network 18, an output normalizing circuit 20, a reference brightness generating circuit 38, and a comparator 22. This 4th
The auto iris circuit 3 of the embodiment is different from the auto iris circuit 2 of the first embodiment in that the input of the A / D conversion circuit 28 is connected to the output of the AGC circuit 11 instead of the output of the CCD 10. A signal for controlling the AGC circuit 11 by comparing the reference luminance generation circuit 39 for generating the second reference luminance signal with the reference luminance signal output from the reference luminance generation circuit 39 and the output of the normalization circuit 20. Is further included.

The auto iris circuit 3 of the fourth embodiment
It is assumed that the neural network 18 of (3) has finished learning by previously providing some luminance patterns (input patterns) and the corresponding desired iris control signals and AGC control signals as teacher signals. At this time, since the AGC circuit 11 operates only when the desired luminance signal is not obtained even if the iris 24 is opened, the teacher signal given to the neural network 18 is
One teacher signal is sufficient for one input pattern.

As described above, in the learning process, the strength of the connection between the neurons is changed by using the error between the input pattern and the desired output value. By repeating this for a plurality of brightness patterns, each brightness pattern is separated. After the learning is completed, the neural network 18 determines the desired iris control signal and AGC for the luminance pattern which is not learned.
The strength of the coupling between the neurons is adjusted so that the signal control signal can be output. That is, the neural network 18 becomes
This means that the input luminance pattern is classified into several categories, and even when a luminance pattern that has not been learned is input, it is possible to classify this input pattern into a certain category and obtain a desired output value. become.

In FIGS. 11 and 1, the same parts are given the same reference numerals and names. Their functions are also the same. Therefore, detailed description thereof will not be repeated here.

In the auto iris circuit 3 of the fourth embodiment shown in FIG. 11, the operation of the part for controlling the iris 24 is the same as that of the first embodiment. Therefore, it is not repeated here.

When the desired luminance signal level cannot be obtained only by controlling the iris 24, the AGC circuit 11 is controlled by the auto iris circuit 3 to operate to raise the luminance signal level. This operation will be described below. The comparator 23 compares the output of the normalization circuit 20 with the reference brightness signal supplied from the reference brightness generation circuit 39, and changes the voltage from the normalization circuit 20 so that the voltage becomes equal to the reference brightness signal. To control the AGC circuit 11. Therefore, even if the iris 24 is opened, the CCD 10
When the luminance signal output from the AGC circuit 11 has not reached a predetermined level, the AGC circuit 11 is controlled by the comparator 23 and amplifies the luminance signal to a desired level.

As described above, in the auto iris circuit 3 of the fourth embodiment, each part on the screen is not weighted by its position on the screen. Instead, the screen is divided into a plurality of blocks, the brightness of each block is standardized by the average value of the brightness of all blocks, and given to the neural network, and the iris control and the AGC control are performed according to the output of the neural network. Therefore, it is possible to correct the iris control signal at the time of backlighting or when spot light is incident, which is difficult with the conventional technique. Further, since the correction of the AGC control signal is also performed by the auto iris circuit 3, a luminance signal of a desired luminance level can be obtained even in a subject having low illuminance such as a dark place in the fourth embodiment. Also in the fourth embodiment, the input value and the output value of the neural network are standardized by the average value of the brightness of all blocks. Therefore, even if the brightness level output from the CCD 10 changes, the change in the pattern of the input data given to the neural network 18 is small. Therefore, there is an effect that the load for the pattern separation applied to the input neural network 18 can be reduced even when the brightness level changes.

FIG. 12 is a block diagram of a part of an auto-iris circuit 5 and a camera-integrated VTR using the auto-iris circuit 5 according to the fifth embodiment of the present invention. The auto iris circuit 5 of the fifth embodiment is an application of the auto iris circuit 4 of the second embodiment shown in FIG. 5 to a conventional auto iris circuit 91 shown in FIG. The auto iris circuit 5 of the fifth embodiment
Is similar to the auto iris circuit 4 shown in FIG.
The / D conversion circuit 28, the m × n block integration circuit 12,
It includes an average value circuit 14, a brightness value normalization circuit 16, a histogram circuit 30, a neural network 32, an output normalization circuit 20, a reference brightness generation circuit 38, and a comparator 22. Furthermore, this auto iris circuit 5
Similarly to the auto iris circuit 3 shown in FIG. 11, the second
Reference luminance generation circuit 39 and second reference luminance generation circuit 3
Second reference luminance signal output from the output terminal 9 and output normalization circuit 2
0 output is compared, and based on the comparison result, the AGC circuit 1
And a comparator 23 for outputting a signal for controlling 1. The input of the A / D conversion circuit 28 is the CCD 10
Is connected to the output of the AGC circuit 11 instead of the output of.

The auto iris circuit 5 of the fifth embodiment
The operation of the part that controls the iris 24 is
This is similar to that of the second embodiment already described. Also,
The operation of the part that controls the AGC circuit 11 based on the output of the output normalization circuit 20 is the same as that of the fourth embodiment already described with reference to FIG. Therefore, detailed description thereof will not be repeated here.

According to the auto iris circuit 5 of the fifth embodiment, the image pickup screen 82 of the CCD 10 (see FIG. 4).
Is divided into a plurality of m × n blocks, which are not weighted by position, and the brightness of each block is obtained, and the iris is controlled based on the brightness. There is no fear that the iris 24 is inappropriately controlled only according to the position of the high-intensity part in the subject as in the conventional case.

Further, since the brightness value of each block obtained by the block integration circuit 12 is once standardized by using the average value of the brightness values of all the blocks, only the brightness level is classified by the histogram circuit 30. Irrespective of the change in the luminance distribution pattern, a histogram having a characteristic shape can be obtained. If the classification result by the histogram is input to the neural network 32 for various input luminance patterns and the neural network 32 is learned by the error between the output and the desired output, the neural network 32 outputs the output of the histogram circuit 30. The iris 24 can be controlled in an appropriate manner according to the characteristics of the brightness distribution appearing in the. Moreover, by using the histogram as the input data to the neural network 32, the pattern of the input data is reduced as compared with the case where the histogram is not used, and the neural network 32 is used in determining the brightness pattern.
This has the effect of reducing the burden on workers.

Further, in the fifth embodiment, the AGC is further added.
By correcting the control of the circuit 11 also using the output of the neural network 32, a brightness signal having an appropriate brightness level is obtained as an output of the AGC circuit 11 even for a brightness pattern of a subject having low illuminance such as a dark place. be able to.

FIG. 13 is a block diagram of a part of an auto-iris circuit 7 according to the sixth embodiment of the present invention and a camera-integrated VTR using the auto-iris circuit 7.
The auto iris circuit 7 of the sixth embodiment is similar to the auto iris circuit 6 of the third embodiment shown in FIG.
This is an embodiment applied to the conventional auto iris circuit 91 shown in FIG.

The auto iris circuit 7 is similar to the auto iris circuit 6 of FIG. 8 in that it has an A / D conversion circuit 28 and m × n.
The block integration circuit 12, the center-weighted photometry circuit 34, the average value circuit 18, the luminance value normalization circuit 16, the histogram circuit 30, the area division circuit 34, the neural network 36, and the output normalization circuit 20. And a reference luminance generation circuit 38 and a comparator 22. Since the auto iris circuit 7 further generates the second reference luminance signal, like the auto iris circuit 3 of the fourth embodiment shown in FIG. 11 and the auto iris circuit 5 of the fifth embodiment shown in FIG. The second reference luminance generation circuit 39 of
The voltage level of the second reference brightness signal provided from the second reference brightness generation circuit 39 is compared with the voltage level of the signal output by the output normalization circuit 20, and based on the comparison result,
The comparator 23 further includes a comparator 23 for controlling the AGC circuit 11 so that the voltage level of the signal output from the normalization circuit 20 changes in the same direction as the voltage level of the second reference luminance signal. The input of the A / D conversion circuit 28 is the CCD 10
Is connected to the output of the AGC circuit 11 instead of the output of.

In FIG. 13 and FIGS. 8, 11, and 12, the same parts are given the same reference numerals and names. Their functions are also the same. Therefore, detailed description thereof will not be repeated here.

The operation of the part for controlling the iris 24 in the auto iris circuit 7 of the sixth embodiment is the same as that of the auto iris circuit 6 already described with reference to FIG. The operation of the part that controls the AGC circuit 11 is similar to that of the auto iris circuit 3 of the fourth embodiment already described with reference to FIG. Therefore, detailed description thereof will not be repeated here.

In the auto iris circuit 7 shown in the sixth embodiment, the A / D conversion circuit 28, the m × n block integration circuit 12, the center-weighted photometry circuit 34, the comparator 22, and the comparator 23. And form a main loop for controlling the iris 24 and the AGC circuit 11. Iris 24 and AG by this main loop
The control itself of the C circuit 11 is the same as that of the conventional center-weighted photometry method.

However, as described above, with only such a main loop, depending on the brightness distribution of the subject, the iris 2
4 and the AGC circuit 11 cannot be controlled properly. In order to properly correct this control, the average value circuit 18
And luminance value normalization circuit 16 and histogram circuit 30
, Area dividing circuit 34, and neural network 3
6 and 6 form an auxiliary loop for correcting the iris control and the AGC control. The output of this auxiliary loop is used to output the center-weighted photometry circuit 3 by the output normalization circuit 20.
Since the output of No. 4 is corrected, appropriate control can be performed over a wider range of the brightness distribution of the subject.

The luminance value of each block is once normalized by the output of the average value circuit 18 and the luminance value normalizing circuit 16. The input pattern given to the neural network 36 is less likely to change in response to the change in the brightness level, and the burden of pattern separation on the neural network 36 is reduced.

Further, in this embodiment, the normalized brightness value of each block is directly used in the neural network 36.
Is not entered in. The brightness values are classified into histograms by the histogram circuit 30 and given to the neural network 36 in the form of frequency distribution data. The brightness value of each block is classified for each area to which each block belongs. Then, the brightness value data is integrated for each area, and the integrated brightness value data is given to the neural network 36. The brightness level distribution and the brightness level position information are both stored in the neural network 36.
Is given as information indicating the characteristics of the input pattern. Therefore, there is an effect that the input patterns can be easily separated by the neural network 36. Therefore, even if a small neural network 36 is used, a sufficient effect can be obtained.

As described above, according to the sixth embodiment, even for an object for which the iris control and the AGC control which are insufficient with only the main loop are difficult, the auxiliary loop using the neural network 36 corrects the control. Can be done. Therefore, appropriate iris control and AGC control can be performed for a wider variety of subjects.

In the sixth embodiment, the circuit using the center-weighted photometry is shown as the circuit forming the main loop. However, the present invention is not limited to this, and the dual-weighted metering method with the lower-weighted metering method as described in the related art may be used, or another weighted metering method may be used.

[0106]

As described above, according to the first aspect of the invention, the image pickup area of the image pickup device is divided into a plurality of partial areas, and the brightness value of each partial area is related to the position on the image pickup area. Instead, it is used as input to the neural network. Therefore, the influence of the brightness distribution of the subject on the iris control is less than that in the case of using the prioritized weighted photometry method in the conventional imaging area. In addition, the brightness value of each partial area is not as it is but is standardized according to a predetermined standard and is input to the neural network, so that the pattern change of the input data is small. The load on the neural network is small, and proper iris control can be performed. As a result, it is possible to provide an auto iris circuit capable of performing an appropriate iris control for the brightness distribution of any subject.

According to the second aspect of the invention, the brightness value of each partial area is standardized by the average value of the brightness values of the partial areas and given to the neural network. The output of the neural network is also standardized by the same standard. Therefore, in addition to the effect of the invention described in claim 1, there is an effect that the pattern change of the input data due to only the change of the brightness level is small and the load on the neural network is reduced.

According to the third aspect of the present invention, the image pickup area of the image pickup device is divided into a plurality of partial areas, and the luminance value obtained for each partial area is input to the neural network and the iris is output from the neural network. Is controlled. Therefore, unlike the conventional weighted metering system, there is less risk that the iris control is inappropriately performed when the luminance distribution becomes irregular. Further, the luminance value of each partial area is subjected to histogram classification, and the frequency distribution information is input to the neural network. Since the histogram classification is performed, the features appearing in the brightness distribution depending on the brightness pattern of the subject can be clearly extracted, and the load on the neural network in performing the iris control on the brightness pattern can be reduced. As a result, it is possible to provide an auto iris circuit that can appropriately perform iris control for various luminance patterns of a subject.

According to the fourth aspect of the present invention, the image pickup area is divided into a plurality of partial areas, and then the basic iris control is performed by using the same weighted metering method as the conventional one. Further, based on the brightness value of each partial area, the iris control by the weighted photometry method can be corrected by using the neural network. Therefore, it is possible to perform appropriate iris control even for a luminance pattern of a subject that could not be sufficiently controlled by the conventional weighted metering method. The normalized luminance values of each partial area are classified into histograms, and the results are input to the neural network, so that the features of the luminance pattern are clearly extracted by the histograms. The load for correcting the iris control by the neural network is small. Similarly, the standardized luminance values of the partial areas are integrated for each partial area group to which the partial areas belong and are input to the neural network in a collected form. It is possible to extract the feature due to the position change of the brightness distribution and input it to the neural network, and the burden on the neural network when correcting the iris control can be reduced. Therefore, it is possible to perform sufficient iris control using a small neural network. As a result, it is possible to provide an auto iris circuit capable of performing accurate iris control even for various luminance patterns of a subject.

According to the fifth aspect of the present invention, the image pickup area of the image pickup device is divided into a plurality of partial areas, and the brightness value of each partial area is input to the neural network regardless of the position on the image pickup area. I am trying. Therefore, the influence on the iris control and the control of the automatic gain control circuit, which is affected by the luminance distribution of the subject, is less than that in the case of using the prioritized light metering method based on the position on the imaging area as in the related art. Also,
The luminance value of each partial area is not directly changed but is standardized according to a predetermined standard and is input to the neural network, so that the pattern change of the input data is small.
The load on the neural network is small, and the iris and the automatic gain control circuit can be controlled appropriately over a wide range of brightness levels.

According to the sixth aspect of the present invention, the brightness value of each partial area is standardized by the average value of the brightness values of the partial areas and given to the neural network. The output of the neural network is also standardized by the same standard. Therefore, in addition to the effect of the invention described in claim 1, the change of the pattern of the input data due to the change of only the brightness level is small, and the iris and the automatic gain control circuit can be controlled while reducing the load on the neural network. There is. As a result, it is possible to provide an auto iris circuit capable of properly controlling the iris and the automatic gain control circuit over various brightness patterns of the subject over a wide range of brightness levels.

According to the invention described in claim 7, the image pickup area of the image pickup device is divided into a plurality of partial areas, and the brightness value obtained for each partial area is input to the neural network and the iris is output from the output. And the automatic gain control circuit is controlled. Therefore, unlike the conventional weighted metering system, there is less risk that the iris and the automatic gain control circuit will be improperly controlled when the luminance distribution becomes irregular. Also,
The luminance value of each partial area is subjected to histogram classification, and the frequency distribution information is input to the neural network. Since the histogram classification is performed, the features appearing in the luminance pattern of the subject can be clearly extracted, and the load on the neural network in controlling the iris and the automatic gain control circuit for the luminance pattern can be reduced. As a result, it is possible to provide an auto iris circuit capable of properly controlling the iris and the automatic gain control circuit over various brightness patterns of the subject over a wide range of brightness levels.

According to the eighth aspect of the present invention, after the image pickup area is divided into a plurality of partial areas, the basic iris and automatic gain control circuit can be controlled by using the same weighted metering method as the conventional one. Done. Further, based on the brightness value of each partial area, it is possible to correct the control of the iris and the automatic gain control circuit by the weighted photometry method using the neural network. Therefore, it is possible to properly control the iris and the automatic gain control circuit even for a luminance pattern of a subject which could not be sufficiently controlled by the conventional weighted metering method. The normalized luminance values of each partial area are classified into histograms, and the results are input to the neural network, so that the features of the luminance pattern are clearly extracted by the histograms. The burden for correcting the control of the iris and the automatic gain control circuit by the neural network is small. Similarly, the standardized luminance values of the partial areas are integrated for each partial area group to which the partial areas belong and are input to the neural network in a collected form. Therefore, it is possible to extract the feature due to the position change of the luminance distribution and input it to the neural network, and it is possible to reduce the burden on the neural network when correcting the control of the iris and the automatic gain control circuit. Therefore, it is possible to sufficiently control the iris and the automatic gain control circuit using a small neural network. As a result, it is possible to provide an auto iris circuit capable of accurately controlling the iris and the automatic gain control circuit for various brightness patterns of a subject over a wide range of brightness levels.

[Brief description of drawings]

FIG. 1 is a block diagram showing an auto iris circuit and its peripheral portion according to a first embodiment of the present invention.

FIG. 2 is a block diagram of the neural network shown in FIG.

FIG. 3 is a diagram of a neuron model of the neural network shown in FIG.

FIG. 4 is a diagram schematically showing block division of an imaging region according to an embodiment of the present invention.

FIG. 5 is a block diagram showing an iris control circuit and its periphery according to a second embodiment of the present invention.

FIG. 6 is a block diagram of the neural network shown in FIG.

7 is a diagram schematically showing a neuron model used in the neural network shown in FIG.

FIG. 8 is a block diagram showing an auto iris circuit and its periphery according to a third embodiment of the present invention.

FIG. 9 is a block diagram of the neural network shown in FIG.

FIG. 10 is a graph of an example of a sigmoid function.

FIG. 11 is a block diagram showing an auto iris circuit and its periphery according to a fourth embodiment of the present invention.

FIG. 12 is a block diagram showing an auto iris circuit and its periphery according to a fifth embodiment of the present invention.

FIG. 13 is a block diagram showing an auto iris circuit and its periphery according to a sixth embodiment of the present invention.

FIG. 14 is a block diagram of a conventional auto iris circuit and its surroundings.

FIG. 15 is a block diagram of a conventional auto iris circuit that also controls the AGC circuit and its surroundings.

[Explanation of symbols]

 2-7 Auto iris circuit 10 CCD 11 AGC circuit 12 Block integration circuit 14 Average value circuit 16 Normalization circuit 18, 32, 36 Neural network 20 Normalization circuit 22, 23 Comparator 24 Iris 26 Lens 28 A / D conversion circuit 30 Histogram Circuit 34 Area division circuit

Claims (8)

[Claims] 1. An auto iris circuit for controlling the iris so as to adjust the output of the image pickup device to a desired level in response to a luminance signal output from the image pickup device whose incident light amount is regulated by the iris. And a standardized brightness value acquisition unit for acquiring a standardized brightness value according to a predetermined standard for each of a plurality of predetermined partial areas of the imaging area of the imaging device based on the brightness signal. An auto-iris circuit including: a neural network having the standardized luminance value as an input; and an iris control unit for controlling the iris in response to an output of the neural network. 2. The normalized brightness value acquisition means includes a brightness value acquisition means for acquiring a brightness value of each of the partial areas based on the brightness signal, and the partial area output by the brightness value acquisition means. A brightness value averaging means for obtaining an average value of the brightness values, and a brightness value of each of the partial regions, including a brightness value normalizing means for normalizing based on the output of the brightness value averaging means, The iris control means compares the output of the neural network with an output normalizing means for normalizing the output of the luminance value averaging means, and compares the output of the output normalizing means with a predetermined reference signal, The auto iris circuit according to claim 1, further comprising: a comparison unit configured to control the iris based on a comparison result. 3. Based on the standardized brightness value of the partial area output from the standardized brightness value acquisition means, each of the partial areas is classified into one of a plurality of predetermined brightness sections, and the frequency is classified. The auto iris circuit according to claim 1, further comprising a classifying unit for outputting distribution information, wherein the neural network receives an output of the classifying unit as an input. 4. The image pickup device has an image pickup region divided into one or a plurality of partial region groups, each of which includes one or a plurality of the partial regions, and the auto iris circuit has the normalized luminance value. The brightness value of each of the partial areas outputted by the acquisition means is further classified into each of the partial area groups, and further includes brightness value integrating means for outputting one integrated brightness value for each of the partial area groups, The neural network receives the output of the brightness value integrating means in addition to the output of the classifying means, and the iris control means divides the imaging area into a plurality of predetermined photometric areas,
A weighted photometric signal output unit for outputting a predetermined weighted photometric signal by weighting the luminance signal for each of the photometric regions with a predetermined coefficient, and the output of the neural network is standardized by the weighted photometric signal. 4. The auto iris circuit according to claim 3, further comprising: output normalizing means for comparing the output of the output normalizing means with a predetermined reference signal, and comparing means for controlling the iris based on the comparison result. . 5. The output of the image pickup device is adjusted to a desired level in response to a luminance signal output from the image pickup device whose incident light amount is regulated by an iris and whose level is adjusted by an automatic gain control circuit. An auto iris circuit for controlling an iris and the automatic gain control circuit, which is standardized according to a predetermined standard for each of a plurality of predetermined partial areas of an imaging area of the imaging device based on the luminance signal. Standardized brightness value acquisition means for acquiring the standardized brightness value, a neural network that receives the standardized brightness value, and the iris and the automatic gain control circuit in response to the output of the neural network. An auto iris circuit including control means for controlling. 6. The normalized brightness value acquisition means includes a brightness value acquisition means for acquiring a brightness value of each of the partial areas based on the brightness signal, and the partial area output by the brightness value acquisition means. A brightness value averaging means for obtaining an average value of the brightness values, and a brightness value of each of the partial regions, including a brightness value normalizing means for normalizing based on the output of the brightness value averaging means, The control means is an output normalizing means for normalizing the output of the neural network based on the output of the luminance value averaging means, and the output of the output normalizing means is a first reference signal which is predetermined. First comparing means for comparing and controlling the iris based on the comparison result, the output of the output normalizing means is compared with a predetermined second reference signal, and the automatic gain based on the comparison result. control And a second comparison means for controlling the road, auto iris circuit of claim 5. 7. Based on the standardized brightness value of the partial area output from the standardized brightness value acquisition means, each partial area is classified into one of a plurality of predetermined brightness sections, and the frequency is classified. The auto iris circuit according to claim 5, further comprising a classifying unit for outputting distribution information, wherein the neural network receives an output of the classifying unit as an input. 8. The image pickup device has an image pickup region divided into one or a plurality of partial region groups each including one or a plurality of the partial regions, and the auto iris circuit is configured to provide the normalized luminance value. The brightness value of each of the partial areas outputted by the acquisition means is further classified into each of the partial area groups, and further includes brightness value integrating means for outputting one integrated brightness value for each of the partial area groups, The neural network receives the output of the brightness value integrating means in addition to the output of the classifying means, and the control means divides the imaging area into a plurality of predetermined photometric areas,
A weighted photometric signal output unit for outputting a predetermined weighted photometric signal by weighting the luminance signal for each of the photometric regions with a predetermined coefficient, and the output of the neural network is standardized by the weighted photometric signal. An output normalizing means for comparing the output of the output normalizing means with a predetermined first reference signal, and a first comparing means for controlling the iris based on a comparison result; and the output normalizing 8. An auto-iris circuit according to claim 7, including second comparing means for comparing the output of the means with a predetermined second reference signal and controlling the automatic gain control circuit based on the comparison result.