JPH1075384A - Signal compression circuit - Google Patents

Abstract

(57) [Summary] A knee characteristic is smoothly approximated by a broken line using a third straight line. Kind Code: A1 A first input / output characteristic straight line having a slope a of 1 so that an input level is output as it is, and a second input / output characteristic straight line having an input level compressed and output by a slope b. The knee characteristic is approximated by a broken line between Lb and a third input / output characteristic line Lc that contacts with a slope c so as to straddle between the first and second input / output characteristic lines. Third straight line Lc
Is selected such that C = (1 + b) / 2. With this inclination, the knee characteristic is smoothly approximated by a polygonal line, so that harmonic components generated at the knee point are reduced, and it is possible to prevent image quality from being degraded due to aliasing. When the inclination c is selected, the hardware scale for obtaining a polygonal line approximation output can be reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a signal compression circuit suitable for application to a knee circuit (knee) provided in a digital video camera device. More specifically, the present invention relates to a signal compression circuit in which when a nonlinear compression characteristic is approximated by a plurality of polygonal lines to give a predetermined nonlinear compression characteristic, signal degradation at a break point is prevented by gradually joining the polygonal lines.

[0002]

2. Description of the Related Art In a video camera apparatus or the like, a signal compression circuit for compressing an input signal to provide a gamma characteristic or a knee characteristic is used. For example, in the knee characteristic, signal compression processing is performed because it is necessary to keep a wide dynamic range of natural light within a narrow dynamic range for an imaging signal (video signal). This signal compression process focuses on the fact that the higher the luminance of a person becomes, the lower the visibility becomes, and compresses the signal level of the high luminance portion.

For example, as shown in FIG. 12, when a level K0 (= p), which is a high luminance input, is determined as a compression start point (knee point), the input signal level is directly used as an output signal level up to the level K0. For an input x equal to or greater than K0, the signal is compressed as input / output characteristics having a predetermined slope.

When an image signal to be handled is an analog signal, since a diode or the like is used in the signal compression processing system, a smooth curve as shown by a broken line in FIG. However, in the case of a digital signal, the output level changes greatly before and after the knee point p, and a harmonic component is generated near the knee point p, which is the level change point. It is known that this harmonic component is turned back to the baseband component side of the imaging signal, and the image quality is degraded due to this.

In order to improve the image quality deterioration, a straight line L having a third input characteristic is compared with a straight line La having a first input / output characteristic and a straight line Lb having a second input characteristic as shown in FIG.
c is used as a section for cornering.

Although the knee point is increased to two by the straight line Lc, the intersection angle between the straight lines La and Lc and the straight line Lc
Since the intersection angle between c and Lb is smaller than the intersection angle at the knee point p, the generation of harmonic components is reduced, and the aliasing component to the baseband component is reduced accordingly, so that deterioration of image quality can be prevented. Become.

In order to realize a polygonal line approximation using the third straight line Lc, if the slopes of the straight lines are a, b, and c, y = ya = ax · x = x Kp1 until x <K1. Up to ≤x≤Kp2, y = yc = cx + (1+
c) If K1 x ≦ Kp2 or more, y = yb = b × x + (1 + b) K
It becomes 2.

In order to realize such a polygonal line approximation, for example, a technique disclosed in Japanese Patent Application Laid-Open No. Hei 4-23569 can be used. FIG. 14 shows the specific configuration. In the figure, an input signal (digital imaging signal) is applied to a terminal 11.
x is supplied and supplied to the minimum value selection circuit 19 and to the first and second calculation means 12 and 13.

The first calculating means 12 is a means for realizing the input / output characteristic yc of the straight line Lc.
Reference numeral 3 denotes means for realizing the input / output characteristics yb of the straight line Lb. The first calculating means 12 multiplies the input signal x by the gradient c, and outputs the multiplied output and (1 + c)
And an adder 15 for adding K1.

Similarly, the second calculating means 13 comprises a multiplier 17 for multiplying the input signal x by a gradient b, and an adder 18 for adding this multiplied output to (1 + b) K2. . By selecting the minimum value among these outputs ya, yc and yb by the minimum value selection circuit 19 at the next stage, a knee characteristic curve approximated by a broken line is obtained at the terminal 20 as shown in FIG. Will be.

[0011]

By the way, the image pickup signal is usually subjected to various signal processings as three primary color signals of R, G and B, so that the same three signal processing systems are required in total. Three knee circuits are also required. On the other hand, as shown in FIG. 14, one knee circuit 10 includes at least two multipliers 1
Since 4 and 17 are required, the number of multipliers used in the entire camera apparatus becomes large. This increases the circuit scale and increases the cost.

Further, in the case of FIG. 13, the slope c of the third straight line Lc is mostly determined by an empirical rule. For example, when b = 0.3, twice as many times, c = 0.6 is often selected.

[0013] Depending on the subject, it may be necessary to emphasize the gradation or the dynamic range. In such a case, if the value of the gradient b is selected to be a different value, as the value departs from the value of 0.3, the intersection angles between the straight lines Lc and La and between Lc and Lb become uneven, and appropriate squaring is performed. Has become difficult to achieve.
As a result, the harmonic components cannot be effectively suppressed, and a problem arises that the deterioration of the image quality due to the harmonic components cannot be sufficiently improved.

The present invention has been made to solve such a conventional problem, and proposes a signal compression circuit capable of sufficiently suppressing harmonic components in spite of a simple circuit configuration. .

[0015]

According to the present invention, there is provided a signal compression circuit for digitizing an input signal and subjecting the digitized input image signal to signal compression. A first input / output characteristic in which the slope a is 1 such that the level is output as it is, a second input / output characteristic in which the input level is compressed and output by the slope b, and the first and second characteristics. A signal compression circuit having a third input / output characteristic which is in contact with the input / output characteristic at a gradient c so as to straddle the input / output characteristic of the signal compression circuit, wherein the gradient c is approximately c = (1 + b) / 2 The third input / output characteristic is selected.

In the present invention, since the inclination c of the third straight line Lc is selected as described above, the third straight line Lc is set to the first straight line Lc.
Points q and r that intersect with the second straight lines La and Lb
The intersection angles at are approximately equal. When the two intersection angles are substantially equal, a smooth approximate curve can be obtained with the third straight line Lc. Therefore, generation of harmonics at a new knee point is reduced, and image quality deterioration can be improved. This relationship is the same even if the inclination b of the second straight line Lb changes.

The signal level at the knee points q and r is represented by Kp
Assuming that 1, Kp2, Kp−Kp1 = Kp2−Kp = △ Kp, and a = 1, so that until x <K1, finally, y = ya = ax · x = Kp1 ≦ x ≦ Kp2 Is y = yc = (x−Kp1) (1
+ B) / 2 + Kp1 When x ≦ Kp2 or more, y = yb = {x− (Kp1 + Kp2) / 2} b + (Kp1 + K
p2) / 2.

As is clear from the above, since the multiplication process is performed once, the circuit scale is greatly simplified.
Further, since the processing of (1 + b) / 2 is merely a bit shift processing, and the processing of (Kp1 + Kp2) / 2 requires only one addition and bit shift processing, these circuit configurations are also simple. The variables are only b, Kp1, and Kp2.

[0019]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS A case where an embodiment of a signal compression circuit according to the present invention is applied to a knee circuit provided in the above-described digital video camera device will be described in detail with reference to the drawings.

FIG. 1 shows an embodiment of a digital video camera apparatus 30 to which the present invention is applied. In this example, three solid-state imaging devices (hereinafter, a case of a CCD image sensor) are used as imaging devices, and G, G
It is assumed that the R and B primary color signals are obtained as imaging signals.

An object is three CCDs via an optical system 31.
Image sensors (hereinafter CCD) 31R, 31G, 31B
And image signals of R, G, and B are obtained from the respective images. When the imaging signals R, G, and B obtained from the CCDs 31R, 31G, and 31B are digitized, the R signal and the B signal are applied to the CCD 31G that obtains the G signal, in order to remove the aliasing component mixed in the fundamental wave component. The CCDs 31R and 31B that obtain signals are 1 /
The same subject image is captured while being relatively shifted by two pixel pitches P / 2.

The image pickup signal is output from a correlated double sampling circuit (C
DS) 32R, 32G, and 32B (hereinafter, R, G, and B are omitted) to perform waveform shaping and remove mixed nozzles. The image signals subjected to the waveform shaping processing and the like are supplied to an A / D converter 34 via an amplifier 33 and digitized.

Each digitized image signal is supplied to a single masking circuit 37 via a one-line delay circuit 36 to perform color correction processing, and then supplied to a knee circuit 10 as a signal compression circuit. Is done. Since the knee circuit 10 performs an adaptive nonlinear calculation process according to the level of the imaging signal, the A / D-converted imaging signal is supplied to a control unit 40 including a CPU. And a multiplication coefficient β are obtained, and these are supplied to the knee circuit 10 as an adaptive control signal.

On the other hand, the imaging signals of the adjacent lines derived from before and after the one-line H delay circuit 36 are sent to an image enhancer 41 for enhancing the outline of the image, and an outline emphasis signal composed of low-frequency components and a high-frequency An edge enhancement signal composed of the components is generated. These signals are, for example, G
The signal may be generated using only the signal and the R signal.

Each of the low-frequency contour emphasizing signals is combined with an image signal to which a knee characteristic has been added by an adder 42, and the image signal subjected to low-frequency compensation is supplied to a gamma correction circuit 43 and subjected to nonlinear processing.

The adder 44 adds the high-frequency augmentation signal to each of the image signals after the non-linear processing, and has broad frequency characteristics up to the high frequency. Then, after the clipping process of the signal level is performed in the clipping circuit 46, each imaging signal is supplied to the single matrix circuit 47,
In this example, a luminance signal Y and a pair of chroma signals Cr (= R−
Y) and Cb (= BY) are formed. By performing matrix processing such as addition and subtraction, the sideband components mixed in the fundamental wave components of the imaging signal under specific imaging conditions are canceled.

These imaging signals are output as digital signals to a terminal 48, and are returned to analog signals by a corresponding D / A converter 49.

Now, in the digital video camera device 30 configured as described above, the above-described knee circuit 10 is configured as follows. First, a description will be given of the broken line approximation of the knee characteristic.

FIG. 2 shows knee characteristics based on the polygonal line approximation.
The input / output characteristic of the input signal x, which is an image pickup signal, is given by a straight line La up to a certain high luminance level (knee point p). The slope a of the straight line La at this time
Since the input level becomes the output level as it is, a = 1.

A second straight line Lb crossing the knee point p
Is b. A chamfered straight line Lc is used for these two straight lines La and Lb. This inclination is defined as c.
New knee points q and r appear by using the straight line Lc. And each knee point p,
Input levels at q and r are Kp, Kp1, and Kp2 as shown in the figure.

In the case where the knee point p is chamfered by the third straight line Lc, when the inclination c of the third straight line Lc is substantially selected as c = (1 + b) / 2, as shown in FIG. The angles θ4 and θ5 intersecting the first and second straight lines La and Lb are very close values. When the two intersection angles θ4 and θ5 are approximate values, the cornering becomes smoothest. It becomes clear from the following developments.

The slope c of the straight line Lc is obtained. In FIG. 3, θ ₂ + θ ₅ = θ ₃ (1) θ ₃ + θ ₄ = θ ₁ (2) where θ ₄ = θ ₅ (3) ), (2), θ ₁ + θ ₂ = 2θ ₃ (4) Taking both sides tan, tan (θ ₁ + θ ₂ ) = tan 2θ ₃ (5) When expanded, (tan θ ₁ + tan θ) ₂ ) / (1-tan θ ₁ tan θ ₂ ) = ² tan θ ₃ / (1-tan ² θ ₃ ) (6) Here, the inclination of each straight line is as follows: a = tan θ ₁ (α> 0) b = tan θ ₂ (b> 0) c = tan θ ₃ ( c> 0) because it is (7), when this is substituted into equation (6), to be obtained (a + b) / (1 -ab) = 2c / (1-c 2) ··· (8) Is the inclination c of the straightening straight line Lc.
Equation (8) is solved for c.

(A + b) c ² +2 (1-ab) c− (a + b) = 0 (9)

[0034]

(Equation 1)

Here, a + b = m 1−ab = n (11)

[0036]

(Equation 2)

Since a> 0, b> 0 and c> 0,

[0038]

(Equation 3)

## EQU1 ##

That is, the straight lines La and L of the slope a and the slope b
The two straight lines La and Lb are obtained by squaring the polygonal line constituted by b with a straight line Lc having a slope c obtained by Expression (13).
Can be connected smoothly.

In a signal range where the image pickup signal level is low, a = 1 is set in order to obtain a linear characteristic.

Thus, equation (13) is

[0043]

(Equation 4)

Is as follows. The inclination c shown in this equation is as shown in FIG. 4 when the horizontal axis is b.

In order to realize the equations (13) and (14) by hardware and software, arithmetic processing such as square root and division is required, and the size of the hardware circuit becomes large. Therefore, these approximate expressions will be considered next.

The knee characteristic is usually used in a range of b = about 0.2 to 0.4 in many cases. When the camera is used outdoors, the dynamic range is emphasized. In this case, b = 0.2 is selected. On the contrary, when the gradation is emphasized, the value of b = 0.4 is selected. Because it is chosen.

The inclination of the approximated polygonal line is represented by c with respect to the ideal inclination of the polygonal polygonal line ci.

As an example of approximation, equation (13) is approximated by a straight line focusing on the actual use area. The purpose of this approximation is to reduce the amount of computation in hardware and software, so it is preferable that the word length be as short as possible. When the actual use area is wide, such as in a case other than knee correction, it is conceivable to switch and use the approximate expression in order to increase the approximation accuracy.

It is assumed that the slope ci of the approximated cornering straight line Lc is expressed as ci = p × b + q (15) using constants p and q as a linear combination of b. If p and q are represented by, for example, 5 bits from the graph value around 0.2 ≦ b ≦ 0.4 in equation (14), the following equation is obtained: ci = c1 = (13 + 20b) / 32 (16) Assuming that the word length is further reduced, ci = c2 = (1 + 2b) / 3 (17) ci = c ≒ (1 + b) / 2 (18)

The absolute value abs of the error between the slope ci of the approximated straight line Lc and the ideal slope of the equation (13) is shown in FIG.
Shown in In the figure, a curve c3 is an error of a conventionally used one, and the error becomes a very large error unless used around b = 0.3.

The curve c1 is the error when using equation (16). Curve c2 is the error when using equation (17).

The curve c is an approximation formula used in the present invention, and is an error when the formula (18) is used. When the curves c and c1 are compared with each other, the error becomes maximum around b = 0. In order to convert this error into an angle, it is sufficient to take the arc value of the error value, which is approximately 5 °. In actual use, such an error is sufficiently acceptable. Incidentally, since tan (1 °) = 0.017 tan (2 °) = 0.035 tan (5 °) = 0.087 (19), the expression (16) is also 1 in the actual use area. It is within the error of less than °.

Equation (17) also fits within an error of 5 ° or less.

However, when importance is placed on the circuit scale, equation (16) is most effective. This is because the hardware can be easily configured. Next, the hardware configuration will be described. First, the slope c shown in Expression (18) will be considered from another viewpoint. As shown in FIG. 3, the straight line Lc is changed to a straight line La or Lb.
Equation (20) must be satisfied in order to smoothly connect

Kp−Kp1 = Kp2−Kp (20) The amplitude transfer characteristic of FIG. 2 (FIG. 3) is expressed by the following equation.

X <Kp1, y = x (21) Kp1 ≦ x <Kp2, y = (x−Kp1) c + Kp1 (22) Kp2 ≦ x, y = (x−Kp) b + Kp (23) From equation (20), if we put ΔKp = Kp−Kp1 = Kp2−Kp (24), then c = (ΔKp + ΔKp · b) / 2ΔKp = (1 + b) / 2 (25) And becomes the same as Expression (18).

Here, by rewriting the equations (21) to (23) based on the equations (20) and (25), x <Kp1, y = x (26) Kp1 ≦ x <Kp2, y = (X−Kp1) × {(1 + b) / 2} + Kp1 (27) Kp2 ≦ x, y = {x− (Kp1 + Kp2) / 2} × b + {(Kp1 + Kp2) / 2} (( 28) FIG. 6 shows a part of the knee circuit 10 for realizing the equations (26) to (28). As shown in the figure, an input signal x is supplied to a terminal 51, a subtractor 52 that subtracts the input signal x from the first signal α, and a gradient b that determines a third input / output characteristic to the subtracted output. Multiplier 5 which multiplies the compression coefficient β by
4 and an adder 56 for adding the first signal to the multiplied output.
It is composed of

The first signal α and the multiplication coefficient β are adaptively selected for the input signal x, and the relationship is shown in FIG. (A) When the input signal x is equal to or lower than the level Kp1, an arbitrary signal level (Kp1 itself in this example) is assigned to the first signal α, and a value “1. "0" is assigned. In this example, Kp1 is set to reduce the circuit scale. (B) When the input signal is equal to or higher than level Kp1 and equal to or lower than Kp2, level Kp1 is used as first signal α, and a coefficient indicating slope c 傾 き = (1 + b) / 2｝ is used as multiplication coefficient β. (C) When the input signal is equal to or higher than the level Kp2, the first
The average level of the levels Kp1 and Kp2 is used as the signal α, and the slope b is used as the multiplication coefficient β.

Then, in the case of (a), equation (26)
Is output, and in the case of (b), equation (2)
The signal y represented by 7) is output, and in the case of (c), the signal y of Expression (28) is output.

As shown in FIG. 8, the average value calculating means 60 for the average value of the levels Kp1 and Kp2 discards the adder 64 for both levels and the lower one bit of the added output (bit shift). Bit shift means 65 for
And can be composed of The terminals 61 and 63 are input terminals, and the terminal 62 outputs the level Kp1 itself.
An averaged output (Kp1 + Kp2) / 2 is output to the terminal 66.

The calculating means 70 for calculating the inclination c can be configured as shown in FIG. First, the level “1” (integer) from the terminal 71 and the slope b from the terminal 72 are wired-ORed to form (1 + b), which is supplied to a bit shift means 73 that discards the lower one bit of the OR output. Is done. By doing so, (1+
b) / 2 output can be obtained.

The reason why the configuration can be simplified in this way is that since the slope of the knee characteristic is always smaller than 1, the operation of adding 1 can be realized only by placing 1 in an integer bit. It is assumed that the slope b is a bit assignment in which the first decimal place is the MSB.

As described above, both the average value calculating means 60 and the inclination calculating means 70 can be constituted by simple logic circuits.

The actual knee circuit 1 is calculated by using these calculating means.
FIG. 10 shows the configuration of 0. Average value calculation means 6
Either the output of 0 is selected by the first selector 80, and the output of the slope calculating means 70, the slope b itself and the signal of level “1” are supplied to the second selector 81, and the alternative processing is performed. Done.

Select signal generating means 8 for controlling these
5 includes a pair of level comparators 86 and 87,
The input signal x and the level Kp1 are supplied to the level comparator 86, and the select signal (bit for the upper bit) is a bit "0" until the input signal x exceeds the level Kp1, and becomes a bit "1" otherwise. ) Is generated.

The input signal x and the level Kp2 are given to the second level comparator 87. The bit is "0" until the input signal x exceeds the level Kp2, and the bit is "1" otherwise.
A select signal (for lower bits) is generated as follows. Therefore, both the select signal and the input signal x
Is as shown in FIG. 11, and the relationship between the select signal and the select terminal at that time is as shown in FIG.

As a result, (a) When the input signal x is equal to or lower than the level Kp1 (select signal is "00"), Kp1 is selected as the first signal α as shown in FIG. A value “1.0” that does not affect the multiplication result is selected. (B) When the input signal x is equal to or more than the level Kp1 and equal to or less than Kp2, that is, when the select signal is “10”, the level Kp1 is selected as the first signal α, and the slope c ｛= (1 + b) as the multiplication coefficient β. / 2} is selected. (C) When the input signal x is equal to or higher than the level Kp2, that is, when the select signal is “11”, the first signal α
Selects the average output of the levels Kp1 and Kp2, and selects the slope b itself as the multiplication coefficient β.

The bit assignment of the coefficient input to the multiplier 54 is such that the integer part is 1 bit and the remainder is a decimal part.
That is, only one bit of the MSB is an integer part, and the rest is a decimal part.

By configuring such a selecting means, the knee circuit 10 can be configured with a simple logic circuit without using a microcomputer such as a CPU. Therefore, even if three circuits are used, the circuit scale does not increase as a whole.

With the configuration shown in FIG. 10, it is possible to output the polygonal line approximation characteristic only by designating the two knee points Kp1 and Kp2 that define the width of the chamfering and the slope b. There is no need for complicated arithmetic processing, such as calculating a broken line approximation characteristic by a microcomputer.

Since it is known in advance that the knee point of the main knee curve is the average of Kp1 and Kp2, the imaging output state before and after the knee point can be verified, and the result (behavior) of the knee characteristic can be easily grasped. .

Although the calculation of the squaring is performed by hardware, there is obtained such a feature that the scale is extremely small.

Since the present invention proposes a signal compression circuit for performing non-linear compression processing, the non-linear processing can be applied to other than knee processing.

[0074]

As described above, according to the present invention, when performing non-linear signal compression, signal compression is performed by smooth broken line approximation.

According to this, the difference in the intersection angle with the third straight line contacting with the slope c is reduced, and the generation of the harmonic component at the intersection can be effectively suppressed. Therefore, there is a feature that the image quality deterioration due to the aliasing distortion due to the harmonic component can be improved. Further, since the present invention can be realized without increasing the circuit scale, the cost can be reduced as compared with the conventional case. Therefore, when the present invention is applied to a digital video camera device or the like, a significant cost reduction can be achieved.

[Brief description of the drawings]

FIG. 1 is a system diagram showing an embodiment in which a signal compression circuit according to the present invention is applied to a knee circuit in a digital video camera device.

FIG. 2 is an explanatory diagram of a broken line approximation of knee characteristics.

FIG. 3 is an explanatory diagram showing the details.

FIG. 4 is an explanatory diagram for calculating the inclination of a third straight line.

FIG. 5 is a characteristic diagram showing an error between an ideal inclination and an approximate inclination.

FIG. 6 is a connection diagram of a main part of a knee circuit.

FIG. 7 is a diagram illustrating a relationship among a first signal, a multiplication coefficient, and an input signal.

FIG. 8 is a specific configuration diagram of an average value calculation unit.

FIG. 9 is a specific configuration diagram of a slope calculation unit.

FIG. 10 is a connection diagram of a knee circuit according to an embodiment of the present invention.

FIG. 11 is a diagram showing a truth table of a select signal.

FIG. 12 is an explanatory diagram of a knee line approximation of knee characteristics.

FIG. 13 is an explanatory diagram of a broken line approximation of knee characteristics.

FIG. 14 is a connection diagram of a conventional knee circuit.

[Explanation of symbols]

10 ... knee circuit, 52 ... subtractor, 54 ... multiplier, 56 ... adder, 60 ... average value calculation means,
70... Inclination calculating means, 80, 81.
85 ... select signal generating means, 86, 87 ... level comparator

Claims (9)

[Claims] 1. A signal compression circuit for digitizing an input signal and performing signal compression on the digitized input image signal, wherein a first slope having an inclination a such that an input level is output as it is is 1. A second input / output characteristic in which the input level is compressed and output by the gradient b, and a gradient c so as to straddle between the first and second input / output characteristics.
Wherein the third input / output characteristic is selected such that the gradient c is approximately c = (1 + b) / 2. And a signal compression circuit. 2. An input signal level Kp at which the first and second input / output characteristic lines intersect is defined as a first bending point, and an input signal level Kp1 at which the first and third input / output characteristic lines intersect is defined as a first bending point. When the input signal level Kp2 at which the third and second input / output characteristic lines intersect is taken as a third bending point, the third input / output characteristic line becomes Kp-Kp1 = Kp2. 2. The signal compression circuit according to claim 1, wherein the signal compression circuit is set to -Kp. 3. A subtracter for subtracting an input signal from a first signal, a multiplier for multiplying the subtracted output by a slope b or the like for determining the third input / output characteristic as a compression coefficient, and 2. The signal compression circuit according to claim 1, further comprising an adder for adding a first signal, wherein the first signal is adaptively selected according to an input signal. 4. When the input signal is at or below the level Kp1, an arbitrary signal level is assigned to the first signal, "1" is assigned as the multiplication coefficient, and the input signal is at or above the level Kp1 and at most Kp2. , The level Kp1 is used as the first signal, and the coefficient indicating the slope c is used as the multiplication coefficient. When the input signal is equal to or higher than the level Kp2, the level is used as the first signal. 2. The signal compression circuit according to claim 1, wherein an average value level of Kp1 and Kp2 is used, and said gradient b is used as said multiplication coefficient. 5. The means for calculating the average value of the average value of the levels Kp1 and Kp2 comprises adders of both levels and bit shift means for bit-shifting the added output. 2. The signal compression circuit according to 1. 6. The inclination calculating means for calculating the inclination c,
2. The signal compression circuit according to claim 1, comprising a wired OR having a level 1 and a gradient b, and bit shifting means for bit shifting the OR output. 7. A first selector to which two outputs from the average value calculating means are supplied, and a second selector to which the level Kp2 and the level 1 are supplied together with the gradient c from the gradient calculating means. And select signal generating means for controlling these selectors, wherein an output from the first selector is used as the first signal and an output from the second selector is used as the multiplication coefficient. The signal compression circuit according to claim 1, wherein: 8. The select signal generating means includes a first level comparator to which the input signal and the level Kp1 are supplied, and a second level comparator to which the input signal and the level Kp2 are supplied. 8. The signal compression circuit according to claim 7, wherein a 2-bit signal obtained by combining the outputs of the level comparators is used as a select signal. 9. The apparatus according to claim 1, wherein said input signal is an image pickup signal, and a knee characteristic having a smooth polygonal curve is obtained by said first to third input / output characteristics. Signal compression circuit.