FR2475834A1 - METHOD AND DEVICE FOR REDUCING NOISE IN A SAMPLE SIGNAL - Google Patents

Abstract

THE INVENTION RELATES TO A DEVICE FOR REDUCING NOISE IN A SAMPLE SIGNAL. IT IS FIRST DETERMINED IF THE SIGNAL CONTAINS LOW FREQUENCY INFORMATION ONLY BY CONSIDERING SAMPLES CLOSE TO 44 OR 54 TO SEE IF THEIR AMPLITUDES ARE AT A CHOSEN QUANTITY FROM ONE ANOTHER; THE AVERAGE OF TWO OR FOUR IMAGE SAMPLES IS TYPICALLY DONE DEPENDING ON THE SIZE OF THE AREA WHERE THERE ARE ONLY LOW FREQUENCIES, BUT LARGER NUMBERS OF SAMPLES MAY BE USED. THE INVENTION APPLIES IN PARTICULAR TO THE DIGITAL PROCESSING OF TELEVISION SIGNALS.

Description

The present invention relates to digital television and more

  especially to reducing treatment

the noise of ouantification.

  In digital video, when too few bits are used to quantize each sample of the video waveform, the noise of quantization is large and causes an annoying outline of the image being viewed. This is very annoying in zones of gradual change of intensity where the slope of the video waveform is low.

  compared to a dimension of the quantization step

  -ion. This is unheard of in such low-frequency parts, and dramatically imaged, the eye is very sensitive to noise. By increasing the number of bits per sample, more levels are available for quaitification, and the gradual change is more accurately represented. However, the use of more bits per sample results in a corresponding increase in the frequency of data required in digital video. One knows, by Article 1. i.title "ECIî Lncoded i-T" C Color Television Subjective

  Oests "by A.A. Goldberg, JSJL: PTE, August 1973, pages e49-

  : 54, add either a slot or a sta.istiquausiy signal Jvidèo aval- csantiúicat-onpour reduce the formation of contours daants, and then pass, by a wire re-low-low, the si; -nal analog video reproduced to reduce the

  added signal visibility and quantization noise.

  However, in this system, the low-pass filter reduces

  also high frequency video information.

  it is therefore desirable to reduce the

  _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _

  Without increasing the data frequency and without reducing the high-frequency video information, as is the case with the present invention, this is achieved by determining whether the digital signal is only low-frequency information and in flittering. samples close to this signal if only the low-frequency information is present for

  reduce the noise content of this signal.

  The invention will be better understood, and other purposes, features, details and advantages thereof

  will become clearer during the description

  explanatory following with reference to the accompanying schematic drawings given solely by way of example illustrating an embodiment of the invention and in which: - Figure 1 shows a video signal gradually changing> quantified using only a few levels ; - Figure 2 shows the same signal with more quantization levels; FIG. 3 shows a flat frame with a coarse quantization; FIGS. 4A-4D show various waveforms relating to a flat frame using the amplitude averaging of the present invention; FIG. 5 shows, in block diagram form,

  an adaptive system for training the average of the ampli-

  study; FIGS. 6 and 7 show waveforms used in the system of FIG. 5; Figure X shows a diagram used to explain Figure 5; - Figure 9 shows the details of a circuit

  delay buffer used in the figure -

  FIG. 10 shows the details of a logic circuit for forming the two-point average, used in FIG. 5; FIG. 11 shows the details of a logic circuit for forming the four-point average used in FIG. 5; and - Figure 12 shows in more detail

  of the present invention.

  FIG. 1 shows a graph of an analog video signal 12 sampled at a fixed sampling frequency at sampling points in time 14 and quantized at the next lowest levels of various quantization levels 16 (time of sampling). sampling are shown as abscissa and quantization levels as ordinate). This results in a digital waveform 18. The difference between the analog signal 12 and the digital signal 18 is the quantization error which may be as large as a quantization level. Since the slope of the waveform 12 changes slowly with respect to the difference between the quantization levels 16, sharp edges 20 occur in the digital waveform 3 which are separated from each other a certain distance apart. considerable along the Horizontal axis, a signal of constant amplitude. This results in highly visible contours in the image viewed from signal 18. Figure 2 shows a graph where the number of quantization levels has been increased and benchmarks

  correspondents have been applied to corresponding elements

  graphs. It can be noted that the steps or steps 20 are much smaller in amplitude and occur more frequently than in the case of FIG.

  resulting in a lower quantization error.

  However, increased quantization levels require an increased frequency of data. The situation of Figure 3 o, again, of the parts

  correspondents in the graph have been given corre-

  respondents. It can be seen that given the small number of quantization levels 16, there is a fixed quantization error 22 if the analog signal 12 is flat and occurs between quantization levels like this.

is illustrated in Figure 3.

  Figure 4A shows a signal that can help overcome these problems. An offset signal 24 (on the abscissa) is illustrated, which has a slot shape, a

  frequency equal to half the sampling frequency

  used and a peak amplitude equal to half the difference in amplitude between adjacent quantization levels. (T: sampling time). It should be noted that the center of the upper part of the slot 24a

  coincides with one of two sample points.

  FIG. 4B shows the results when this shift signal 24 is added to an analog video signal 12 having a constant amplitude of 3/4 of a

  quantization before step or quantization step.

  It may be noted that in FIG. 4B, the amplitude of exchange

  Adjacent ticks alternate up and down half of a quantization level. FIG. 4C shows the result of the signal of FIG. 4B after quantization. This signal 25, in the case shown, exists only at the quantization level O and 1 and the adjacent signals do not differ from each other by more than one quantization level. That being the case, and according to the invention, the signal 25 is integrated

  or average, either to the eye or electronically

  only to give the result shown in FIG. 4D if the average is done electronically. This results in a signal level halfway between two levels of ciphering. This remains true for any value of the analog video signal 12 in the

  upper half of any quantization range, that is,

  between half and one, and half and two, two and a half and three, and so on. For a signal value at the lower half of a range of quantification, ie between zero and one-half, one and one and a half, two and two-and-a-half, etc., the video signal and combined offset will always be quantized at the lower end of the range, ie zero, one, two, etc. respectively. In each case, after completing the integration or

  the formation of the average, the maximum error of quan-

  cation and half of the quantization step, instead of the maximum error of one step as described above for Figure 1, This corresponds to an effective doubling of the number of quantization levels without adding of additional bit per sample, which can be used to represent information changing the element. In the case of high-level inatormation, the error is increased to a level and a half (a level of oizine plus half-level of the shift signal), since the average of the signals can not be averaged. However, as mentioned above, the noise in the high-frequency parts of the signal is less annoying than in the

-arzL es U Frequency break.

  The above concept describes how to be extended to the formation of: average of-more samples-to obtain more amp resolution at low frequency. By exeraple, two synchronous offset signals alternating on the lines, each having two levels which differs from the levels of the other signal, out of a total of four levels (ie, the level corresponding to a quarter of a gsradin of o3mn). : you can be added to the ana-soft video signal before quantization. On the condition that the four adjoining clinches are not more than one, the average of the four will produce one of the four possible levels of amplitude, three or more intermediate. of amplitude c, that is to say, a quarter, a half, three quarters of a cuantification step and a level of quantification. It is

  the ecuivalent of the addition of two bits per sample.

  2.5 If the condition and the adjacent samples are not more than one quantization level, then two adjacent teeth of the image may be removed. the two parts of the plant are not separated from more than one level, we can carry out their poyenne to obtain a supplementary oi of resolution of arnp.li tude by Schantil! on. If the condition qut. two] the adjacent elements of 'i.r e will not dimz more than one level is not filled, so s. The content of the imae is signaled at the same frequency, that is to say, when the quantization error is D1. The above-mentioned added signal may be used to explain the invention, when the steps or quantization steps are small or small, compared to the noise in the video signal. received, no offset signal is required Basically, this application relates to filtering or forming the soft average of mutual samples close to signals Figure 5 shows an example of a system for implementing The above operations An analog video signal is applied to an input of an adder 26, while the offset signals are applied at every other line at its other input. FIGS. 6 and 7. During alternate lines or every other line, an offset signal 28, such as that represented in FIG. 6, is applied to the adder 26. it alternates between amplitudes of zero and three-quarters of the difference between adj level During the remaining alternating lines, the signal 30 of FIG. 7 is applied to the adder 26. It has an amplitude which varies between one quarter and one half of that of the

  difference between adjacent levels of quantization.

  These signals 2S and 30 are at half the sampling frequency with a phase shift of 90 between them. As can be seen in FIGS. 6 and 7 where the quantization levels are indicated on the ordinate, the sampling clock signal has a phase shift from one line to the other, thus shifting the sampling points. from one line to another. On the other hand, there is a phase shift of 180 for alternate lines of each of the shift signals 28 and 30. These phase shifts 0 occur at each time the respective signals begin a new horizontal line. The portion of the output signal of the ACD 26 due to the signals 28 and 30 is shown schematically in FIG. 8 which shows part of a frame. The quantization levels that have been added to the analog video signal by the adder 26 are shown in FIG. 8, expressed in quarters of a cuantification level. it is necessary

  note that these numbers represent the additional levels

  rather than the absolute value of digital

  coming out of the adder 26. In the following description,

  the sampling point 32 will be considered as the one currently under consideration and reference is made to point 34 on the same line, point 36 on the line above and point 38 on the line below. The video signal

  analogue plus additional levels of quantification

  are applied to an analog / digital converter 40 which, in turn, applies a

  3 bits of the sum of this analogue video signal.

  and additional levels, via a transmission path 41, to a delay buffer circuit 42. This 8-bit buffer circuit 42 applies, at each of its

  outputs 32a, 4a, 36a, 3Ga, 8-bit signals representing

  both the amplitude of the signal occurring at the various sample points 32, 34, 36 and 38 respectively. The outputs are named using corresponding references with the suffix "a" added to indicate which points appear at which outputs. All outputs of the buffer circuit 42 are applied to a four point average logic circuit 44. This circuit produces, at an output 46, a flag signal when the four points 32, 34, 36 and 38 are not different from each other, in amplitude, by more than one quantization step. The output 46 applies the flag signal as the control signal to a multiplexer 48 which includes a switch. The switch is at the position shown when the above condition is true, so the average of the amplitudes of the four points 32, 34, 36 and 38 is provided by the output 50 of the logic circuit 44 and is applied to the data output 52 by the Zl8 multiplexer. The signal at the outputs 50 and 52 has the resolution of a 10-bit signal, which reduces the contour without increasing

  the frequency of interference by the transmission path 41.

  If these four points are not at a difference of a quantization level from each other, then there is no flag signal for selecting the average of the four points at the output 46 and therefore the multiplexer 48 goes to its lower position and thus receives the output of the multiplexer 56 which also contains a switch. The signals representing the points 32 and 34 at the outputs 32a and 34a are applied to the logic circuit for forming the two-point average 54, and

  if these points are at a level of a level of quantification

  of the other, a flag selection signal of the two-point average is applied by the output 58

  multiplexer 56 and is then in the position

  sented. In this case, the output 60 provides a signal representing the actual average on two points of the points

  32 and 34 and applies it to the multiplexer 48 by the multi-

  plexeur 56 and therefore at the data output 52.

  This gives the resolution of a 9-bit signal. In the case where the points 32 and 34 are not at the distance of a step of the quantization level, then no two-point average selection flag signal is present at the output 58 and therefore the multiplexer 56 is at the lowest position (not shown), and only the signal representing the point 32 is applied by the multiplexer 58 and the multiplexer 48 at the output 52,

  this signal being an 8-bit signal.

  The shift signals 28 and 30 added in every other line of Figures 6 and 7 are repeated in Figures 12a and b. By operating on an analog signal 1212 such as that shown in FIG. 12c having a relatively constant amplitude being at a quantization level 1216, the arrangement of FIG. 5 produces during a first horizontal line an added signal illustrated at 1214 in FIG. 12d which represents

  the sum of signals 28 and 1212. During the horizontal line

  Next, the sum signal 1220 is produced which represents the sum of the signals 30 and 1212. When quantized, the signal 1214 takes an equal numerical value

  at the level of quantization 1216 at each point

  as shown by the signal 1222 in Fig. 12f, because the analog signal 1214 never reaches the quantization level 121 &. Similarly, signal 1224 of FIG. 12g represents the value

  At digital resulting from the quantization of signal 1220.

  Siunal 122-24 also remains at quantization level 1216 because the signal 1220 does not reach the level of 121C quaternization. The time constant of the signals 1222 and 1224 is equal to the quantization level 1216 and therefore the numerical value is a

  approximation close to analog values.

  Figs. 12h-12o illustrate the conditions where the input analog signal takes a value slightly above the quantization level 1216 as illustrated by the signal 1226 in Fig. 12h. The signal 1226 is above the quantization level 1216 by a quarter of a quantization level. If the signal 1226 was simply

  quantified as in the prior art, the error in quantifying

  cation would be one quarter of a quantization level.

  The signal 1228 of FIG. 12j illustrates the sum of the shift signal 28 and the analog signal 1226 that is produced by the device of FIG. 5. It will be noted that the signal 1228 reaches the next higher level of quantization at 1218 points. alternating sampling. Signal 1230 of FIG. 12k1 represents the signal 30 and signal 1226. As signal 1226 is one quarter of a quantization level above the 121 level, and that of the signal 50 to which it is added is at the quarter-level of quantization, the signal 1230 does not reach the next higher level of quantization according to 121. The result of the quantization of the signal 12 is illustrated by the signal 12 Fig. 12m, and the result of the quantization of the signal 1230 is illustrated by the signal 1234 in Fig. 12n. The time average of the numerical sum of the digital signals 122 and 123 / produced by the FIG. 5 is a quarter of a quantization level above the level 1216 which is exactly the value of the analog signal 1226. Thus, in this case, the error of

  quantization has been reduced from one quarter level to zero.

  If the analog signal applied to the arrangement of Fig. 5 is midway between quantization levels 1216 and 1218, the sum signals are as illustrated by waveforms 1236 and 1238, shown together on FIG. Figure 12o. It is apparent that the waveform of sum 1236, when formatted numerically, is identical to signal 1232 with half of its residence time at level 1216 and the other half at 1218. Signal 1238 will also take, when it will be put in digital form, a digital value identical to the signal 1228 except for a phase shift. When an addition in the device of Figure 5, the output signal will take an average value

  in time halfway between levels 1216 and 1218.

  This is exactly equal to the value of the analog signal

  received and results in a zero quantization error.

  If the received signal has a magnitude of 1/4 quantization level below level 1218, the quantization error can be expected to be 3/4 of a quantization level. When summed with signal 28, the sum is half time above level 1218 and half the time below. The sum with the signal 30 remains at all times above the 1218 level. When formatted numerically and averaged over time, the output signal will be at a quarter level below the 1218 level.

  while the quantization error will be reduced to zero.

  Fig. 9 shows an embodiment of the delay buffer 42. The 8-bit input signal of the converter 40 is applied to the output 38a _5 directly and has a 63.5 microsecond delay line 62 (a horizontal line). ) minus 70 nanoseconds. The output of line 62 forms output 32a and is also applied

  at the delay line 64 which has a delay of 140 nanoseconds.

  The output of line 64 is applied to the output 34a and also to the delay line 66 which has a delay of 63.5 microseconds minus 70 nanoseconds. The output of line 66 forms the output 36a. It should be noted that all the delays above are for a system with 525 lines per image, 30 images per second and for a frequency

  sampling rate of 7.16 iEz. Delays of 70 nano-

  seconds are required to obtain an offset of the half of a sample interval to the sampling frequency, which is necessary because of the phase difference between the signals 28 and 30. In other systems, other values of delay for

lines 62, 64 and 66.

  Figure 10 shows a detailed circuit diagram

  logic 54 of the formation of the two-point average.

  The signal at the output 32a of the buffer circuit 42 is applied to an input of a subtraction means 68, while the signal at the output 34a is applied to another input of the subtraction means 68. A difference signal is present at the output of the subtraction means 68 and is applied to an input of a digital comparator 70 which receives, at its other input, a logic signal "1" present at the line 71. The comulator 70 produces, at the output 58, a "1I" logic if the difference applied between its two inputs is less than or equal to 1 and a logic "0" in other cases. This signal is the flag signal for selecting the above-mentioned two-point average. The signals at the outputs 32a and 34a are also applied to an adder 72 and their sum, which forms the average

  of the two points, is applied to the output 60.

  FIG. 11 shows the details of the four-point average logic circuit 44. The signals at the outputs 32a, 3a, 36a and 38a are applied to the circuit 44. The output signal 32a is applied to all means of subtraction 74, 76 and 78. The signal at. the output 36a is applied to the subtraction means 74 which applies the difference between this signal and that of the output 32a to the digital comparator 80. This comparator applies the logic level "1" if the difference between its input signals is less than or equal to to 1, to an AND gate 86. The output signal 38a is applied to the subtraction means 76 which applies the difference between this signal and that of the output 32a to the comparator 82. The comparator 82 applies an output signal, if this

  difference is less than or equal to 1, at the AND gate 86.

  The signal at the output 34a is applied to the subtraction means 78 and the difference between this signal and that

  at the output 32a is applied to the digital comparator 84.

  If the difference is smaller than 1, a logic signal is applied to the AND gate 86. It may be noted that a logic signal 11 "I is applied to the digital comparators 80, 82 and 84 by the line 85 so that they can If the difference is less than or equal to 1 at the output of all the comparators 80, 82 and 84, the AND gate 86 produces a signal in the high state, which forms the flag signal of the average of four points. at the output 46. The signals at the outputs 32a and 36a are applied to the adder 88 which, in turn, applies their sum to an input of the adder 92. The signals at the outputs 38a and 34a are applied to the adder 90 which, in turn, applies their sum to another input of the adder 92. Thus, the output of the adder 92 is formed of the signal of the average of the four points to

  bits that is present at output 50.

  It will be appreciated that variations are possible within the scope of the invention. For example, the point to the right of point 32 can be used to derive the average of the two points and the points to the right at the top and bottom of point 32 can be used to derive the average of the four points. Other combinations of surrounding samples may be used. Moreover, the concept can be extended to the average of eight points or more. If the received video signal is already a digital signal, then the converter 40 will be only quantization means for requantizing after the addition of the offset signals by the adder 26. Also, the use of a phase shift of 180 line to line in the sampling frequency is not required. It has been used in a preferred embodiment which is employed with the invention disclosed in U.S. Patent Application No. 132,137 filed March 1530 by Reitmeier and Dischert. The offset signal, if used, need not be synchronous with the sampling signal but

  this can produce interference signals.

  Of course, the invention is not limited to the embodiment described and shown which has been given as an example. In particular, it includes all the means constituting technical equivalents of the means described and their combinations if they are executed according to its spirit and implemented.

  in the context of protection as claimed.

Claims (16)

R E V E N D I C A T IO N S   A method for reducing noise in a sampled signal, characterized in that it determines whether said signal contains low-frequency information (74-86 or 68.70) and that mutually garbled samples are filtered. signal if only low frequency information is present to reduce the noise content of said signal (8892 or 72). 2. Method according to claim 1, characterized in that said step of determining comprises comparing at least two close samples to determine if their amplitudes are in a quantity.   chosen in relation to each other.   3. Method according to claim 21characterized in that the aforementioned step of comparison consists of   comnarer 4 close samples (32-38).   4. Method according to claim 3, characterized in that the aforementioned filtering step consists in averaging at least 4 samples p'oches (32-38) if said 4 samples are in the selected amount nprécitée some by to the others and if not, to average two close samples (32, 34) if said samples are in said selected quantity one compared to each other.   5. Process according to any one of the claims   2; 3 or 4, characterized in that the above-mentioned sampled signal is quantized and in that the quantity chosen   This specifies the quantification stage.   C. The compound of claim 1 wherein said filtering step comprises:   at least two close samples (32, 34).   7. Method according to claim 5, characterized in that the above-mentioned step of filtering consists in making the   average of 4 gold samples (32-38).   The method of claim 1, characterized in   what Filt-er's nested statement is to make the average.   9. P-océdé according to 1'endendication 1l caract4gisé in that leoute read reads a shift signal (28 or) to the aforementioned sampled signal having the aforesaid step of determination.   The method according to claim 9, characterized in that said shift signal (28 or 30) is   synchronous with the sampling frequency.   11. The device according to claim 10, characterized in that the aforesaid offset (28 or 30) comprises a slot having a frequency equal to the   half of the sampling frequency.   12. A method according to claim 11, characterizing the above-mentioned sampled signal (28 or 30) of a television signal having scan lines and thereby the above-mentioned offset signal.   a phase shift of 180 from one line to another.   2C 13. A method according to claim 12, characterized in that said sampled signal is quantized and in that the offset signal (28 or 30) above said peak-to-peak amplitude of alternating lines of ii and a half quarter of the autification steps 14. Apparatus for reducing noise in a signal sampled according to the method of claim 1, characterized by means (78-80 or 68, 70) for determining whether said signal contains only low frequency information and means (88-92 or 72) for mutually filtering orbital samples of said signal if only a low frequency in'ormation is present at the noise content of said signal. 15. Apparatus according to claim 14> characterized in that the aforementioned determination means comprises means (70, 80-84) for comparing at least two random samples (32, 34) to determine if their amplitudes are in a selected amount, one for each do not tell each other.   Device according to claim 15, characterized in that the comparison means (80-84) compares 4 close samples (32-38).   17. Device according to claim 16, characterized in that the aforementioned filtering means comprises means (88-92) for averaging 4 close samples (32-38) if said samples are in the abovementioned chosen quantity, some by compared to others. 18. Device according to any one of claims 15 to 177characterized in that the sampled signal mentioned above is quantified and in that the   chosen quantity includes, a step of quantifying   cation. 19. DisDositif according to claim 14, characterized in that the aforementioned means of -iltrage comprises means (72) for averaging at least two samples close (32,34).   20. The device as claimed in claim 19, characterized in that the aforementioned filtering means comprises means (88-92) for averaging four close samples. 21. Device according to claim 13, characterized in that the said tilting means comprises a   average to average (72 or 88-92).   22. The device according to claim 14 further characterized by means for adding an offset signal.   to the aforementioned sampled signal before the means of - aforementioned determination.   23. Device according to claim 221characterized in that said shift signal (28 or 30) is   synchronous with the sampling frequency.   24. The device as claimed in claim 23, wherein the shift signal comprises a slot having a frequency equal to half the length of the slot. sampling frequency.   25. Device according to claim 24, characterized in that the sampled signal mentioned above is composed of a siena! t4.1 ('ision having scan lines and in that the dWcalage signal (28 or 30) precit4   has a phase difference of 180 from one line to another.   2C. DisnosJi? according to claim 25, characterized in that the aforesaid sampled signal is quantized so that the offset signal (28 or 30) precedes, 6 i. an amplitude rte to crete lines atIternées of a half and an ocuart of the steps of quantification