JPH0432585B2 - - Google Patents

Description

[Detailed description of the invention]

The present invention relates to restoration and reproduction processing for reproducing an analog signal of an image from a digital signal for transmission after AD conversion of an image signal representing a document to be transmitted. This invention is particularly applicable to the digital transmission of image information sent and received by a transmission method called Belin type by adding an AD conversion circuit and a DA conversion circuit.
It is not limited to this. In the Bellin-type transmission system used to date, the document to be transmitted is wound without overlapping onto a rotating cylinder and is moved on an axis parallel to the axis of this cylinder. A reader (phototube, photomultiplier tube, etc.) connected to the plate detects the reflected light of the illuminated document at the focal point of the reader. Similarly, in the receiving section, the rotary cylinder is A document is reproduced on a photosensitive support driven by. More specifically, if the transmission of the signal along the transmission line is carried out in a digital path, at the transmitting end the document to be transmitted is represented by:
The analog signal that is the output signal of the emitter is an image signal, and this image signal amplitude modulates or frequency modulates the carrier wave. The transmission band of this analog signal is limited by the phototube of the reader and the resolution of the emitter. This transmission band can be limited to 800 Hz without significant deterioration (decoupling of signals processed at 800 Hz does not result in significant degradation in the quality of the received image). After demodulation, the image signal is analyzed at frequency f. This frequency f is a function of the modulation rate φ used and the number n of quantization bits according to the following equation. f=φ/n For example, in this formula, φ=4800 baud or 9600 baud n=4 or 5 bits The frequency f represents the sampling frequency,
According to Shannon's theorem, this is too small when operating at 4800 baud and bit counts > 4. In this case, the reconstructed signal would have to exhibit strong distortions, but this is forgetting or not taking into account the inherent characteristics of the image signal. If we consider a photograph to be transmitted, this photograph consists of parallel lines, and these lines are in fact a series of points or pixels. The number of points is limited by the resolving power of the emitter and is therefore a direct function of the transmission band (this number is equal to twice the transmission band). That is, if the emitter cylinder rotates once per second, approximately 1600 points are determined, and 2 points are determined per second.
When rotating, it is limited to 800 points. When selecting a processing method (φ and n), the sampling frequency is defined by f=φ/n. The data in question is preferably fixed and the image signal processed in such a way that it can be analyzed, stored, transmitted and reproduced without significant alterations. When the picture is transmitted at a rate of one revolution per second, each line is analyzed with one increment. That is, a definition of 960 points or pixels is obtained.
(where φ=4800 baud and n=5 bits). Since the number of points corresponds to the number of samples per rotation of the cylinder, it is equal to the frequency f divided by the number of rotations per second. Since the line has a constant length (e.g. L = 200 mm), we calculate the spacing of the points (L/ff = 0.21 mm in this example) and from this we can determine the finer details than this for the reproduced photo. We can conclude that does not exist (this is a matter of probability). Furthermore, it should be noted that the fineness obtained in this manner is superior to that obtained with analog two-revolutions-per-second devices. The problem is much more difficult for photographs transmitted at a rate of two revolutions per second. If the sampling frequency remains unchanged, a fineness of 960 points can be obtained, but in this case, since two lines are analyzed, each line has 480 points, which means that the fineness that can be avoided is 0.4 mm or more. It is extremely inadequate due to its maximum dimensions. The reproduced image forms a mosaic of coarse lines. This is because each pixel ultimately occupies a small rectangle about 0.43mm long with constant density. The first thing to do is to carry out the pixel integration before sampling and thus between the two quantization operations (AD conversion). It is necessary to calculate the average density of the pixels (after quantization) as described in French Patent Application No. 8109240 of May 8, 1981 in the name of the applicant.
It is this average value that is transmitted digitally. When using a microprocessor, this microprocessor selects a reference frequency (4800 or 9600Hz) consistent with Shannon's theorem, performs sampling at this frequency, calculates the average value of the samples forming a pixel, and This average value would have to be quantized and transmitted after a log/linear transformation. This method exhibits the great advantage of allowing even the most minute details implicit in the transmitted signal to appear, even if their values are compressed by integration. In this way,
The concepts of probability and chance are removed. A second method of the invention consists in applying true signal processing to the level of the receiving section of the system.
The transmission band of analog signals can be reduced to 800Hz, and in the most undesirable cases (2
rotation), but 480 samples per line, so
It should be noted that 480 pixels are obtained. Therefore, the signal must be processed at the receiving end to prevent mosaic effects. This processing can be performed in an analog manner in a DA converter or in a digital manner prior to this conversion. In addition to the above, there are four main drawbacks that should be corrected: 1. Deterioration of the sharp transitions of the analog signal corresponding to large amplitude differences and thus boundaries between control areas. 2 Minute parts that are excluded during transmission processing but implicitly exist. 3 Contouring effects of blurred areas, as seen in the transition from one quantization level to an adjacent quantization level. 4 Undesirable mosaic effect. The invention therefore aims to provide a method by which these drawbacks can be eliminated. Therefore, in this invention, the most important characteristic of the image analog signal (at the transmitter) is the rise time t _n (10%
~90%), and this rise time is based on the fact that if the rotation speed of the cylinder is 2 revolutions per second, this rise time is about 550 μs (if the rotation speed of the cylinder is 1 revolution per second, this rise time is 700 μs). That's all). If we take these characteristics into account and compare the analog signal and its digitized signal, we conclude that there is no intersection between the analog signal and its average value more than twice during the pixel integration time. be able to. The present invention takes advantage of this characteristic, which, unlike other processing methods prior to transmission, enables restoration and reproduction processing at the receiver, ie, rather than proactive processing. The advantage of this method is that it does not require complex and expensive memories, and that it guarantees almost complete immunity to noise since each pixel is quantized as a unit and the transmission is therefore synchronous. be. It should also be noted that the receiver section of this device includes the inverse function of the logarithmic/linear transformation described above. That is, the method according to the invention provides for the purpose of reproducing a clear transition part of an analog signal by using an adjacent Considering the pixels and processing the center pixel according to the relative and absolute amplitudes of these pixels into m "sub-pixels" with weights (and hence densities) corresponding to the pixels surrounding this center pixel. It is in the division. Therefore, the average density value of the first pixel processed is saved and redistributed to the corresponding sub-pixels to be reconstructed according to the following equation: In this equation, C is the absolute amplitude of the central pixel to be processed, C _j is the amplitude of the lower pixels to be reproduced, and m is the number of lower pixels. In this way, the pixel size can be reduced to create an intermediate quantization level. Therefore, this method solves the mosaic problem and the contour problem at the same time. More particularly, according to a first embodiment of the invention, the number of pixels considered is at least 3.
, i.e. the pixel P to be processed, its preceding pixel P-1 and its successor pixel P+1 are considered, and the pixel P
forms a quantization difference x for the pixel P-1 and a quantization difference y for the pixel P+1. In this case, quantization of lower pixels created for processing pixel P is performed as follows. If the product xy is equal to 0, the amplitude value of the lower pixel C _j
remains unchanged and remains equal to the absolute amplitude C being processed. If the product x・y is positive, the value of the amplitude C _j of the lower pixel is calculated to ensure a stepwise transition between pixels P-1 and P+1, and the steps of this transition are determined experimentally. , changes depending on the difference between the values x and y. When the product x・y is negative, the value of the amplitude C _j of the lower pixel is calculated so as to ensure the reproduction of the minute part by a transition part showing a maximum value or a minimum value, and the shape of this transition part is determined by experiment. , and varies as a function of the difference between the values x and y. According to a second embodiment of the invention, the pixels considered are equal to five significant pixels, namely the pixel P to be processed, the preceding pixels P-1 and P-2, and the following pixels P+1 and P+2. Then, pixel P is pixel P
−1, or the pixel P
Pixel P-2 has a quantization difference y with respect to +1, pixel P-2 has a quantization difference α with respect to P-1, and pixel P+2 has a quantization difference α with respect to pixel P+1. Forms β. In this case, quantization of the lower pixels created for processing pixel P is performed as follows. If the quantization difference x is equal to zero, as in the previous embodiment, the amplitude value C _j of the lower pixel remains unchanged and remains equal to the absolute amplitude of the pixel P being processed. If the quantization difference is different from zero, many methods can be used. In this case, when the product x・y is equal to zero, if β・x=0, it remains unchanged unless the quantization difference |x| is less than or equal to a predetermined value (for example, 1) or equal to this predetermined value. It remains at a value equal to the absolute amplitude of the pixel P being processed. In this case, the amplitude C _j changes according to the following linear form. C _j = C-m+1-j/m+1x where, C is the absolute amplitude of pixel P, j is the number of the lower pixel, and m is the number of lower pixels of pixel P. If β·x>0, the value of the amplitude C _j of the lower pixel remains unchanged and remains equal to the amplitude of the pixel P being processed. If β·x<0, the value of the amplitude C _j of the lower pixel is calculated to ensure fine reproduction for pixels P and P+1 at the same time. This is because in this case there is a double point. When b x·y is negative, the amplitude values C _j of the lower pixels are calculated in such a way as to guarantee a fine reproduction for the pixel P and possibly for the pixels P and P+. c If x・y is positive, the amplitude value Cj of the lower pixel
is calculated to ensure a stepwise transition between P-1 and P+1 as described above. To simplify the clock device, the number of lower pixels m
is preferably chosen equal to the number n of quantization bits. Moreover, in order to obtain a high-performance system, the number m must be 3 or more than 3 (m>3). However, there are cases where it is not possible to treat it properly. These cases are limited to small fluctuations in high-frequency analog signals (480 Hz or higher for two rotations, 960 Hz or higher for one rotation). In conventional image processing,
It has been determined that these exceptional cases are extremely rare and that such processing does not result in additional blur compared to the unprocessed signal. Another object of the present invention is to apply the above-mentioned method to a device for reproducing an analog signal, such as a signal output from a Beran type analyzer, after digital transmission, and this system includes the following: a. In the transmitting section, it consists of a logarithmic/linear converter that receives the analog signal demodulated in advance, and two integrators that are installed in parallel to the output of the converter and operate alternately, and the luminance signal output from each pixel is The unit that integrates the whole and the time required for AD conversion for each pixel,
A digital signal representing the value of each pixel is serially connected to an AD converter connected to the output of the unit via a sample-and-hold circuit for recording the signal given by either one of the integrators. and a parallel-serial interface capable of transmitting this digital signal to one or both of a multiplexer and a modem for sending the signal to a transmission line according to current standards. b. In the receiving section, a serial-to-parallel converter that receives a serial digital signal sent through a transmission line, an A converter connected to the output of the serial-to-parallel converter, and a signal obtained from the DA converter. a linear/logarithmic converter for processing and obtaining at its output an exponential signal similar to the demodulated analog signal transmitted to the input of said emitter, and for applying the method of the invention upstream of said DA converter or and a downstream calculation/processing unit. According to another characteristic of the invention, the modulated signal provided by the emitter is transmitted via a demodulation/white signal detection/leveling circuit to a log/linear converter, which circuit is connected to the emitter and the logarithm. /includes a variable gain amplifier and a full-wave rectifier in series with the linear converter, and a switching device is provided between the full-wave rectifier and one or both of the logarithmic/linear converter and the transmission line. . In this case, the invention comprises a circuit acting on a switching device for interrupting the connection between the full-wave rectifier and the log/linear converter when there is no emitter transmission. It also detects the presence of a white signal issued at the start of transmission, and adjusts the gain of this amplifier so that the variable gain amplifier outputs a signal with an amplitude equal to the maximum value allowed by the AD converter, so that the gain of the amplifier remains constant during transmission. A circuit is provided to store a signal determining this amplifier gain so that it remains constant and to connect between the full wave rectifier and the log/linear converter when locking of the amplifier gain is performed. Furthermore, after the gain locking, the transmitting section
In addition, before the digital processing of the document, a circuit for generating a redundant binary word to be used as a synchronization signal of the receiving section so that the start and end of the binary word for quantizing the pixels of the receiving section can be recognized, and b. The receiver includes a circuit capable of generating a recognition language that causes the receiver to accept or not accept the message according to the code. In this case, the receiver itself consists of a circuit for decoding the recognition language, which allows or disallows the reception of the signal transmitted by the transmission line, along with the code used, and a return-to-zero circuit for the clock that controls the DA converter. and a synchronization word decoder for transmitting signals. The deductive processing circuit according to the invention is installed after or before the DA conversion. DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described in detail below with reference to embodiments shown in the drawings. Referring to FIG. 1, the signal from emitter 1 (eg, an amplitude modulated 1800 Hz carrier) is applied to the input of variable gain amplifier 2 through isolation transformer 3. When the white signal is continuously sent by the emitter 1 for 3 seconds, the white signal detection circuit 4 detects the presence of the start of transmission, and the amplifier 2 outputs a signal with an amplitude equal to the maximum allowable value of the AD converter 5. Adjust the gain of this amplifier 2 so that the This is because the signal of maximum amplitude corresponding to "white" produced by the emitter 1 corresponds to the highest digital value that the AD converter 5 can supply. However, if the document read signal exhibits an amplitude greater than the amplitude originally determined by the white signal, an amplitude clipper may be provided to match this amplitude to a preset white signal level. can. The signal determining the gain of amplifier 2 is stored in such a way that this gain always remains the same during the transmission of the document. As soon as this gain locking is achieved,
A circuit 6 generates a redundant binary word, which is provided with a synchronization signal to the receiver (FIG. 2) so that the receiver (FIG. 2) can recognize the beginning and end of the binary word that quantized the pixels. useful as. A full-wave rectifier circuit 7 equipped with a low-pass filter for demodulating the image signal of the 1800 Hz carrier wave is connected to the output of the amplifier 2. The demodulated analog signal provided by the rectifier circuit 7 is transmitted via a switching device 9 to a log/linear converter 8 . This converter 8 is intended to linearize the image signal. In fact, this image signal is exponentially combined with the various grays of the document from black to white. By linearizing this ratio, each pixel can be quantized with only 5 bits. At its output, the converter 8 is connected to a double integrator 10
connected to. Each of the multiple integrators operates alternately to integrate all the image signals output from each pixel from one sampling time to the next sampling time to obtain the average value of the image signals during that period. This method of using a double integrator eliminates the integration uncertainty caused by the single integrator requiring time to return to zero each time it integrates a pixel's luminance signal anew. As soon as this integration is finished, the signal is transferred into the sample and hold circuit 11 and stored therein for the time when the next connected AD converter 5 finishes its conversion. The digital signals representative of the pixel values are then serialized by the serial-parallel interface 12 and sent to one or both of the multiplexer and modem 13 at the output of the system. This modem is capable of sending signals on line 14 according to current standards. In this embodiment, the clock frequency that controls the AD converter 5 and the parallel-serial interface 12 is 4800 Hz, which is output by the modem 13, synchronously with N.
It's equal to divided by. N is the number of binary bits that make up the binary word that quantizes the pixel. In this way, a synchronous transmission of binary information is obtained, ie without additional bits at the beginning and end. On the receiving side, shown in FIG. 2, the serial digital signal transmitted by line 14 is applied to the input of a 5-bit serial-to-parallel converter (circular register 16). The signal output from this converter 16 is the circuit 6
Decoder 1 of the recognized binary word output from (Fig. 1)
7, which decoder either allows or disallows the recipient to receive the document, depending on the code 18 used. In parallel thereto, the synchronization word decoder 18' generates a signal upon this synchronization word after its plausibility check 19, which signals the zero of the counter 20 which divides the clock frequency of the modem 13 (4800 Hz) by five. Added to return input. The output of this counter 20 is AD
Commands the converter 21 to acquire the signal present at its input. 2 added to the DA converter 21
The binary word will be considered synchronously with the binary word generated by the emitter. The signal thus obtained is, except in the case of digital reconstruction, an analog image of the emitted signal. This signal is then passed to the linear/field number converter 22
and a calculation/processing unit 23 applying the method according to the invention. This exponential signal is then modulated in a modulator 25 with a carrier wave of, for example, 1800 Hz, and the signal is sent to the reproducing device 24 via an isolation transformer 26. This signal is therefore equivalent to the modulated analog signal output from emitter 1. To prevent interference effects appearing as moiré on the document reproduced by device 24, modulator 25 is slaved to the clock signal of modem 13. When the sending device finishes parsing the document,
1800Hz signal disappears. At this moment, the image signal detection circuit 27 detects that the modulation level of the modulation circuit 25 is -
Short-circuit the output of the converter 22 so that it becomes 50 dB,
The reproducing device (block 27'), which only takes signals above -37 dB, records the end of the transmission in that case. It should be noted that line-by-line reading of a document in the transmission system described above can be carried out according to principles other than the Verin method. Furthermore, another important advantage of the system according to the invention is that the clock frequency can be varied without disrupting the operation of the system. Further, the production of this transmission system can be implemented using a microprocessor. In this case, the order of processing performed on the signal is that on the receiving side, all processing performed on the signal is performed upstream of the DA converter 21, and on the transmitting side, all processing performed on the signal is performed on the signal. All processing is modified to be performed downstream of the analog-to-digital converter 5. First, as mentioned above, the diagram in Figure 3 shows that during the integration time of a pixel, the analog signal (analog base paper signal z(t)) and its average value (digital signal z(t)) This is to show that there is no communion. In this regard, recall that these two signals satisfy the following equation. Z(t _i )=∫ ^ti+1 _ti z(t)dt where t _i indicates the sampling instant and t _i+1 −t _i =1/f where f is the sampling frequency. In FIGS. 4 and 5 the number of pixels considered for signal processing was chosen equal to seven. This appears to correspond to the best compromise between effectiveness and complexity of the processing formula. However, 7 pixels, P-3, P-2, P-1,
Among P, P+1, P+2, P+3, the most significant pixels are the central five pixels, P-2, P-1, P, P
+1, P+2, and the two pixels at both ends P-3, P
+3 is only useful in certain limited cases. When limited to 3 pixels (P-1, P, P+1),
Although the processing is weaker, it results in a significant improvement in the quality of the reconstructed signal and the quality of the image reproduction. Such processing will be explained below. The variables that intervene in the processing of the digital signals illustrated in FIGS. 3 and 4 are as follows. That is, (1) the base frequency φ, which is chosen to be, for example, 4800 Hz, (2) the number p of pixels considered, in this case 7, (3) the number m of lower pixels of the central pixel P, in this case 5, ( 4) the number of quantization bits n, in this case 5; (5) the absolute amplitude of the considered pixel (which is a ₀ , a,
b, c, d, e, e ₀ ), (6) quantization level difference, which is α ₀ , α, x,
y, β, β ₀ (these differences satisfy the following formula), α ₀ =a-a ₀ , a=ba-a, x=c-b, y=d-
c, β = e - d, β ₀ = e ₀ - e, (7) Amplitudes C _j (j = 1,...・
5). Receiving Processing Algorithm A General Principle First, note that the differences x and y determine the type of processing to be performed, and α and β indicate parameters. α ₀ and β ₀ are used only in specific cases. The amplitude C of pixel P is used to account for the effectiveness of the processing so that the calculated value remains between the levels corresponding to photoblack and photowhite. It must be noted that since this process is iterative, the value considered at instant t is shifted by one pixel due to the processing corresponding to the pixel at the next instant t+1. Therefore, the following formula is obtained. α _(t+1) = α _0(t) , x _(t+1) = α _(t) , y _(t+1) = x _(t) , β _(t+1) = y _(t) , B Type 1 of processing with difference values x, y, β If x=0, no changes are made. That is, no matter what j is included in the range [1, 5], the following equation can be obtained. C _j =C 2 If x is different from zero, the type of treatment will vary depending on the value and sign of the products x·y and β·x. a When x・y=0: If β・x=0, the change is made only when x=±1, in which case it is included in the range [1, 5] and no matter what j is. , C _j must satisfy the following equation: C _j =C-6-j/6.x. If β·x>0, no changes are made. That is, whatever j is included in the range [1,5], the equation C _j =C will be realized. If β.x<0, then the "fine regeneration" process described below with respect to FIG. 6 is applied to the two pixels P and P+1 simultaneously. This is because in that case, a double point c=d exists. b When x·y<0, “fine reproduction” processing is realized for pixel P, and possibly for pixels P and P+1. If c x·y>0, a "step decomposition" process described below with respect to FIG. 7 is performed to facilitate the transition between pixels P-1 and P+1. Therefore, it may be sufficient to define the following two different processing methods. - “Fine reproduction” processing (x・y<0 and double points) −“Stepwise decomposition” (x・y>0) C Stepwise decomposition; (x・y>0) Regenerated lower pixels of the 5th stage are at level b and shows the decomposition of the transition between and level d. These sublevels have values determined by coefficients as a function of the differences x, y, α, β. . α=0 or β
= 0, use α ₀ or β ₀ instead of these in the calculation. Since the processing is iterative and concerns only one pixel at a time, steps between pixels must represent logical transitions. Therefore, a point must be defined that corresponds to the passage between two adjacent pixels. In the case of processing pixel P, this point P _i takes the following coordinates. t _i on the abscissa and P _(bc) on the ordinate, i.e. P _i = P _i
(t _i ,,P _(bc) ). Here

[Formula] is. This point P _i corresponds to the optimal graph passing point between pixel P-1 and pixel P. Similarly, an optimal passing point can be defined for the transition between pixels c and d. For d,
This optimal passing point P _i+i will have the following coordinates. The abscissa is t _{i +1} and the ordinate is P _(cd) . Here

[Formula] is. For processing pixel P, it is necessary to know the coordinates of this point P _i+1 with respect to pixel P. This simple movement of the origin is limited by setting d=c+y, and from this, P _i+1 (t _i+1 , P _(cd) ), where,

[Formula] Since the binary values ε _(bc) and ε′ _(cd) are known in this way, the fluctuations of the lower pixels at both ends are limited, and the ratio ε _(bc) /ε′ _(cd) As a function of , the amplitudes of all subpixels to be reproduced can be determined. However, if the values of C ₁ or C ₅ are ε _(bc) and ε′ _(cd)
There are two special cases in which the limit defined by is exceeded. These cases were determined by the following two tests, but it must be emphasized that in these cases, even if α or β is zero, it cannot be replaced by α ₀ or β ₀ . If α·x<0, and the ratio ε _(ab) /ε′ _(bc) is 1.4 or more, x is used instead of ε _(bc) . If y·β<0, and the ratio ε' _(de) /ε _(cd) is 1.4 or more, y is used instead of ε' _(cd) . It is clear that these substitutions are made in the calculations, formulas, etc. below. For convenience, let C _j =C+ε _j for all j (j=1, . . . 5). ｜ε′(cd)/ε(bc)｜If ≥4, ε ₁ = ε _(bc) , ε ₅ = −4ε _(bc) , ε ₂ = ε _(bc) , ε
₃ =
ε _(bc) , ε4=ε _(bc) , and if 4≧｜ε′(cd)/ε(bc)｜≧3/2, ε ₁ = ε _(bc) , ε ₅ = ε′ _{(cd )} , then ε ₂ = 1/10 (6ε ₁ − ε ₅ ), ε ₃ = −1/10 (2ε ₁ −3ε ₅ ), ε ₄ = −1/5 (7ε ₁ −3ε ₅ ), and 3/ If 2≧｜ε′ _(cd) /ε _(bc) |≧2/3, then ε ₁ ε _(bc) , ε ₅ = ε′ _(cd) , then ε ₂ = −1/2ε ₅ , ε ₃ = -1/2 (ε ₁ + ε ₅ ), ε ₄ = -1/2ε ₁ , and if 2/3≧｜ε' _(cd) /ε _(bc) |≧1/4, ε ₁ = ε _(bc) , ε ₅ = ε′ _(cd) , ε ₂ = −1/5 (3ε ₁ +7ε ₅ ), ε ₃ = −1/5 (3/1ε ₁ +ε ₅ ), ε ₄ = −1/ 5 (1/2ε ₁ −3ε ₅ ), and if 1/4>|ε′ _(cd) /ε _(bc) |, then ε ₁ = −4ε′ _(cd) , ε ₅ = ε′ _(cd) Then, ε ₂ = ε′ _(cd) , ε ₃ = ε′ _(cd) , and ε ₄ = ε′ _(cd) . C. Fine regeneration treatment (x·y<0) For this treatment, the same principle as described above is adopted. That is, the coordinates of the optimal passing points: P _i and P _i+1 are calculated. P _i (t _i , P _(bc) , here

[Formula] P _i+1 (t _i , P _(cd) ), where

[Equation] Similarly, since the amplitude of the lower pixel directly depends on the values ε _(bc) and ε' _(bc) , we can set the reproduced lower pixel value to the voltage corresponding to photoblack and photowhite, respectively. In order to limit the processing to hold between 1 and 2, the absolute amplitude C of the pixel P must be taken into account (Δ is the balance factor). As mentioned earlier, we have to add tests for (y・β) and (x・α), and if α and β are zero, we can replace them in the calculation with α ₀ and α, respectively. Note that β ₀ is used. If y β < 0, this corresponds to “fine reproduction type” processing for the P+1 pixel, but in order to locate the maximum value of the reproduced peak, the ratio ε′ _(de) /ε _(cd) is Must be researched. If this ratio is greater than or equal to 1.4, y must be used instead of ε′ _(cd) in the calculations below. Similarly, if x・α<0, the ratio ε _(ad) /ε′ _(bc) is 1.
Four
In the above case, -x is used instead of ε _(bc) . In all calculations below, all j(j
=1,... ₅ ) is used. Obtain C _j = C + △・ε _j . Therefore, the original processing is performed as follows. |ε′ _(cd) /ε _(bc) | If >1.4, ε ₁ =2ε _(bc) −4ε′ _(cd) , ε ₂ =−2ε _(bc) +ε′ _(cd)
′ ε ₃ = ε ₄ = ε ₅ = ε′ _(cd) . 1.4≧｜ε′ _(cd) /ε _(bc) | If >1.1, ε ₁ = 2ε _(bc) −4ε′ _(cd) , ε ₂ = −3ε _(bc) +2ε′ _(cd)
, ε ₃ = ε _(bc) ,
ε ₄ = ε ₅ = ε′ _(cd) . 1.1≧｜ε′ _(cd) /ε _(bc) | If >1, ε ₁ = 1/3ε′ _(bc) , ε ₂ = −2ε′ _(bc) , ε ₃ = 0, ε ₄ = 2 /3ε′ _(bc) , ε ₅ =ε′ _(bc) . |ε _(cd) /ε _(bc) | If = 1, ε ₁ = ε′ _(cd) , ε ₂ = 0, ε ₃ = −2ε′ _(cd) , ε ₄ = 0
, ε ₅ =
ε′ _(cd) . 1>｜ε′ _(cd) /ε _(bc) |≧0.9, ε ₁ = ε _(bc) , ε ₂ = 2/3ε _(bc) , ε ₃ = 0, ε ₄ = −2
ε _(bc) , ε ₅ =1/3ε _(bc) . 0.9＞｜ε′ _(cd) /ε _(bc) ｜≧0.7, ε ₁ = ε _(bc) , ε ₂ = ε _(bc) , ε ₃ = ε′ _(cd) , ε ₄ = 2ε _(
bc) −3ε′ _(cd) ,
ε ₅ = −4ε _(bc) +2ε′ _(cd) . If 0.7＞｜ε′ _(cd) /ε _(bc) ｜, then the following two cases are possible. If y・β<0, ε ₁ = ε _(bc) , ε ₂ = ε _(bc) , ε ₃ = ε _(bc) , ε ₄ = 1/3ε _(bc) −2ε′ _(cd) , ε ₅ = −10/3ε _(bc)
+2ε' _(cd) If y·β>0, the two pixels P and P+1 are processed as a block. Pixel P+1 has already been processed previously (iterative processing), and in this case, the calculation data is to be changed, so before outputting the data, the following five pixels corresponding to the result of processing pixel P+1 are A buffer containing the values is required, and these values can be erased during processing of P. A case in which this process is performed simultaneously for two pixels will be described below. E “Fine reproduction” processing performed simultaneously on two pixels in the case of double points (1) (y≠0) This processing complements the above case corresponding to the condition y・β>0. be. The five sub-pixels reconstructed for pixel (d) are written as follows. dj=d+Δ·ε _dj , where j takes a value from 1 to 5. To complete the calculation, calculate ε′ _(de) = β・|y|/|β ₀ |+|y|,
Perform the comparison below. If |ε′ _(de) |≦|y|, ε _dj takes the following value. That is, ε _d1 = −3ε′ _(de) , ε _d2 = 0, ε _d3 = ε _d4 = ε _d5 = ε
′ _(de) . If |ε′ _(de) |>|y|, the following two cases are possible. (a) ｜ε′ _(de) ｜＋｜y｜≦ε _(bc) , then ε _dj
takes the same value as above. That is, ε _d1 = −3ε′ _(de) ,
ε _d2 = 0, ε _d3 = ε _d4 = ε _d5 = ε′ _(de) . (b) ｜ε′ _(de) ｜＋｜y｜＞｜ε _(bc) ｜, then
ε _dj
is also changed as follows. That is, ε _d1 = −3(ε _(bc) −y), ε _d2 = 0, ε _d3 = ε _d4 = ε
_d5 =
ε _(bc) −y. (2) (y=0) In this case, the effective reference point is as follows. P _i (t _i , P _(bcd) ) and P _i+2 (t _i+2 , P _(cde) ) where

[expression] and

[Formula] This is because y is zero and the point P _i+1 has no reason to exist. Since these points overlap, it is assumed that the maximum value of the peak to be reproduced is in the center of this large pixel consisting of P and P+1. The 10 subpixels are written as follows: c ₁ , c ₂ , c ₃ , c ₄ , c ₅ , d ₁ , d ₂ , d ₃ , d ₄ , d ₅ . Here, ε _j and ε _dj correspond to jε[1,5]. Let ε=inf(ε _(bcd) , ε′ _(cde) . In that case, ε _1c = ε, ε _2c = ε, ε _3c = ε, ε _4c = 0, ε _5c = −3
ε,
ε _1d = −3ε, ε _2d = 0, ε _3d = ε, ε _4d = ε, ε _5d =
ε Note: In this entire processing section, the balance coefficient △ must be calculated for the following 10 pixels.
Therefore, this is the same for P and P+1. Calculation of equilibrium coefficient △ If M is the value corresponding to photowhite, C _j
In order to balance the values of , the balance factor Δ must be calculated. If x>0, the maximum floor difference εj is compared with the difference MC. If sup(ε ₁ ,...ε ₅ )<MC, Δ=1 If sup(ε ₁ ,...ε ₅ )>MC, Δ=
If MC/sup(ε ₁ , . . . ε ₅ )−x<0, the maximum floor difference −ε _j is compared with C. If sup(-ε ₁ ,...-ε ₅ )>C, △=1 If sup(-ε,...-ε ₅ )>C, △=
C/sup (-ε ₁ ,...-ε ₅ ) In conclusion, the following three points must be noted. --A field of application that serves as a limit for the study of amplitude variations of subpixels is the result of experiments and preliminary calculations taking into account the basic characteristics of portrait documents. The amplitude fluctuation of each lower pixel was roughly approximated to a linear function. This is to facilitate processing, whether analog or digital.
By changing the first derivatives of these functions, the effectiveness of the process can be easily manipulated while maintaining the same average density value. The second consideration concerns immunity to noise. The analog signal transmission band is approximately 800
Hz, so some transitions are prohibited, and by observing the digital signal as it is received, most of the harmful noise that shows unusual transitions, such as dark dots on bright ground, is removed. I can say that it is possible. For a given transmission band,
When x·y<0, there is a threshold value that cannot be exceeded by the sum |x|+|y|. Empirically, for n=5, this threshold is
It turns out that it can be taken equal to 15. As a final note, the signal obtained at the output of this processing system will be used for amplitude modulation of the 1800 Hz signal so that a conventional optical receiver can be connected. This 1800Hz is subordinated to the mother frequency φ from the modem to avoid interference,
It forms a natural wave of the modulated signal. F Processing when only 3 pixels are considered In this case, the only values considered are x and y, and these values play a symmetrical role. This is because even if x and y are reversed, the order of C _j can be changed. When x・y=0, the value of C remains unchanged, i.e. C _j
= 0 If x・y>0, refer to table (1) - “Transition” If x・y<0, refer to table (2) – “Transition”

【table】

[Table] Bye. The output signals are limited by levels corresponding to photoblack and photowhite, respectively.

[Brief explanation of drawings]

Figures 1 and 2 are block diagrams of the transmitter and receiver, respectively, of a digital transmission system for portrait documents transmitted and received by the Bellin type transmission method, and Figure 3 is a block diagram of the analog atomic signal and the corresponding digital transmission system. Figure 4 is a time function voltage diagram of a digital signal that displays variables. Figure 5 is a time function voltage diagram of a digital signal in which the central pixel is broken down into lower pixels. Figure 6 is a time function voltage diagram of a digital signal. is a time function diagram for showing the fine regeneration principle, and FIG. 7 is a time function time diagram for explaining step decomposition. 1... Emitter, 2... Variable gain amplifier, 3...
...Isolation transformer, 4...White signal detection circuit, 5
...A/D converter, 6...Redundant binary word generation circuit,
6'... Recognition word generation circuit, 7... Full wave rectifier, 8
... Logarithmic/linear converter, 9 ... Switching device, 10 ... Integrating circuit, 11 ... Sample hold circuit, 12 ... Parallel-to-serial interface, 13 ...
Modem, 14...Transmission line, 16...Serial to parallel converter, 17...Recognition binary word decoder, 18...Code, 18'...Synchronization word decoder, 19...Validity check circuit, 20...Clock , 21...D/A converter, 22...Linear/logarithmic converter, 23...Calculation/processing unit, 24...Reproduction device, 25...
Modulator, 27... image detection circuit.

Claims (1)

[Scope of Claims] 1. A received signal is processed as a function of the relative amplitude and absolute amplitude of p consecutive pixels, and the central pixel of the successive pixels is processed so that the density of each pixel is a function of the density of surrounding pixels. AD of the luminance signal representing the transmitted document, characterized by its resolution into subpixels of
A method of reproducing an image signal from a digital signal that is used for transmission after conversion. 2 Compression of data is performed by logarithmic/linear conversion in the transmitting unit, linear/logarithmic conversion is performed in the receiving unit, and processing of the received signal is performed before the linear/logarithmic conversion in the receiving unit. A method for reproducing an image signal according to claim 1. 3 Save the average value of the density of the original pixels to be processed,
3. The method for reproducing an image signal according to claim 1, wherein the signal is redistributed to the corresponding reproduced lower-order pixel according to the following formula. However, C is the absolute amplitude of the central pixel to be reproduced,
C _j is the amplitude of the reproduced lower pixel, and m is the number of lower pixels. 4 said number of pixels is at least equal to 3, i.e. the pixel P to be processed, the preceding pixel P-1 and the following pixel P+1, the pixel P forming a quantization difference x with respect to the pixel P-1; However, the quantization difference y is formed for pixel P+1, and the quantization of the lower pixel created for the processing of pixel P is as follows: When the product x・y is equal to zero, the amplitude value C of the lower pixel _j remains constant and remains equal to the absolute amplitude of the pixel P to be processed, and if the product x・y is positive, the amplitude value C _j of the lower pixel is equal to the value x between pixels P-1 and P+1. If the product x y is negative, the amplitude value C _j of the lower pixel is
calculated to ensure fine regeneration by means of transitions exhibiting local maxima or local minima, the shape of said local maxima or local minima being determined experimentally and varying as a function of the deviation between the values x and y; A method for reproducing an image signal according to any one of claims 1 to 3, characterized in that the method is performed as follows. 5 The number of said pixels is equal to 5, i.e. the pixel P to be processed, the preceding pixels P-1 and P-2, and the following pixels P+1 and P+2, where pixel P is equal to pixel P-1. The pixel P+1 has a quantization difference x, and the pixel P-2 has a quantization difference α with respect to the pixel P-1.
The pixel P+2 has a quantization difference β with respect to the pixel P+1, and the quantization of the lower pixels created for processing the pixel P is as follows: When the quantization difference x is equal to zero, The amplitude value C _j of the lower pixel remains unchanged and remains equal to the absolute amplitude of the pixel P to be processed, and if the quantization difference is different from zero, a, and if the product x・y is equal to zero, then , () If β x = 0, quantization difference |x|
remains constant and equal to the absolute amplitude of the pixel P being processed, unless C j is less than or equal to a predetermined value (e.g. 1), in which case the amplitude C _j varies according to the following linear formula: Then, C _j = C-m+1-j/m+1x where C is the absolute amplitude of pixel P, j is the number of lower-order pixels, m is the number of lower-order pixels of pixel P, and () If β・x>0 , amplitude value of lower pixel
C _j remains unchanged and remains equal to the amplitude of the pixel P being processed; () If β x < 0, the amplitude value of the lower pixel
Since there is a double point in that case, C _j is calculated to ensure fine reproduction for pixels P and P+1 at the same time, and if b x・y is negative, the amplitude value of the lower pixel C _j is for pixel P, and possibly for pixels P and P+
1, and if c x・y is positive, the amplitude value of the lower pixel C _j
is calculated to ensure a stepwise transition between P-1 and P+1. . 6 To perform a stepwise transition between pixel P-1 and pixel P+1, transition P-1→P and transition P→P
The optimal passing points P _(bc) = C + ε _(bc) and P _(cd) = C + ε' _(cd) for +1 are respectively expressed as ε _(bc) = -x・|y|/|y|+|α| and Determine by ε′ _{( cd )} = −y | 6. The method for reproducing an image signal according to claim 5, characterized in that the image signal is determined. 7 The above-mentioned fine reproduction processing is performed by calculating the value of the lower pixel to be reproduced by using the amplitude of the lower pixel of the voltage corresponding to photo black and photo white as a function of the values ε _(bc) and ε' _(cd) of the pixel P. 7. The method for reproducing an image signal according to claim 5, wherein the determination is made taking absolute amplitude into consideration. 8a In the transmitting section, it consists of a logarithmic/linear converter 8 receiving a pre-modulated analog signal, and two integrators 10 installed in parallel and acting alternately on the output of said converter 8, each pixel a unit capable of integrating the entire luminance signal output by the integrator 10, and a sample hold for storing the time required to AD convert the signal supplied by either of the two integrators 10 for each pixel. connected to the output of said unit via a circuit 11
An AD converter 5 and a serial/parallel interface that serializes a digital signal representing the value of each pixel and transmits the serialized signal to one or both of a multiplexer 13 and a modem that sends the signal to a transmission line 14 according to current standards. ace 12; b. in the receiving section, a serial-to-parallel converter 16 that receives the digital/serial signal output from the transmission line 14;
a DA converter 21; a linear/logarithmic signal capable of processing the signal obtained by the DA converter 21 and outputting an exponential signal equivalent to the demodulated analog signal transmitted to the input of the transmitting section; 1. A reproducing device for an image signal obtained during digital transmission of an analog signal, such as a signal output from a Beran type analyzer, characterized in that it comprises a converter 22. 9. The image signal reproducing apparatus according to claim 8, wherein the modulator 25 is made dependent on the clock of the modem 13 in order to prevent interference that appears as moiré in the reproduced document. 10 The modulated signal emitted by the emitter 1 is transmitted via the demodulation/white signal detection 4/leveling circuit to the log/linear converter 8, which circuit connects the emitter 1 and the log/linear converter 8. A switching device 9 is provided between the full-wave rectifier 7 and one or both of the logarithmic/linear converter 8 and the transmission line 14. An image signal reproducing device as claimed in claim 8. 11. Claim characterized in that the transmitter comprises a circuit acting on a switching device 9 for interrupting the connection between the full-wave rectifier 7 and the logarithmic/linear converter 8 in the absence of an emitter transmitter. The image signal reproducing device according to item 8. 12 The transmitter detects the presence of a white signal transmitted at the start of transmission, and adjusts the gain of variable gain amplifier 2 so that it outputs a signal with an amplitude equal to the maximum value that can be recognized by AD converter 5. and store the signal determining this gain so that the gain of amplifier 2 always remains the same during transmission, and when the locking of the gain of amplifier 2 is carried out, the full-wave rectifier 7 and the log/linear converter 9. The image signal reproducing apparatus according to claim 8, further comprising a circuit for connecting between the image signal and the image signal. 13, comprising a clipper that aligns the amplitude of the white level of the read signal to the predetermined white level when the amplitude of the white level of the read signal is higher than the predetermined white level amplitude. An image signal reproducing device according to claim 8. 14 After the gain locking, the transmitting unit:
Before the digital processing of the document, a redundant binary word is generated which generates a redundant binary word which acts as a synchronization signal for the receiving part so that the receiving part recognizes the beginning and end of the binary word which quantizes the pixels.
b) a recognition word generation circuit that generates a recognition word 6' that allows or disallows the receiver to receive the message according to the code used; An image signal reproducing device according to claim 8. 15. The receiving section has a circuit 17 for decoding the recognition word, which allows or disallows the receiver to receive the signal transmitted by the transmission line, depending on the code used.
9. The image signal reproducing apparatus according to claim 8, further comprising: a synchronization word decoder 18 for transmitting a lock return-to-zero signal for controlling the DA converter 21. 16. The image signal reproducing apparatus according to claim 8, wherein the receiving section further includes an image detection circuit 27 that short-circuits the output of the converter 22. 17 When implemented using a microprocessor, the processing order for signals is as follows in the receiving section:
The signal processing is performed upstream of the DA converter 21, and in the transmitting section, the signal processing is changed to be performed downstream of the AD converter 5. Image signal reproducing device.