JP3751645B2 - Noise removal apparatus and method - Google Patents

Description

[0001]
[Industrial application fields]
The present invention relates to a noise removal apparatus and a noise removal method for removing noise from a signal containing noise, in particular, removing noise from a digital signal including noise with strong impulse characteristics.
[0002]
[Prior art]
Conventionally, as a noise removal method for removing noise from a digital signal mixed with noise, for example, a moving average method or a median method is known.
[0003]
Here, the noise removal method using the moving average is based on the average value of signals (hereinafter referred to as neighborhood signals) that are temporally or spatially close to the signal to be processed (the signal from which noise is to be removed). It is like performing a replacement operation with a signal.
[0004]
Further, the noise removal method by the median is such that an operation of replacing the median value when the neighboring signals are arranged in the order of ascending or descending and the signal related to the processing is performed.
[0005]
[Problems to be solved by the invention]
By the way, for example, noise (error) that occurs during recording / reproduction of a recording / reproducing apparatus that handles digital signals has such characteristics that the impulse characteristics are strong (random noise) and that the noise signal is isolated. Note that “isolated” means that there is no other noise signal in the neighboring signal.
[0006]
However, when noise removal processing using the conventional noise removal method as described above is performed on such isolated noise with strong randomness, most signals that are not affected by noise are also processed in the same way. It will end up. That is, for example, when an image signal is taken as an example, if a signal not affected by noise is processed in the same manner as described above, image quality deterioration such as image blur occurs as a result.
[0007]
In view of the above, the present invention has an object to provide a noise removal apparatus and method that can remove only isolated noise with strong impulse characteristics.
[0008]
[Means for Solving the Problems]
The present invention has been proposed in order to achieve the above-described object, and the noise removal apparatus according to the present invention performs prediction based on past input data and calculates prediction data corresponding to the current input data. To adaptively determine the presence or absence of noise based on the first prediction means, the input data processed earlier in time and the predicted data corresponding to the input data processed earlier in time. A second prediction means for predicting a threshold value to be used, and a comparison for comparing a difference between the current input data and the prediction data corresponding to the current input data with a threshold value from the second prediction means; And means for selecting either the current input data or the prediction data corresponding to the current input data based on the comparison result of the comparison means.
[0009]
Here, the second prediction means predicts the threshold value based on past input data and past prediction data corresponding to the past input data. The first prediction means includes a coefficient calculation unit that adaptively determines a prediction coefficient by performing a least square operation on the past input data, and a signal that performs linear prediction using the current input data and the prediction coefficient It consists of a value prediction unit.
[0010]
The noise removal apparatus of the present invention further includes first storage means for storing input data, and second storage means for storing prediction data from the first prediction means, and the second storage means. The prediction means adaptively changes the threshold value based on the difference between the data from the first storage means and the prediction data from the second storage means.
[0011]
The input data is preferably divided into blocks of a plurality of lines and a plurality of pixels, and it is preferable to perform the process such as the prediction using the data in units of blocks.
[0012]
Next, the noise removal method according to the present invention obtains prediction data corresponding to the current input data based on the past input data, and the input data processed earlier in time and processed earlier in time. A threshold value used to adaptively determine the presence or absence of noise based on the predicted data corresponding to the input data, and the difference between the current input data and the predicted data corresponding to the current input data Is compared with the threshold value, and either the current input data or the prediction data corresponding to the current input data is selected based on the comparison result.
[0013]
Here, when the threshold value is predicted, the threshold value is predicted based on past input data and past prediction data corresponding to the past input data. The prediction data is obtained by performing a least square operation on the past input data, adaptively determining a prediction coefficient, and performing prediction using the current input data and the prediction coefficient.
[0014]
The threshold value is stored in the input data and the prediction data, respectively, and is adaptively changed based on the difference between the stored input data and the prediction data.
[0015]
In other words, the noise removal device of the present invention is a device that removes noise from a digital signal, and estimates a signal estimation unit that estimates a signal to be processed and a criterion for the accuracy of the output of the signal estimation unit. And a switching means for switching between the output of the signal estimation means and the input signal for processing based on the outputs of the signal estimation means and the threshold value estimation means.
[0016]
Here, a memory for accumulating past input signals and past signal estimation means is provided, and the signal estimation means and threshold value estimation means perform estimation using the accumulated past signals. I have to.
[0018]
Further, the signal estimation means is a least square prediction coefficient determination means and a linear prediction means, and the threshold value estimation means is an absolute value of a difference between a past input signal value and a past prediction value accumulated corresponding thereto. A value proportional to the sum is used as a threshold value.
[0019]
[Action]
According to the present invention, the threshold value is adaptively predicted based on the input data and the corresponding prediction data, and the threshold value is compared with the difference between the current input data and the prediction data. At this time, if the input data includes, for example, noise with strong impulse characteristics, the value of the difference between the current input data and the prediction data is separated from the threshold value. In this case, by outputting prediction data instead of input data, the obtained data becomes data without noise (data from which noise has been removed). Further, by removing only the noise having a strong impulsive property in this way, it is possible to prevent a signal not including noise (having less noise) from being processed.
[0020]
【Example】
Embodiments of the present invention will be described below with reference to the drawings.
[0021]
As shown in FIG. 1, the noise removal apparatus according to the embodiment to which the noise removal method of the present invention is applied includes a blocking processor 3 that blocks input data supplied via a terminal 11, and the blocking processing described above. A data memory 9 (first storage means) for storing the block data from the unit 3 and a coefficient calculator 13 for adaptively determining a prediction coefficient by applying a least square operation to the past block data from the data memory 9 And a signal value predictor 4 for multiplying the current block data from the block processor 3 by the prediction coefficient (the first prediction means comprises the coefficient calculator 13 and the signal value predictor 4), A data memory 10 (second storage means) for storing prediction data from the signal value predictor 4, block data from the data memory 9, and block data from the data memory 9 A threshold predictor 5 (second predictor) that adaptively predicts a threshold based on the prediction data stored in the corresponding data memory 10, and the input data and the prediction of the signal value predictor 4 Comparator 7 that compares the difference with the data with the threshold value from the threshold predictor 5, and selects either the input data or the predicted data based on the comparison result of the comparator 7 (for example, the difference > Selector 6 for selecting predicted data when threshold value and selecting input data when difference ≦ threshold value.
[0022]
The threshold predictor 5 adaptively changes the threshold based on the difference between the block data from the data memory 9 and the predicted data from the data memory 10.
[0023]
In this embodiment, a digital image signal is assumed as input data, and the case where noise is noise due to a random bit error during reproduction of a digital VTR (a VTR that does not compress an image signal) is described.
[0024]
In FIG. 1, the input terminal 11 is supplied with a one-dimensional digital image signal which is reproduced from a recording medium (magnetic tape) by a digital VTR and subjected to operation conversion through a predetermined process according to the preceding configuration (not shown). Therefore, the digital image signal supplied to the input terminal 11 may contain noise due to random bit errors during VTR playback.
[0025]
The digital image signal supplied to the input terminal 11 is sent to the blocking processor 3. The block processor 3 forms a two-dimensional block of 8 lines × 8 pixels, for example, and performs order conversion so that the image signal in the two-dimensional block becomes a temporally continuous one-dimensional signal. Note that the size of blocking can be arbitrarily selected, and can be 3 × 3 pixels, for example.
[0026]
The output data si3 from the block processor 3 is sent to a signal value predictor 4 and a data memory 9 described later. As will be described later, the data memory 9 controls whether or not to store the data si3 from the blocking processing circuit 3 in accordance with the comparison result signal si7 from the comparator 7. .
[0027]
Data si9 read from the data memory 9 is sent to the coefficient calculator 13. The coefficient calculator 13 uses the data si9 read out from the data memory 9 (that is, data, for example, one block before the current output data si3 from the block processor 3), and uses, for example, the least square method. To calculate the prediction coefficient.
[0028]
In other words, the prediction coefficient is determined by the least-squares method using data that has been processed earlier in time and determined from data determined not to be noise, and that has been stored in the memory 9. It is a thing.
[0029]
The prediction coefficient from the coefficient calculator 13 is sent to the signal value predictor 4.
[0030]
The signal value predictor 4 is also supplied with the current output data si3 from the block processor 3 together with the prediction coefficient obtained by the coefficient calculator 13, and the signal value predictor 4 receives the data. An operation of multiplying si3 by the prediction coefficient is performed. That is, the signal value predictor 4 multiplies the output data si3 related to the current processing from the blocking processor 3 by the prediction coefficient obtained from the past data si3 stored in the data memory 9, The current prediction data (signal estimation value) si4 is obtained.
[0031]
Here, the flow of processing in the coefficient calculator 13 and the signal value predictor 4 will be described with reference to the flowchart of FIG.
[0032]
In FIG. 2, in step S <b> 1, the past output data si <b> 3 of the blocking processor 3 stored in the data memory 9 is made into the form of simultaneous linear equations. In the next step S2, a prediction coefficient (coefficient solution) is obtained by solving the simultaneous linear equations. The processing of step S1 and step S2 is performed by the coefficient calculator 13.
[0033]
In step S3, the present output data si3 from the blocking processor 3 is multiplied by the coefficient solution (prediction coefficient). The signal value predictor 4 performs the process in step S3.
[0034]
Prediction data si4 from the signal value predictor 4 obtained as described above is sent to the comparator 7 and also to the data memory 10.
[0035]
Data si10 (predicted data) read from the data memory 10 is sent to the threshold predictor 5. The threshold predictor 5 is also supplied with output data si9 from the data memory 9. However, in this case, the data si10 read from the data memory 10 is data corresponding to the output data si9 from the data memory 9.
[0036]
In the threshold predictor 5, the output data si9 (data processed earlier in time) from the data memory 9 and the data for all the data stored in the data memories 9 and 10 and the data Subtraction is performed with the data si10 from the data memory 10 corresponding to si9, the sum of absolute values of the subtraction values is taken, and division is performed by the total number of data, and then a predetermined value is determined for the value. And a value obtained thereby is output as the threshold value si5.
[0037]
That is, the processing in the threshold predictor 5 is as shown in the flowchart of FIG.
[0038]
In FIG. 3, in step S11, the output data si4 (si10) of the signal value predictor 4 stored in the data memory 10 is subtracted from the output data si3 (si9) of the blocking processor 3 stored in the data memory 9. To do.
[0039]
In the next step S12, the absolute value of the subtraction value is taken, and in step S13, the data sum of all of them (for all the data stored in the data memories 9 and 10) is obtained.
[0040]
In step S14, the total value is divided by the total number of data, and in step S15, the divided value is multiplied by a predetermined proportional constant. The proportionality constant is appropriately determined in advance and does not change during processing. As this predetermined proportionality constant, for example, a value larger than 1.0 such as 1.7, 2.0,... Is used.
[0041]
The threshold value si5 obtained by the threshold value predictor 5 as described above and the prediction data si4 from the signal value predictor 4 described above are sent to the comparator 7.
[0042]
The comparator 7 is also supplied with an input digital image signal from the input terminal 11. The comparator 7 takes the difference between the prediction data si4 from the signal value predictor 4 and the image signal (input signal value) from the input terminal 11, and uses the absolute value of the difference as the threshold predictor. Comparison using the threshold value from 5 (comparison with an adaptively determined threshold value).
[0043]
When the comparator 7 determines that the absolute value of the difference is larger than the threshold value in the comparison, the comparator 7 indicates that the signal from the input terminal 11 includes noise (impulsive noise). Conversely, when it is determined that the absolute value of the difference is not greater than the threshold value, comparison result data si7 indicating that the signal from the input terminal 11 does not contain noise is output. The comparison result data si7 is, for example, 1 when indicating that noise is included, and 0 when indicating that noise is not included. Of course, it can be 0 when indicating that noise is included, and 1 when indicating that noise is not included.
[0044]
The data si7 is sent to the selector 6 as a selection control signal. The selector 6 selects the prediction data si4 from the signal value predictor 4 and the input data from the input terminal 11 according to the selection control signal, and the selection control signal is the above-mentioned selection control signal. When the signal indicates that it includes noise, the prediction data si4 from the signal value predictor 4 is conversely. When the signal indicates that the selection control signal does not include noise, the input terminal 11 Select input data from.
[0045]
Therefore, the signal selected by the selector 6 becomes image data that does not contain noise (that is, noise is removed), and the data from which this noise has been removed is taken out from the terminal 12.
[0046]
The data si7 from the comparator 7 is also sent to the data memory 9 as a control signal as to whether or not the data si3 from the blocking processor 3 is stored. That is, when the data si7 indicates that it includes noise, the data si3 including the noise (data si9) is used by preventing the output data si3 from the blocking processor 3 from being stored in the data memory 9. Thus, the calculation processing by the coefficient calculator 13 is prevented from being performed. Thereby, it can prevent that the prediction calculation using the prediction coefficient based on the signal containing noise in the signal value predictor 4 is performed. The data si7 can also be considered as a control signal for determining whether or not to read data from the data memory 9.
[0047]
Note that the coefficient calculator 13 and the threshold predictor 5 of the noise removal apparatus of this embodiment hold a predetermined prediction coefficient and threshold as initial values. Therefore, for example, at the initial time such as when the power is turned on, the prediction by the signal value predictor 4 is performed by the prediction coefficient predetermined and held in the coefficient calculator 13 and the threshold value prediction is performed. The comparator 7 makes a comparison (determining whether or not there is noise) based on a predetermined threshold value stored in the comparator 5. Thereafter, as described above, the prediction data is stored in the data memory 10 and the next block data is processed.
[0048]
That is, in the noise removing apparatus of this embodiment, by performing the processing as described above, relatively conspicuous large noise is removed, and at the same time, the input signal and its predicted data (estimated signal) are gradually stored in the data memory. 9 and 10, and after the data is accumulated to some extent in the data memories 9 and 10 in this way, the prediction coefficient corresponding to the input data to be processed at the present time is set as the past coefficient. Determine from data using least squares method. The prediction data and the input data predicted by the prediction coefficient determined by the least square method are threshold values determined by the absolute value of the difference between the past prediction data (estimated value) and the corresponding input data, that is, a value proportional to the error. And whether it is noise or not is determined. By repeating such an operation, the signal that is gradually removed approaches true noise (noise with a strong impulse).
[0049]
In other words, the noise removal apparatus according to the embodiment of the present invention estimates the time or position where noise is present in advance, adaptively determines the presence or absence of noise according to the nature of the noise, and selectively processes the presence or absence of noise. By doing so, a noise-free signal is obtained. That is, in the noise removal apparatus according to the present embodiment, the signal value that should be present is estimated from the past property of the signal value, and the difference between the estimated value and the actual signal value exceeds a variation that can be considered from the property of the signal. By performing correction only on the signal, correction is made only to noise that deviates significantly from the nature of the signal (noise with strong impulse characteristics), and inadvertent signal deterioration can be suppressed.
[0050]
【The invention's effect】
As described above, in the present invention, the prediction data corresponding to the current input data is obtained based on the past input data, and the threshold value is predicted based on the input data and the prediction data corresponding to the input data. The difference between the current input data and the predicted data corresponding to the current input data is compared with the threshold value, and the current input data or the predicted data corresponding to the current input data is compared based on the comparison result. By selecting one of them, for example, it is possible to remove only isolated noise with strong impulse characteristics.
[Brief description of the drawings]
FIG. 1 is a block circuit diagram showing a schematic configuration of a noise removing apparatus according to an embodiment of the present invention.
FIG. 2 is a flowchart showing a process for obtaining prediction data in the embodiment.
FIG. 3 is a flowchart illustrating processing for predicting a threshold value in the present embodiment.
[Explanation of symbols]
3 ... Blocking processor 4 ... Signal value predictor 5 ... Threshold predictor 6 ... Selector 7 ... Comparators 9 and 10 ... Data memory 13 ... Coefficient calculator

Claims (7)

First prediction means for performing prediction based on past input data and calculating prediction data corresponding to the current input data;
Predicts a threshold value used to adaptively determine the presence or absence of noise based on input data processed earlier in time and predicted data corresponding to the input data processed earlier in time A second prediction means;
A comparison means for comparing a difference between the current input data and the prediction data corresponding to the current input data with a threshold value from the second prediction means;
A noise removal apparatus comprising: selection means for selecting either the current input data or prediction data corresponding to the current input data based on a comparison result of the comparison means. The noise removing apparatus according to claim 1, wherein the second predicting unit predicts the threshold value based on past input data and past predicted data corresponding to the past input data. . The first prediction means may be composed of a coefficient calculator for determining adaptively prediction coefficients by performing a least-squares calculation to the past input data, the signal value predictor for performing prediction using the prediction coefficients The noise removing apparatus according to claim 1. First storage means for storing input data;
A second storage means for storing prediction data from the first prediction means;
The second prediction means adaptively changes the threshold value based on a difference between data from the first storage means and prediction data from the second storage means. The noise removing device according to 1.   The first prediction means performs prediction based on input data determined to have no noise and prediction data corresponding to the input data, and calculates prediction data corresponding to the current input data. The noise removal device according to claim 1.   The second prediction means predicts a threshold value for adaptively determining the presence or absence of noise based on input data determined not to have noise and prediction data corresponding to the input data. Item 2. The noise removing device according to Item 1. Based on the past input data, find the prediction data corresponding to the current input data,
Based on the input data processed earlier in time and the prediction data corresponding to the input data processed earlier in time, a threshold used to adaptively determine the presence or absence of noise is predicted. ,
The difference between the current input data and the predicted data corresponding to the current input data is compared with the threshold value,
A noise removal method, comprising: selecting either the current input data or the prediction data corresponding to the current input data based on the comparison result.

Images