JP2001285882A - Device and method for reducing noise - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve the prediction accuracy and to enhance the noise reduction performance at application of a classification adaptive prediction processing to a digital information signal that is subjected to coding decoding processing. SOLUTION: A decoder 1 outputs a decoded image signal and decoding-use attached information. The attached information means signal type information, image format information, image quality information, and a motion vector or the like. A class based on the attached information is created. An area segmentation section 4 extracts pixel data of a class tap to extract a feature quantity. A class code is generated on the basis of the attached information class and the feature quantity. A prediction coefficient ROM 8 outputs a prediction coefficient set corresponding to the received class code to a prediction arithmetic section 9. The prediction coefficient is decided in advance by learning and stored. Using the pixel data of the prediction tap extracted by a prediction tap data generating section 5 and the prediction coefficient set received from the ROM 8 to apply a prediction arithmetic operation to an output image signal of the decoder 1 can generate a signal whose noise is reduced.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a noise reduction apparatus and method for reducing noise after decoding an encoded digital image signal or an encoded digital audio signal.

[0002]

2. Description of the Related Art MPEG2 (Moving Picture Expert Group phase) is one of the compression coding methods for image signals.
2) is used. In a transmission / reception or recording / reproducing system based on MPEG2, an image signal is subjected to compression / encoding processing based on MPEG2 to be transmitted or recorded, and a received / reproduced image signal is subjected to decompression corresponding to compression / encoding processing based on MPEG2. By decoding, the original image signal is restored.

[0003] In the encoding process by MPEG2, in order to make the encoding process versatile and to improve the efficiency of compression by encoding, together with the encoded image data,
The additional information for the decoding process is transmitted. Additional information
It is inserted into the header in the MPEG2 stream and transmitted to the decoding device.

[0004] In addition to MPEG, the characteristics of image signals obtained by decoding differ greatly depending on the coding / decoding system applied. For example, physical characteristics (frequency characteristics and the like) greatly differ depending on signal types such as a luminance signal, a color difference signal, and three primary color signals. This difference will remain in the decoded signal that has undergone the encoding / decoding process. In general, in image coding / decoding processing, the number of pixels to be coded is often reduced by introducing space-time thinning processing. The characteristics of the spatio-temporal resolution of an image greatly differ depending on the thinning method. Further, even when the difference between the spatio-temporal resolution characteristics is small, the image quality characteristics such as S / N and the amount of coding distortion vary greatly depending on the compression rate (transmission rate) condition in coding.

The applicant of the present application has previously proposed a classification adaptive processing. This is because, in a learning process (offline), a prediction coefficient is obtained for each class using an actual image signal (teacher signal and student signal) and accumulated, and in an actual image conversion process, an input image signal is obtained. , And an output pixel value is obtained by a prediction operation of a prediction coefficient corresponding to the class and a plurality of pixel values of the input image signal. The class is determined according to the distribution and waveform of pixel values in the spatial and temporal vicinity of the pixel to be created. Calculates prediction coefficients using actual image signals and calculates prediction coefficients for each class to prevent degradation in resolution of input image signals compared to noise reduction processing using low-pass filters with fixed coefficients It has the characteristic that the noise can be reduced while performing.

[0006]

By applying the classification adaptive processing to the decoded image signal,
When the noise is reduced, the target image signal has the above-described characteristic difference. As a result, the prediction accuracy of the classification adaptive processing is reduced, and there is a problem that sufficient noise reduction performance cannot be obtained.

Further, in the classification adaptive processing, the prediction performance can be improved by introducing the motion information of the target image signal into the class. As the motion information, an expression form of detailed motion information such as a motion vector is effective. However, when detecting a motion vector from an image signal that has undergone encoding / decoding processing, the detection accuracy of the motion vector is reduced due to distortion of the decoded image signal, and
There is a problem that a large amount of arithmetic processing is required for motion vector detection.

Accordingly, it is an object of the present invention to perform a noise reduction process satisfactorily by performing a classification adaptive process using additional information on a digital information signal that has undergone encoding and decoding processes. An object of the present invention is to provide a noise reduction device and method.

[0009]

SUMMARY OF THE INVENTION In order to solve the above-mentioned problems, the present invention reduces the noise of an input digital information signal generated by decoding an encoded digital information signal. And a class information generating means for generating class information based on the additional information for decoding processing, storing a predetermined prediction coefficient, and generating class information from the stored prediction coefficients. Coefficient storage means for outputting a prediction coefficient corresponding to the output of the means, area extraction means for extracting an area consisting of a plurality of samples from the input digital information signal, and a plurality of samples and prediction coefficients extracted by the area extraction means. And a calculation processing means for performing a prediction calculation based on the calculation result to generate a sample value.

The invention according to claim 17 is a noise reduction method for reducing noise of an input digital information signal generated by decoding an encoded digital information signal. A class information generating step of generating class information based on
Storing a predetermined prediction coefficient, from among the stored prediction coefficients, outputting a prediction coefficient corresponding to the class information, and extracting a region consisting of a plurality of samples from the input digital information signal; And performing a prediction operation on the basis of the plurality of samples extracted in the region extracting step and the prediction coefficient to generate a sample value.

According to the first and 17th aspects of the present invention, by using the additional information for the decoding processing, it is possible to improve the prediction accuracy in the noise reduction processing to which the classification adaptive processing is applied.

According to a second aspect of the present invention, there is provided a noise reduction apparatus for reducing noise of an input digital information signal generated by decoding an encoded digital information signal. First region extracting means for extracting an area composed of samples, feature amount extracting means for extracting a characteristic amount based on a sample from the first region extracting means, and a feature amount and additional information for decoding processing. Class information generating means for generating class information, and storing a predetermined prediction coefficient,
A coefficient storage unit that outputs a prediction coefficient corresponding to the output of the class information generation unit from the stored prediction coefficients, a second region extraction unit that extracts a region including a plurality of samples from the input digital information signal, 2 based on the plurality of samples extracted by the region extracting means and the prediction coefficient,
And a calculation processing means for performing a prediction calculation to generate a sample value.

[0013] The invention of claim 18 is a noise reduction method for reducing noise of an input digital information signal generated by decoding an encoded digital information signal. A first area segmentation step of extracting an area composed of samples, a feature quantity extraction step of extracting a feature quantity based on the sample extracted in the first area segmentation step, and a feature quantity and decoding process. A class information generating step of generating class information based on the additional information of
Storing a predetermined prediction coefficient, outputting a prediction coefficient corresponding to the class information from the stored prediction coefficients, and extracting a second region from the input digital information signal by extracting a region including a plurality of samples; Steps and
Performing a prediction operation on the basis of the plurality of samples extracted in the step of extracting the second area and the prediction coefficient to generate a sample value.

According to the second and eighteenth aspects of the present invention, it is possible to perform the classification adaptive processing using the additional information for the decoding processing together with the characteristic amount of the input digital information signal, and the classification adaptive processing is applied. The prediction accuracy in the noise reduction processing can be improved.

According to a fourth aspect of the present invention, there is provided a noise reduction apparatus for reducing noise of an input image signal generated by decoding an encoded digital image signal. Class information generating means for generating class information based on the stored prediction coefficients, from among the stored prediction coefficients, a coefficient storage means for outputting a prediction coefficient corresponding to the output of the class information generation means, Region extracting means for extracting an area consisting of a plurality of pixels from an input image signal, and an arithmetic processing means for performing a prediction operation to generate a pixel value based on the plurality of pixels and the prediction coefficient extracted by the area extracting means; A noise reduction device characterized by having:

According to a nineteenth aspect of the present invention, there is provided a noise reduction method for reducing noise of an input image signal generated by decoding an encoded digital image signal. A step of generating class information based on the class information; a step of storing a predetermined prediction coefficient; and a step of outputting a prediction coefficient corresponding to the class information from the stored prediction coefficients. A region extracting step of extracting a region consisting of the pixels of the above, and a step of performing a prediction operation to generate a pixel value based on the plurality of pixels extracted in the region extracting step and the prediction coefficient. This is a noise reduction method.

According to the fourth and 19th aspects of the present invention, by using the additional information for the decoding processing, it is possible to improve the prediction accuracy in the noise reduction processing to which the classification adaptive processing is applied.

According to a fifth aspect of the present invention, there is provided a noise reduction apparatus for reducing noise of an input image signal generated by decoding an encoded digital image signal. A first region extracting unit for extracting a region, a feature amount extracting unit for extracting a feature amount based on pixel data from the first region extracting unit, and a feature amount and additional information for decoding processing. Class information generating means for generating class information, coefficient storing means for storing a predetermined prediction coefficient, and outputting a prediction coefficient corresponding to the output of the class information generation means from the stored prediction coefficients, and an input image A second area extracting unit for extracting an area including a plurality of pixels from the signal, and a prediction operation based on the plurality of pixels and the prediction coefficient extracted by the second area extracting unit. A noise reducing apparatus characterized in that it comprises a processing means for generating a pixel value by performing a.

According to a twentieth aspect of the present invention, there is provided a noise reduction method for reducing noise of an input image signal generated by decoding an encoded digital image signal. A first area segmentation step of extracting a pixel area, a feature quantity extraction step of extracting a feature quantity based on the pixel data extracted in the first area segmentation step, and a feature quantity and decoding process. Generating class information based on the additional information of the class information, storing a predetermined prediction coefficient, and outputting a prediction coefficient corresponding to the class information from the stored prediction coefficients; and A second region extraction step of extracting a pixel region composed of a plurality of pixels from the image signal, and a plurality of pixel regions extracted in the second region extraction step On the basis of the prediction coefficient element, the noise reduction method is characterized by comprising the step of generating a pixel value by performing a prediction calculation.

According to the fifth and twentieth aspects of the present invention, it is possible to perform the classification adaptive processing using the characteristic information of the input digital image signal and the additional information for the decoding processing. The prediction accuracy in the noise reduction processing can be improved.

[0021]

DESCRIPTION OF THE PREFERRED EMBODIMENTS One embodiment of the present invention will be described below. First, a configuration related to generation of a predicted image signal will be described with reference to FIG. An input bit stream is supplied to the decoder 1. Here, the input bit stream is transmitted and received by a transmitting / receiving system (or a recording / reproducing system,
Hereinafter, the same applies. ), Image data compressed and encoded by MPEG2 and other data such as additional information. The decoder 1 outputs a decoded image signal and additional information for decoding.

The additional information is additional information necessary for the decoding process.
Inserted in the headers of the P layer and the picture layer, the decoder 1 performs decoding processing using the additional information, and separates and outputs the additional information.

The additional information is supplied to the additional information extracting unit 2, and the additional information used for the classification adaptive processing is selectively output from the additional information extracting unit 2. The extracted additional information is supplied to the additional information class generator 3. For example, there is the following as additional information used in the classification adaptive processing.

(1) Signal type information: each component of a component signal (Y, U, V components, Y, Pr, Pb
(2) Image format information: Interlace / progressive identification information, field or frame frequency (time resolution information), image size information of horizontal pixels and vertical lines (spatial resolution) Information) 4: 3, 16: 9, etc. aspect ratio information (3) Image quality information: Transmission bit rate (compression rate) information (4) Motion vector: Horizontal and vertical motion amount information There are various types, and decoding on the receiving side is realized by transmitting various control signals including the above-described additional information. The signal characteristics of the decoded image signal greatly differ depending on various specifications and attributes indicated by the above-described additional information. Therefore, by introducing this characteristic information into the classification adaptive processing, the prediction performance is improved.

The decoded image signal from the decoder 1 is supplied to an area extracting section 4 and a prediction tap data generating section 5.
The region cutout unit 4 extracts a region including a plurality of pixels from the input image signal and supplies pixel data relating to the extracted region to the feature amount extraction unit 6. The feature amount extraction unit 6 generates an ADRC code by performing processing such as 1-bit ADRC on the supplied pixel data, and supplies the generated ADRC code to the class code generation unit 7. The plurality of pixel regions extracted in the region extracting unit 4 are called class taps. A class tap is an area composed of a plurality of pixels existing in a spatial and / or temporal vicinity of a target (target) pixel. As described later, the class is determined for each attention (target) pixel.

ADRC calculates the maximum value and the minimum value of pixel values in a class tap, obtains a dynamic range that is the difference between the maximum value and the minimum value, and requantizes each pixel value according to the dynamic range. It is. 1 bit A
In the case of DRC, the pixel value is converted into one bit depending on whether it is larger or smaller than the average value of a plurality of pixel values in the tap. The ADRC process is a process for making the number of classes representing the level distribution of pixel values relatively small.
Therefore, the present invention is not limited to ADRC, and may use encoding for compressing the number of bits of a pixel value such as vector quantization.

The additional information class generated on the basis of the additional information in the additional information class generator 3 is also supplied to the class code generator 7. The class code generation unit 7
Based on the additional information class and the ADRC code, a class code indicating the result of the class classification is generated, and the class code is supplied to the prediction coefficient ROM 8 as an address. R
The OM 8 outputs a prediction coefficient set corresponding to the supplied class code to the prediction calculation unit 9. The prediction coefficient set is
It is determined in advance by a learning process described later, and for each class,
More specifically, it is stored in the prediction coefficient ROM 8 in a form using a class code as an address. The prediction coefficients may be stored in a memory having a RAM configuration from which the prediction coefficients can be downloaded from the outside.

On the other hand, the prediction tap data generator 5 extracts a predetermined area (prediction tap) including a plurality of pixels from the input image signal, and supplies the extracted prediction tap pixel data to the prediction calculator 9. The prediction tap is an area including a plurality of pixels existing in the spatial and / or temporal vicinity of the target (target) pixel, like the class tap. The prediction operation unit 9 performs a product-sum operation according to the following equation (1) based on the pixel data supplied from the prediction tap data generation unit 5 and the prediction coefficient set supplied from the ROM 8 to obtain a predicted pixel value. And outputs a predicted pixel value.
The prediction tap and the class tap described above may be the same or different.

Y = w ₁ × x ₁ + w ₂ × x ₂ + ‥‥ + w _n × x _n (1) where x ₁ , ‥‥, and x _n are pixel data of prediction taps, and w ₁ , ‥‥ and w _n are prediction coefficient sets. The prediction operation is not limited to the linear expression represented by the expression (1), but may be a second-order or higher-order expression, or may be nonlinear.

The predicted image signal is obtained by reducing noise in the output image signal of the decoder 1. Unlike classification noise removal using a low-pass filter with fixed coefficients, the classification adaptive processing uses prediction coefficients that are obtained in advance using actual image signals, so that noise is reduced without deteriorating resolution. This is a process that can be performed.

FIG. 2 shows an example of an arrangement of class taps extracted by the area extracting section 4. A class tap is set by a total of seven pixels including the pixel of interest and a plurality of pixels around the pixel of interest in the decoded image signal. FIG. 3 shows an example of the arrangement of prediction taps output from the prediction tap data generation unit 5. In the decoded image signal, a prediction tap is set by a total of 13 pixels including a target pixel and a plurality of pixels around the target pixel. 2 and 3
, The solid line indicates the first field, and the broken line indicates the second field.
Indicates a field. Further, the illustrated arrangement of taps is merely an example, and various arrangements can be used.

Next, referring to FIG. 4, a class code (prediction coefficient ROM) formed in the class code
An example of the prediction coefficient stored in the prediction coefficient ROM 8 will be described. In the class information shown in FIG. 4, a signal type class, a format class, a compression rate (transmission rate) class, a first motion vector class, and a second motion vector class are generated by the additional information class generation unit 3. Class. The signal feature class is
A class based on the feature amount extracted by the feature amount extraction unit 6, for example, an ADRC class. In the table of FIG. 4, the leftmost signal type class is the highest order of the address, and the rightmost signal feature class is the lowest order.

The signal type classes are, for example, two types of Y, U, V and Y, Pr, Pb. Prediction coefficients are separately obtained for each signal type.
0 and K1. The format class corresponds to the spatio-temporal resolution characteristic of the image to be processed, and is, for example, of two types.
The class of F1 is defined. For example, for an interlaced image, F0, for a progressive image, F0
One class is assigned. Other examples of image format classes are field or frame frequency, horizontal pixels or vertical lines. As an example, F0,
The larger the F1, F2,... Number, the higher the spatio-temporal resolution.

The compression rate (transmission rate) class is a class based on image quality information, and includes i types of classes R0 to Ri-1.
Is prepared. The higher the compression ratio, the greater the amount of coding distortion. The first motion vector class is a class corresponding to a motion vector between a frame including the pixel of interest (current frame) and a temporally previous frame, and j types are prepared. The second motion vector class is a class corresponding to a motion vector between a frame including the target pixel (current frame) and a temporally later frame, and m types are prepared. The compression ratio class and the first and second motion vector classes may be individual values. In that case, however, the number of classes is large, so that they are grouped into a plurality of representative values. For example, one representative value may be set for each of a plurality of ranges formed by appropriate threshold values, and a class corresponding to the representative value may be set. More specifically, a three-stage class of still, small, and large motions may be formed from motion vectors expressing horizontal and vertical motions.

The above five types of classes are classes generated by the additional information class generation unit 3. However, the classes described above are merely examples, and only some of the classes may be used. For example, only the additional information class may be used as a class. Then, a signal feature class (for example, a class based on an ADRC code) generated by the feature extractor 6 is added to the lower side of the five types of classes described above. As the signal feature class, k kinds are prepared.

As described above, five types of additional information classes and one
A prediction coefficient set is stored in the ROM 8 for each class determined by the type of signal feature amount class. Equation (1) described above
When performing the prediction operation represented by, w ₁ , w ₂ , ‥‥,
There are n sets of prediction coefficients of w _n for each class.

Another embodiment of the present invention will be described with reference to FIG. Parts corresponding to FIG. 1 showing the configuration of one embodiment are denoted by the same reference numerals. Another embodiment is based on the characteristics of the decoded image signal from the decoder 1,
The prediction performance of the classification adaptive processing is improved by changing the data extraction method for class classification and the structure of the prediction tap.

In order to change the class tap structure for extracting the characteristic amount of the decoded image signal based on the additional information extracted by the additional information extracting unit 2, as shown in FIG. The pattern of the class tap to be performed is switched. The feature extraction unit 6 uses ADR
When extracting a waveform and a level distribution as a feature value by C, the size of the region to be subjected to ADRC is changed according to the spatial resolution of the target image. Further, since the signal characteristics vary depending on the type of signal, the class tap structure is changed. Furthermore, it is also possible to change the class tap structure according to the aspect ratio of the image.

The additional information also includes a compression rate (transmission rate information) indicating distortion of an image due to encoding and decoding.
Information on the compression ratio can be extracted from the additional information. It is difficult to detect the amount of coding distortion in an image signal once decoded. When the classification adaptive processing is applied to signals having different encoding distortion amounts, it is difficult to improve the prediction performance. Therefore, the configuration of the class tap is changed according to the compression rate (transmission rate information). Furthermore, by changing the configuration of class taps based on the first and second motion vector information, it is possible to realize a class tap structure having high spatio-temporal correlation characteristics. For example, in the case of stillness, a class tap is formed in a frame, and when there is movement, a class tap is formed in the previous and next frames in addition to the current frame.

Further, as shown in FIG. 5, the class code formed by the class code generator 7 is supplied to the prediction tap data generator 5 as a control signal. As a result, an optimal prediction tap pattern is set for each class in which additional information is added as shown in FIG. The structure of the prediction tap is changed according to the additional information in the class in the same way as the above-described class tap structure is changed by the additional information, and the prediction tap is changed by changing the prediction tap similarly to the case of the class tap. Performance can be improved.

FIG. 6 schematically shows an example in which the area of the tap (class tap or prediction tap) is changed according to the additional information. FIG. 6 shows an example in which a spatio-temporal tap is set by spatial taps belonging to the current frame and the frames before and after the current frame, and a frame of a broken line represents a tap region.

FIG. 6 shows a case where the motion vector MV1 between the previous frame and the current frame is downward, and the motion vector MV2 between the current frame and the next frame is upward. These motion vectors MV1 and MV
The tap area is corrected according to 2. By this motion correction, it is possible to configure a tap using a plurality of pixels having a strong correlation. Further, the area of the range including the tap is changed according to the image format information, for example, the spatial resolution information F0, F1, F2. Spatial resolution information F
0, F1, and F2 are included in the class information generated by the class code generation unit 7 as the additional information or the additional information class of interest. In the example of FIG. 4 described above, F0, F1
There are two types of classes.

As an example, F0 has the lowest spatial resolution, F1 has an intermediate spatial resolution, and F2 has the highest spatial resolution. As the spatial resolution increases, the area including the tap is gradually enlarged. When the spatial resolution is low, the range in which pixels having a strong correlation exists is narrow, so that the tap region is also narrow. Thereby,
The performance of the noise reduction processing by the classification adaptive processing can be improved.

Next, learning, that is, a process of obtaining a prediction coefficient for each class will be described. Generally, an image signal (hereinafter, referred to as a teacher signal) having the same signal format as an image signal to be predicted by the classification adaptive processing, and a processing targeted for the classification adaptive processing in the teacher signal (ie, A prediction coefficient is determined by a predetermined calculation process based on an image signal (student signal) obtained by performing a process related to the noise reduction process). In the classification adaptive processing performed on an image signal that has been subjected to encoding / decoding of an image signal according to the MPEG2 standard or the like, learning is performed by, for example, a configuration illustrated in FIG. FIG. 7 shows a configuration for learning prediction coefficient data in another embodiment shown in FIG.

For learning, a teacher signal and an input image signal are used. The teacher signal is a signal that has no noise or has less noise than the input image signal. The input image signal may be formed by adding noise to the teacher signal. The input image signal is converted by the encoder 21 into, for example, an MPE.
G2. The output signal of the encoder 21 corresponds to the received signal in FIG. An output signal of the encoder 21 is supplied to a decoder 22. The decoded image signal from the decoder 22 is used as a student signal. Decoder 2
2 is added to the additional information extracting unit 23 for decoding.
And additional information is extracted.

The extracted additional information is supplied to an additional information class generator 24 and an area extractor 25. As described above, the additional information includes signal type information, image format information, image quality information, motion vectors, and the like.

The decoded image signal from the decoder 22, that is, the student signal, is supplied to the area extracting section 25 and the prediction tap data generating section 26. As in the configuration of FIG. 5, the area cutout unit 25 is controlled by the additional information extracted by the additional information extraction unit 23, and the prediction tap data generation unit 26 generates the additional information of the class generated by the class code generation unit 28. Controlled by class. Thereby, a tap can be set by a plurality of pixels having high temporal and / or spatial correlation. The data of the class tap extracted by the region extracting unit 25 is supplied to the feature amount extracting unit 27, and the feature amount extracting unit 27 extracts a feature amount by processing such as ADRC. This feature amount is supplied to the class code generation unit 28. Class code generator 28
Generates a class code representing the result of the classification based on the additional information class and the ADRC code. The class code is supplied to the normal equation adding unit 29.

On the other hand, the pixel data of the prediction tap extracted by the prediction tap data generation unit 26 is supplied to the normal equation addition unit 29. The normal equation addition unit 29 performs a predetermined operation process based on the output of the prediction tap data generation unit 26 and the teacher signal to generate a normal coefficient that sets a prediction coefficient set corresponding to the class code supplied from the class code generation unit 28 as a solution. Generate equation data. Normal equation adder 2
9 is supplied to the prediction coefficient calculation unit 30.

The prediction coefficient calculation unit 30 performs an operation for solving a normal equation based on the supplied data. The prediction coefficient set calculated by this calculation process is supplied to the memory 31 and stored. Prior to performing the image conversion process related to the prediction estimation, the contents stored in the memory 31 are loaded into the prediction coefficient ROM 8 in FIG.

The normal equation will be described below. In the above equation (1), before learning, the prediction coefficient sets w ₁ , ‥
‥ and w _n are undetermined coefficients. Learning is performed by inputting a plurality of teacher signals for each class. When the number of types of the teacher signal is denoted by m, the following equation (2) is obtained from the equation (1).
Is set.

Y _k = w ₁ × x _k1 + w ₂ × x _k2 + ‥‥ + w _n × x _kn (2) (k = 1, 2, ‥‥, m)

If m> n, the prediction coefficient sets w ₁ , ‥
Since ‥ and w _n are not uniquely determined, the element e _k of the error vector e is defined by the following equation (3), and the prediction coefficient set is set so as to minimize the error vector e defined by the equation (4). To be determined. That is, the so-called minimum 2
A prediction coefficient set is uniquely determined by multiplication.

E _k = y _k − {w ₁ × x _k1 + w ₂ × x _k2 + Δ + w _n × x _kn } (3) (k = 1, 2, Δm)

[0054]

(Equation 1)

As a practical calculation method for obtaining a prediction coefficient set that minimizes e ² in Equation (4), partial differentiation of e ² with a prediction coefficient w _i (i = 1, 2 ‥‥) is performed. (5)), each prediction coefficient w _i such that the partial differential value becomes 0 for each value of _i.
Should be determined.

[0056]

(Equation 2)

A specific procedure for determining each prediction coefficient w _i from equation (5) will be described. If X _ji and Y _i are defined as in equations (6) and (7), equation (5) can be written in the form of a determinant of equation (8).

[0058]

(Equation 3)

[0059]

(Equation 4)

[0060]

(Equation 5)

Equation (8) is generally called a normal equation. The prediction coefficient calculation unit 30 calculates a prediction coefficient w _i by performing a calculation process for solving the normal equation (8) according to a general matrix solution such as a sweeping method.

The generation of the prediction coefficients can also be performed by software processing as shown in the flowchart of FIG. The process is started from step S1, and in step S2, a student signal is generated to generate necessary and sufficient learning data for generating a prediction coefficient. In step S3, it is determined whether sufficient learning data necessary to generate a prediction coefficient has been obtained. If it is determined that sufficient learning data has not been obtained yet, the process proceeds to step S4. I do.

In step S4, a class is determined from the feature amount extracted from the student signal and the additional information. In step S5, a normal equation is generated for each class, and the process returns to step S2 to repeat the same processing procedure, thereby generating a normal equation necessary and sufficient to generate a prediction coefficient set.

If it is determined in step S3 that necessary and sufficient learning data has been obtained, the process proceeds to step S6. In step S6, the prediction coefficient sets w ₁ , w ₂ ,.
.., W _n are generated for each class. Then, at step S7, stores the prediction coefficient set w ₁ to w _n for each class generated in the memory, and ends the learning processing in step S8.

The present invention is not limited to the above-described embodiment of the present invention, and various modifications and applications can be made without departing from the gist of the present invention. For example, M
The present invention is applicable not only to the case of using PEG2 but also to the case of using another encoding method such as MPEG4.

[0066]

As described above, according to the present invention, when the classification adaptive processing is applied to a decoded signal to reduce noise, the attributes and characteristics of the target decoded signal are reduced. By using the additional information for decoding, the prediction accuracy of the classification adaptive processing can be improved, and the performance of the noise reduction processing can be improved. According to the present invention, the use of the additional information for decoding makes it possible to perform class classification reflecting the attributes and characteristics of the target signal, thereby improving the prediction accuracy of the class classification adaptive processing and improving the performance of the noise reduction processing. it can. According to the present invention, by using the additional information for decoding, the attribute of the target signal,
An appropriate prediction tap configuration reflecting characteristics can be configured, the prediction accuracy of the classification adaptive processing can be improved, and the performance of the noise reduction processing can be improved.

Further, according to the present invention, by using the motion vector information of the target decoded signal, a detailed class classification and an appropriate prediction tap configuration can be performed, and the prediction accuracy of the classification adaptive processing can be improved. It is possible to improve the performance of the noise reduction processing. Since the motion vector information transmitted as the additional information is used instead of detecting the motion vector information from the decoded signal, it is possible to avoid an enormous amount of calculation required for detecting the motion vector. In addition, when a motion vector is detected from a decoded signal, the accuracy of the motion vector may be reduced due to coding distortion. According to the present invention, since the motion vector information included in the additional information is used, high-precision motion vector information can be used, whereby the prediction accuracy of the classification adaptive processing can be improved, and the performance of the noise reduction processing can be improved. it can.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a configuration of an embodiment of the present invention.

FIG. 2 is a schematic diagram illustrating an example of a pixel arrangement of a class tap.

FIG. 3 is a schematic diagram illustrating an example of a pixel arrangement of a prediction tap.

FIG. 4 is a schematic diagram illustrating an example of a class based on additional information and a feature amount.

FIG. 5 is a block diagram showing a configuration of another embodiment of the present invention.

FIG. 6 is a schematic diagram for explaining another embodiment of the present invention.

FIG. 7 is a block diagram illustrating an example of a configuration related to a prediction coefficient learning process when performing a class classification adaptive process.

FIG. 8 is a flowchart showing processing when the learning processing is performed by software.

[Explanation of symbols]

1,22 ... decoder, 2,23 ... additional information extraction unit, 3,24 ... additional information class generation unit, 4,25
..Area extraction unit, 5, 26: prediction tap data generation unit, 6, 27: feature amount extraction unit, 7, 28: class code generation unit, 8: prediction coefficient ROM, 9.・ ・
Prediction calculation unit

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5C059 KK01 KK23 MA00 MA19 MA28 PP16 RC12 TA29 TA30 TA69 TC02 TC10 TC12 TC24 TC25 TD09 TD16 UA05 UA12 UA18 UA38

Claims (24)

[Claims] 1. A noise reduction device configured to reduce noise of an input digital information signal generated by decoding an encoded digital information signal, wherein class information is determined based on additional information for decoding processing. Class information generating means for generating a predictive coefficient, a coefficient storing means for storing a predetermined predictive coefficient, and outputting a predictive coefficient corresponding to the output of the class information generating means from the stored predictive coefficients, Area extracting means for extracting an area consisting of a plurality of samples from a digital information signal; and an arithmetic processing for generating a sample value by performing a prediction operation based on the plurality of samples extracted by the area extracting means and the prediction coefficient Means for reducing noise. 2. A noise reduction apparatus for reducing the noise of an input digital information signal generated by decoding an encoded digital information signal, comprising the steps of: First region extracting means for extracting; feature amount extracting means for extracting a feature amount based on a sample from the first region extracting means; class information based on the feature amount and additional information for decoding processing; Class information generating means for generating a predictive coefficient, a coefficient storing means for storing a predetermined predictive coefficient, and outputting a predictive coefficient corresponding to the output of the class information generating means from the stored predictive coefficients, Second area extracting means for extracting an area consisting of a plurality of samples from the digital information signal; A noise reduction apparatus comprising: a calculation processing unit configured to perform a prediction calculation based on the plurality of samples obtained and the prediction coefficient to generate a sample value. 3. The method according to claim 1, wherein the prediction coefficient corresponds to a noise-free teacher signal and the input digital information signal, and is generated in advance using a student signal containing noise. Noise reduction device. 4. A noise reduction device configured to reduce noise of an input image signal generated by decoding an encoded digital image signal, wherein class information is determined based on additional information for decoding processing. Class information generating means for generating, a coefficient storing means for storing a predetermined prediction coefficient, and outputting a prediction coefficient corresponding to the output of the class information generating means from the stored prediction coefficients; and the input image Region extraction means for extracting an area consisting of a plurality of pixels from a signal, and operation processing means for performing a prediction operation to generate a pixel value based on the plurality of pixels extracted by the area extraction means and the prediction coefficient, and A noise reduction device comprising: 5. A noise reduction apparatus configured to reduce noise of an input image signal generated by decoding an encoded digital image signal, wherein a region including a plurality of pixels is extracted from the input image signal. A first region extracting unit, a feature amount extracting unit that extracts a feature amount based on the pixel data from the first region extracting unit, and class information based on the feature amount and the additional information for the decoding process. Class information generating means for generating, a coefficient storing means for storing a predetermined prediction coefficient, and outputting a prediction coefficient corresponding to the output of the class information generating means from the stored prediction coefficients; and the input image A second area extracting means for extracting an area composed of a plurality of pixels from the signal; and a plurality of pixels extracted by the second area extracting means and the prediction coefficient. And a calculation processing means for performing a prediction calculation to generate a pixel value. 6. The noise reduction device according to claim 4, wherein the additional information is information indicating a type of an image signal to be processed. 7. The noise reduction device according to claim 4, wherein the additional information is format information of an image signal to be processed. 8. The noise reduction device according to claim 4, wherein the additional information is image quality information. 9. The noise reduction device according to claim 4, wherein the additional information is motion vector information. 10. The position of an area to be extracted by at least one of the first area extracting means and the second area extracting means in response to motion vector information included in the additional information according to claim 4 or 5. A noise reduction device. 11. The method according to claim 4, wherein at least one of the first area cutout means and the second area cutout means has a size of an area cut out in response to image format information included in the additional information. A noise reduction device characterized by being changed. 12. The noise reduction device according to claim 11, wherein the image format information is time and / or spatial resolution information of the image signal. 13. The noise reduction device according to claim 11, wherein the image format information is aspect information of the image signal. 14. The noise according to claim 4, wherein the prediction coefficient corresponds to a teacher signal without noise and the input image signal, and is generated in advance using a student signal containing noise. Reduction device. 15. A learning class information generating means for generating class information based on additional information for decoding processing attached to the student signal according to claim 14, wherein Learning area extracting means for extracting an area consisting of pixels of the above, generating data for solving a normal equation based on the teacher signal, the output of the class information generating means, and the output of the area extracting means A noise reduction device that is determined by a normal equation calculation unit that performs a predetermined calculation process based on an output of the normal equation calculation unit to calculate the prediction coefficient. 16. The learning method according to claim 14, wherein the prediction coefficient is a first area extracting means for learning for extracting an area composed of a plurality of pixels from an input image signal, and pixel data from the first area extracting means. Learning feature extraction means for extracting a feature based on the class information; and learning class information generation means for generating class information based on the feature and the additional information for decoding associated with the student signal. A second area extracting means for learning for extracting an area composed of a plurality of pixels from the student signal; a teacher signal; an output of the class information generating means; and an output of the second area extracting means. Calculating a prediction coefficient by performing a predetermined calculation process based on an output of the normal equation calculation means; Noise reduction device characterized in that it is determined by the prediction coefficient determining means for. 17. A noise reduction method for reducing noise of an input digital information signal generated by decoding an encoded digital information signal, wherein class information is determined based on additional information for decoding processing. Generating a class information, storing a predetermined prediction coefficient, and outputting a prediction coefficient corresponding to the class information from the stored prediction coefficients; and A region extracting step of extracting a region composed of the samples of the above, and a step of performing a prediction operation to generate a sample value based on the plurality of samples extracted in the region extracting step and the prediction coefficient. A noise reduction method characterized by the following. 18. A noise reduction method for reducing noise of an input digital information signal generated by decoding an encoded digital information signal, comprising the steps of: A first region extraction step to be extracted; a feature amount extraction step of extracting a feature amount based on the sample extracted in the first region extraction step; the feature amount and additional information for decoding processing A class information generating step of generating class information based on: a step of storing a predetermined prediction coefficient, and outputting a prediction coefficient corresponding to the class information from the stored prediction coefficients; and A second area extracting step of extracting an area consisting of a plurality of samples from the digital information signal; Performing a prediction operation on the basis of the plurality of samples extracted in the region extraction step and the prediction coefficient to generate a sample value. 19. A noise reduction method for reducing noise of an input image signal generated by decoding an encoded digital image signal, wherein class information is determined based on additional information for decoding processing. A step of generating class information to be generated; a step of storing a predetermined prediction coefficient; and a step of outputting a prediction coefficient corresponding to the class information from the stored prediction coefficients; and a step of outputting a plurality of pixels from the input image signal. And a step of generating a pixel value by performing a prediction operation based on the plurality of pixels extracted in the region extraction step and the prediction coefficient. Noise reduction method. 20. A noise reduction method for reducing noise of an input image signal generated by decoding an encoded digital image signal, wherein a pixel region including a plurality of pixels is extracted from the input image signal. A first area extraction step, a feature amount extraction step of extracting a feature amount based on the pixel data extracted in the first area extraction step, a feature amount and additional information for decoding processing. A class information generating step of generating class information based on: a step of storing a predetermined prediction coefficient, and outputting a prediction coefficient corresponding to the class information from the stored prediction coefficients; A second area extraction step of extracting a pixel area composed of a plurality of pixels from the image signal; and a second area extraction step. Performing a prediction operation on the basis of the obtained plurality of pixels and the prediction coefficient to generate a pixel value. 21. The noise reduction method according to claim 19, wherein the additional information is information indicating a type of an image signal to be processed. 22. The noise reduction method according to claim 19, wherein the additional information is format information of an image signal to be processed. 23. The noise reduction method according to claim 19, wherein the additional information is image quality information. 24. The noise reduction method according to claim 19, wherein the additional information is motion vector information.