JP5232180B2 - Adaptive sampling apparatus and adaptive sampling program - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To provide a technology for performing sampling at unequal intervals in accordance with the nature of an input signal when sampling frames of the input signal. <P>SOLUTION: This adaptive sampling device 1A includes: a frame storage means 10; a hierarchical signal reduction means 20 to perform sampling at a plurality of different sampling intervals reduced in sampling points to frames to form reduced signals to output the reduced signal group; a hierarchical error calculation means 30 to restore the respective reduced signals to sampling intervals equal to those of the input signal, and quantize errors from the frames on a block basis to be output as error signals; a sampling pattern database 50 to store a plurality of sampling patterns being an allocation method of the sampling intervals applied to the respective blocks of the frames; an optimization means 70 to search for a sampling pattern for optimizing errors based on the respective error signals and the sampling patterns; and a sample selection/arrangement means 80 to sample the frames of the input signal based on the sampling pattern obtained by the optimization means to be output as output signals. <P>COPYRIGHT: (C)2011,JPO&amp;INPIT

Description

  The present invention relates to a technique for sampling a signal, and more particularly, to a technique for sampling at irregular intervals according to the nature of an input signal.

  Sampling is a technique for discretizing an input signal temporally or spatially. Conventionally, sampling is realized by referring to an amplitude value of an input signal at a constant frequency (or time interval) in a one-dimensional signal represented by time versus amplitude such as speech.

For example, an audio CD (Compact Disc) is sampled at 44.1 kHz (about 22.7 μs).
Further, for example, in a still image, the image is divided at equal intervals in the horizontal and vertical two orthogonal axis directions, and sampling is performed at the lattice points.

  Furthermore, in a moving image, a sample point is placed in a space composed of three axes of horizontal, vertical and time. For example, by dividing into 640 pixels in the horizontal direction and 480 pixels in the vertical direction and further dividing at 30 Hz in the time direction, spatiotemporal sampling is realized.

  In the interlace method, a group of still images constituting a moving image is divided into a plurality of (for example, two) fields, and different spatial sampling is performed in each field, thereby realizing visually efficient sampling. Yes.

  For example, in a 1080 / 59.94i system television, sampling is performed at 59.94 Hz in the time direction. At this time, the image at each time is composed of 1920 pixels in the horizontal direction and 1080 pixels in the vertical direction, and an odd field that samples only pixels whose vertical coordinates are on the odd-numbered horizontal line and an even-numbered horizontal line. Alternately, even-numbered fields that sample only those pixels are placed. That is, the spatial sampling position differs between adjacent fields.

  In the MUSE (MUltiple Sub-Nyquist sampling Encoding) method (Patent Document 1), sampling of all pixels in the spatial direction is performed in four fields by quincunx sampling.

  In the quantization in the amplitude direction, nonlinear quantization is performed. For example, in the μ-law and A-law in PCM (Pulse Code Modulation) of an audio signal, a smaller quantization step is set for a portion where the level of the input signal is smaller. In this way, it is possible to improve quantization noise at the small signal level that is conspicuous in linear quantization.

  In addition, there is a method of performing sampling at a finer sampling interval for a locally important region in a signal. For example, when sampling an image, there is a technique of sampling in a polar coordinate system having an origin at the center of interest in the field of view (Patent Document 2). There is also a technique of performing quadratic hierarchical sampling by dividing an image into blocks and repeating operations for dividing important blocks into sub-blocks (Patent Document 3).

JP 60-86994 A (Patent No. 1544848) Japanese Patent No. 3007764 Japanese Patent No. 2955761

  However, in sampling at equal intervals, sampling is always performed at regular intervals regardless of temporal (or spatial) density of the input signal. For this reason, aliasing distortion occurs especially in dense signal sections in time (or space), and high-frequency components are lost due to the effect of the low-pass filter that suppresses aliasing distortion. There is a problem that it is difficult to reproduce. Conversely, when the original signal has only low frequency components, sampling is performed at unnecessarily fine intervals, and there is a problem that waste is increased in terms of information amount and bandwidth.

  The interlace method is efficient sampling using visual characteristics, but does not sample according to the texture and movement of the image.

  In the MUSE system, adaptive interpolation processing is performed on the receiver side according to the magnitude of image movement. However, this is an improvement in image quality by devising an interpolation process and does not adaptively control the Nyquist frequency by sampling. For this reason, there is a problem that a signal having a fixed Nyquist frequency regardless of texture or motion cannot be faithfully reproduced.

  The μ-law and A-law in PCM perform nonlinear quantization in the amplitude axis direction of a signal, and do not make sampling intervals in the time (space) axis direction unequal. For this reason, although there is a problem that quantization noise can be improved, aliasing distortion cannot be improved.

  Although the method described in Patent Document 2 can finely sample a locally important region, there are a plurality of important regions in a signal, or a continuous distribution from an important region to an unimportant region. However, not all important areas can be sampled in detail, and there is a problem that signal degradation occurs.

  Even in the method described in Patent Document 3, although a locally important region can be finely sampled, it is a sampling of a hierarchical structure, and a sampling output that preserves the order of the original signals is not made. There is a problem that the reproduction (reconstruction) of the image becomes complicated.

  The present invention has been made to solve such a problem, and is applied to each block of a frame when a frame, which is a fixed section of an input signal, is divided into blocks and sampled at different intervals for each block. An object of the present invention is to provide a technique for searching for an allocation method for minimizing an error from allocation methods of sampling intervals, and sampling and outputting a frame of an input signal based on the searched allocation method.

  The present invention has been devised to solve the above problems. First, the adaptive sampling device according to claim 1 divides a frame, which is a constant section of an input signal, into a plurality of blocks, and An adaptive sampling device that performs sampling at different sampling intervals or the same sampling interval, hierarchized signal reduction means, hierarchization error calculation means, sampling pattern database, optimization means, and sample selection arrangement means And a configuration comprising:

  According to such a configuration, the adaptive sampling apparatus samples the frame at a plurality of different sampling intervals with a reduced number of sampling points, and outputs the reduced signal group by the hierarchical signal reduction means. .

  Also, the adaptive sampling device returns the respective reduced signals to the same sampling interval as that of the input signal by the hierarchization error calculation means to obtain an enlarged signal, and quantifies the error between each enlarged signal and the frame for each block. Output as a signal. As a result, the adaptive sampling device can quantify signal degradation that occurs when sampling is performed at various sampling intervals.

Further, the adaptive sampling device, in the sampling pattern database, when the frame is divided into a plurality of blocks, the entire frame of the sampling pattern that is a method of assigning the sampling interval to be applied to each block of the frame Only a plurality of sampling patterns that satisfy all the constraint conditions consisting of the enlargement ratio, the number of blocks in the frame, and the enlargement ratio options that can be assigned to each block are stored.

  Further, the adaptive sampling device searches for a sampling pattern for optimizing the error based on each error signal output from the hierarchization error calculation unit and the sampling pattern by the optimization unit. As a result, the adaptive sampling apparatus can select a sample point assignment method in consideration of signal reproducibility from among sample point assignment methods stored in the sampling pattern database.

  The adaptive sampling apparatus samples the frame of the input signal by the sample selection / arrangement unit based on the sampling pattern obtained by the optimization unit, and outputs the sampled frame as an output signal. Thereby, the adaptive sampling apparatus can sample the frames at unequal intervals and output the sampling results by assigning the sampling points in consideration of signal reproducibility.

  The adaptive sampling apparatus according to claim 2, wherein a frame, which is a constant section of an input signal, is divided into a plurality of blocks, and sampling is performed at different sampling intervals or for each block. In this configuration, a hierarchized signal reduction unit, a hierarchization error calculation unit, a sampling pattern database, an ordering unit, an allocation unit, an optimization unit, and a sample selection and arrangement unit are provided.

  According to such a configuration, the adaptive sampling apparatus samples the frame at a plurality of different sampling intervals with a reduced number of sampling points, and outputs the reduced signal group by the hierarchical signal reduction means. .

  Also, the adaptive sampling device returns the respective reduced signals to the same sampling interval as that of the input signal by the hierarchization error calculation means to obtain an enlarged signal, and quantifies the error between each enlarged signal and the frame for each block. Output as a signal.

  Further, the adaptive sampling apparatus stores a record in which the sampling interval of the input signal and the number of blocks to which a plurality of different sampling intervals are assigned for each mode satisfying a predetermined constraint condition in the sampling pattern database. This constraint condition is the ratio of the number of sample points to be assigned to the reference sample points in the frame (enlargement rate as a whole frame), the number of blocks in the frame, and the sampling interval (enlargement rate) that can be assigned for each block. It means three conditions.

  Further, the adaptive sampling apparatus gives priorities to the blocks in order of increasing error based on the error signal output from the hierarchization error calculation means by the ordering means. Further, the adaptive sampling apparatus assigns a sampling interval having a higher number of sampling points to each block of the frame, based on the record for each mode and the priority order, by the assigning unit, for each mode. Generate a sampling pattern of

  Further, the adaptive sampling device searches for a sampling pattern for optimizing the error based on each error signal output from the hierarchization error calculation unit and the sampling pattern by the optimization unit. Further, the adaptive sampling apparatus samples the frame of the input signal by the sample selection / placement means based on the sampling pattern obtained by the optimization means, and outputs it as an output signal.

Further, the adaptive sampling device according to claim 3 is the adaptive sampling device according to claim 1 or 2, wherein the optimization means calculates the total of the entire frame from each error signal for each sampling pattern. An error is obtained, and a sampling pattern that minimizes the total error is searched.
According to such a configuration, the adaptive sampling apparatus can sample the frames at unequal intervals and output the sampling results by assigning sample points with good signal reproducibility.

  Further, the adaptive sampling device according to claim 4 is the adaptive sampling device according to claim 1 or 2, wherein the optimization means performs an error in the entire frame from each error signal for each sampling pattern. And adding a penalty corresponding to the combination of sampling intervals between blocks included in the sampling pattern to obtain a total error, and searching for a sampling pattern that minimizes the total error. To do.

  According to such a configuration, the adaptive sampling apparatus performs sampling at unequal intervals in the frame by assigning the sampling points taking into consideration the reproducibility of the signal and the difference in the sampling interval between the blocks. Sampling results can be output.

Further, the adaptive sampling device according to claim 5 is the adaptive sampling device according to claim 1 or 2, wherein the optimization means performs an error in the entire frame from each error signal for each sampling pattern. And adding a penalty corresponding to the amount of code necessary to describe the sampling pattern to obtain a total error, and searching for a sampling pattern that minimizes the total error.
According to this configuration, the adaptive sampling apparatus can perform sampling at unequal intervals in consideration of signal reproducibility and code amount, and output the sampling result.

The adaptive sampling apparatus according to claim 6 is the adaptive sampling apparatus according to claim 1 or 2, wherein the reduced signal generated by the hierarchical signal reduction means is sampled at the same sampling interval. Are included, and the optimization means optimizes the sampling interval and phase of the reduced signal for each block.
According to this configuration, the adaptive sampling apparatus can optimize the sampling point arrangement in both the sampling interval and the sampling phase, thereby improving the signal reproducibility.

  An adaptive sampling program according to claim 7 is a computer program for dividing a frame, which is a constant section of an input signal, into a plurality of blocks and performing sampling at different sampling intervals or for each block. Are configured to function as hierarchized signal reduction means, hierarchization error calculation means, optimization means, and sample selection arrangement means.

  An adaptive sampling program according to claim 8 is a computer program for dividing a frame, which is a predetermined section of an input signal, into a plurality of blocks and performing sampling at different sampling intervals or for each block. Are configured to function as hierarchized signal reduction means, hierarchization error calculation means, ordering means, allocation means, optimization means, and sample selection / placement means.

According to the first and seventh aspects of the present invention, the frame is sampled while changing the sampling interval for each block based on the optimal sampling pattern for optimizing the error in the frame, and the sampling result is obtained. Can be output.
For this reason, the amount of information is not wasted, and the original signal can be faithfully reproduced with a small amount of information when a sampled signal is reproduced.

Further, since local sampling is performed at regular intervals with respect to the original signal, nonlinear distortion is difficult to be added.
Furthermore, in order to store the waveform of the original signal in a similar manner locally, for example, when this method is applied to an image signal, in general, although the original image can be partially visually recognized, Can not be used as a scramble because it does not hold the original image.

  According to the second and eighth aspects of the present invention, since it is not necessary to obtain the sum of errors in a frame for all sampling patterns that satisfy the constraint conditions, an approximately optimal sampling pattern can be generated at high speed. Therefore, it is possible to speed up the sampling with few errors with respect to the frame.

  According to the third aspect of the present invention, it is possible to sample the frames at unequal intervals and output the sampling results by assigning the sampling points with good signal reproducibility.

  According to the fourth aspect of the present invention, sampling is performed at unequal intervals between frames by assigning sampling points in consideration of signal reproducibility and sampling interval differences between blocks. Can be output.

  According to the invention described in claim 5, it is possible to perform sampling at unequal intervals in consideration of the reproducibility of the signal and the code amount of the sampling pattern, and output the sampling result.

  According to the sixth aspect of the present invention, since the sampling point arrangement can be optimized in both the sampling interval and the sampling phase, the signal reproducibility can be improved.

It is a figure which shows the outline of the adaptive sampling system performed with the adaptive sampling apparatus which concerns on this invention. It is a figure which shows the example of the combination of the number of blocks which should apply each magnification in a flame | frame. It is the figure which showed the example of the combination of the expansion magnification at the time of reduction under the restraint conditions shown in FIG. It is a block diagram which shows the structure of the adaptive sampling apparatus which concerns on 1st Embodiment (and 3rd Embodiment) of this invention. (A) In the hierarchized signal reduction means, it is a block diagram showing a parallel configuration example of the signal reduction means. (B) In the hierarchical signal reduction means, it is a block diagram showing an example of a serial configuration of the signal reduction means. It is a block diagram which shows the structural example of a hierarchization error calculating means. It is a figure which shows the example of the sampling pattern memorize | stored in the sampling pattern database. It is a flowchart which shows operation | movement of the adaptive sampling apparatus which concerns on 1st Embodiment of this invention. It is a figure which shows that the number of sampling patterns will become enormous if the number of blocks in a frame increases. It is a block diagram which shows the structure of the adaptive sampling apparatus which concerns on 2nd Embodiment of this invention. It is a figure which shows the outline of the operation example of the ordering means and allocation means which concern on 2nd Embodiment of this invention. It is a figure which shows the outline of the operation example of the optimization means which concerns on 2nd Embodiment of this invention. It is a flowchart which shows operation | movement of the adaptive sampling apparatus which concerns on 2nd Embodiment of this invention. It is a figure which shows the example of a structure of the hierarchized signal reduction means which concerns on 3rd Embodiment of this invention, and the example of the position of the sample point in the reduction signal output from each signal reduction means. It is a figure which shows the example of the sampling pattern memorize | stored in the sampling pattern database at the time of discriminating a sampling interval and a phase. It is a flowchart which shows operation | movement of the adaptive sampling apparatus which concerns on 3rd Embodiment of this invention.

[Outline of adaptive sampling method]
Hereinafter, embodiments of the present invention will be described with reference to the drawings as appropriate.
In the embodiment, means having the same function are denoted by the same reference numerals and description thereof is omitted.
First, with reference to FIG. 1, the outline of the adaptive sampling system performed with the adaptive sampling apparatus of this invention is demonstrated.

  The adaptive sampling apparatus sets a frame for a certain time interval of the input signal 101 (or an interval in space, hereinafter described in terms of time). The adaptive sampling device takes M sample points (reference sample point 102) as a reference on the time axis in the frame (M is a natural number of 2 or more).

  Although the interval between the reference sample points 102 taken by the adaptive sampling apparatus is arbitrary, for example, it is set to an equal time interval. When the input signal 101 is sampled in the time direction, the reference sample point 102 is the sample point included in the frame.

  Subsequently, the adaptive sampling device divides this frame into N blocks (N is a natural number of 2 or more). In FIG. 1, N = 6. Then, the adaptive sampling device sets a sampling interval for each block. The sampling interval is selected from a plurality of types of sampling grids.

  For example, a plurality of sampling intervals having a relation of a geometric sequence, such as the interval τ of the reference sample points 102 (the interval between the white circles in FIG. 1), the double interval 2τ, and the quadruple interval 4τ are prepared as a sampling grid. Then, the sampling interval to be applied to each block is determined.

  Alternatively, a plurality of sampling intervals having a relation of an arithmetic sequence such as the interval τ of the reference sample point 102, the interval 2τ that is twice the interval τ, and the interval 3τ that is 3 times that are prepared as sampling grids. The sampling interval to apply may be determined.

  Then, the adaptive sampling device takes sample points at the sampling interval assigned to each block and outputs the amplitude value at this sample point. An example of this sample point is indicated by a black circle as the output sample point 104 in FIG. 1 is a ratio of the number of assigned sample points to the reference number of sample points for each block.

  At this time, the adaptive sampling apparatus sets a condition for the total number of sample points assigned to each block so that the total number of sample points in the frame becomes a predetermined number. For example, the total number of sample points to be allocated is set to be a constant multiple (for example, 1/2 times) of the total number of reference sample points. Further, it is preferable to sample more finely as the waveform in each block is more complicated.

  If the amplitude values at the sample points (104 black circles in FIG. 1) thus obtained are arranged so that the sample point intervals are constant (black circles 105 in FIG. 1), the waveform in the frame of the input signal 101 is obtained. Is converted into a signal as shown by the curve 106 in FIG. In FIG. 1, the signal is illustrated by a curve for simplicity, but a strict waveform is a waveform having a value only at each position of the output sample point 104.

A method for assigning the number of sample points to each block will be described.
The frame is divided into N blocks (N is a natural number), and the total number of sample points assigned in the frame is C times the total number of reference sample points (C is a positive real number). This magnification C is referred to as the overall magnification.
The number of sample points assigned to each block is selected from W types (W is a natural number) of c ₁ times, c ₂ times,..., C _W−1 times the reference sample number in the block. These options are called block magnifications. Note that c _w (w = 0, 1,..., W−1) is a positive real number.
Three conditions of the overall magnification C in the frame, the number N of blocks in the frame, and all options of block magnification {c _w } _{w = 0, 1,..., W−1} are referred to as constraint conditions.

Here, a block magnification is assigned to each block. Among all N blocks, the number of blocks a _w to which the block magnification c _w is applied is set. The number of patterns (a _w ) satisfying the constraint conditions _{w = 0, 1,..., W−1} is finite. This number of patterns is called the number of modes, and each pattern is called a mode. A number is assigned to each mode, and these are called mode numbers.
Each mode is each integer solution (a _w ) of equation (1) _{w = 0, 1,..., W−1} . The number of integer solutions is the number of modes.

Here, an example of how to assign the number of sample points to each block will be described with reference to FIG.
In the example of FIG. 2, the frame is divided into N = 6 blocks as shown in FIG. 1, and the overall magnification is C = 1/2. The block magnification is selected from W = 4 options including c ₀ = 1 (equal magnification), c ₁ = 1/2, c ₂ = ¼, and c ₃ = 1/8. Shall be chosen.

At this time, there are a total of four combinations of mode 0 to mode 3 for assigning sample points of respective magnifications to blocks. This is obtained by solving the integer solutions a ₀ , a ₁ , a ₂ , and a ₃ in the equation (2). That is, the number of modes is determined by determining the constraint conditions.

For example, in mode 0 (a ₀ , a ₁ , a ₂ , a ₃ ) = ( ₀ , ₆ , ₀ , ₀ ), a sample point that is 1/2 the reference is assigned to all blocks. That is, this corresponds to the case where the frames are sampled equally.
For example, in mode 3 (a ₀ , a ₁ , a ₂ , a ₃ ) = ( ₂ , ₁ , ₁ , ₁ , ₂ ), there are two equal-sized blocks, one half-sized block, one quarter-sized block Is 1 and there are 2 1/8 times blocks.
In any mode, the number of sample points is exactly ½ times in the entire frame.

  In each mode, when counting to which block each magnification is specifically assigned, there are as many combinations as shown in the rightmost column of FIG. When these are summed up in all modes, there are 256 ways to assign sample points.

  This way of assigning the sampling points to each block in the frame (in other words, the enlargement ratio assigned to each block in the frame) is called a sampling pattern.

  FIG. 3 shows all sampling patterns under the constraint conditions shown in FIG. In this table, ascending order with respect to the mode, and within each mode, all numbers in the denominator of the magnification for each block are considered as digits in each digit (the one with the larger block number is considered as the lower digit) and all are in ascending order. The sampling pattern is shown (omitted on the way). The sampling pattern in FIG.

  Hereinafter, the first to third embodiments of the adaptive sampling device will be described. In the first embodiment, a sampling pattern that minimizes an error is searched from all sampling patterns. On the other hand, in the second embodiment, a mode-specific sampling pattern is generated for each mode, and the speed is increased by approximately searching for a sampling pattern that minimizes the error. In the third embodiment, a sampling pattern that minimizes an error is searched from a sampling pattern that considers not only the enlargement ratio but also the sampling phase.

(First embodiment: Optimization by full search)
[Configuration of Adaptive Sampling Apparatus 1A]
With reference to FIG. 4, the configuration of the adaptive sampling apparatus 1A (1) according to the first embodiment of the present invention will be described.

  When the adaptive sampling apparatus 1A divides a frame in an input signal into blocks and samples each block, the optimal sampling apparatus 1A has the least signal degradation among all the sampling patterns stored in the sampling pattern database. A sampling pattern is obtained, and the optimum sampling pattern and an output signal obtained as a result of sampling using the optimum sampling pattern are output.

  The adaptive sampling apparatus 1A includes a frame storage unit 10, a layered signal reduction unit 20, a layered error calculation unit 30, a sampling pattern database 50, an optimization unit 70, and a sample selection / placement unit 80. .

In FIG. 4, in the number of outputs of the hierarchized signal reduction means 20, the number of inputs and outputs of the hierarchization error calculation means 30, the number of inputs of the optimization means 70, and the number of inputs of the sample selection arrangement means 80, The case where the number of reduced signals output from the reducing means 20 (hereinafter referred to as the number of resolutions) is 3 has been described as an example. I do not care.
In the following, the number of the resolution of the reduced signal when the H-1 = 3, i.e., by including the frame I ₀ which is not reduced, will be described for the case of H = 4 hierarchy.

The frame storage means 10 stores and outputs a sampled amplitude value group in a predetermined section of the input signal I as a frame I ₀ .

  Specifically, in the case where the external input signal I is an unsampled signal, the frame storage means 10 performs sampling at equal intervals with a predetermined clock, and the amplitude at each sampling point in the frame. Store the value. On the other hand, when the input signal I has already been sampled, the frame storage means 10 may store the amplitude value group at the same sampling interval as the input signal I or resample at a different sampling interval. In addition, the amplitude value group may be stored.

Then, the frame storage means 10 outputs the sampled amplitude value group as the frame I ₀ to the hierarchized signal reduction means 20, the hierarchization error calculation means 30, and the sample selection / placement means 80. An amplitude value at a sample point position t (t = 0, 1,..., T−1; T is a natural number) in the frame is denoted as I ₀ (t).
In FIG. 4, the frame I ₀ output from the frame storage unit 10 to the specimen selection / arrangement unit 80 is omitted by the symbol A in a circle.

Layered signal reduction means 20 reduces the frame I ₀ in stages, and outputs the reduced signal I _k.
Specifically, the layered signal reduction means 20 inputs a frame _{I 0} from the frame memory unit 10, a small reduction signal frame _{I 0} of sampling points _{I k (k = 1,2, ...} , H-1 H is a natural number of 2 or more), and the reduced signals I _{1 to} I _H−1 are output to the hierarchization error calculation means 30 and the sample selection arrangement means 80.
In FIG. 4, the reduced signals I _{1 to} I ₃ output from the hierarchized signal reducing means 20 to the sample selecting / arranging means 80 are omitted by the symbols B, C, and D in circles.

The number H-1 of the reduced signals is referred to as the resolution number. Further, the input frame I ₀ is set as the hierarchy 0, and the reduced signal I _k is set as the hierarchy k.
The hierarchical signal reduction means 20 includes signal reduction means 21 for the number of resolutions.

Two examples of the configuration of the hierarchical signal reduction means 20 are shown in FIG.
Layered signal reduction means in FIG. 5 (a) 20 are provided signal reduction of a plurality of different reduction ratio means 21-1 (21) to 21-3 (21) in parallel, an input frame _{I 0} in the independently It is a configuration to reduce.

  5B has a configuration in which a plurality of signal reduction units 21-1 (21) to 21-3 (21) are cascade-connected to sequentially reduce signals. At this time, all the signal reduction means 21 may have the same reduction ratio or may have different reduction ratios.

  The signal reduction means 21 applies a low-pass filter to the input signal as necessary, and then performs sub-sampling to generate and output a signal having a smaller number of sampling points than the input signal. Is. For sub-sampling, any of the following reduction methods by simple thinning, reduction after moving average, or reduction after convolution may be used.

<Reduction by simple decimation>
The signal reduction means 21 thins data in the t direction with respect to the input signal I _in (t) (t = 0, 1,..., T _in ), and outputs the signal I _out (t) (t = 0, 1,..., T _out ). Preferably, T _in > T _out .
For example, when the enlargement ratio is 1 / K times, the signal reduction means 21 performs the equation (3).

  Here, L is a parameter (phase) for determining the sampling start position, and for example, a value such as L = 0, equation (4), or equation (5) can be used.

<Reduction after moving average>
Further, the signal reduction means 21 may perform sampling while applying a moving average filter as shown in equation (6).

Here, L ₀ and L ₁ are a lower limit and an upper limit of a phase range in which moving average is performed, and L ₀ <L ₁ is set. For example, in the case of reduction to 1/3 times, parameters such as K = 3, L ₀ = 0, and L ₁ = 2 can be used.

<Reduction after convolution>
Further, as shown in the equation (7), the signal reduction unit 21 generally applies a digital filter with tap coefficients a (L ₀ ), a (L ₀ +1),..., A (L ₁ ). Then, thinning may be performed.

  In the case of performing reduction with an enlargement ratio of 1/2 using a 7-tap digital filter, for example, the parameter of equation (8) can be used.

The hierarchization error calculation means 30 quantifies the deterioration of one or more input reduced signals I _k received with respect to the frame I ₀ before reduction as an error for each block.

Specifically, the hierarchization error calculation means 30 inputs the frame I ₀ before reduction from the frame storage means 10, and the reduction signal I _k (k = 1, 2,..., H from the hierarchization signal reduction means 20. -1; H is a natural number of 2 or more). Then, the hierarchization error calculation means 30 calculates an error signal F _k (t) between the enlarged signal U _k (t) _obtained by up-sampling each reduced signal I _k and the frame I ₀ (t). Subsequently, the hierarchization error calculation means 30 aggregates F _k (t) for each block i (i = 0, 1,..., N−1) to obtain an error signal E _k (i), and this E _k ( i) is output to the optimization means 70. Here, N is the number of divisions of the frame into blocks.
The hierarchization error calculation means 30 includes signal enlargement means 31 and error calculation means 32 for the number of resolutions.

The hierarchization error calculation means 30 can be configured as shown in FIG. 6, for example.
6 includes signal expanding means 31-1 (31) to 31-3 (31) and error calculating means 32-1 (32) to 32-3 (32). .

In the example of FIG. 6, the case where there are _three reduced signals I _{1 to} I ₃ is illustrated, but the number of reduced signals may be any number as long as it is 1 or more. When there are H-1 reduced signals (H is a natural number of 2 or more), it is assumed that there are also H-1 signal expanding means 31 and error calculating means 32 in FIG.

The signal expanding means 31 upsamples the input reduced signal I _k (k = 1, 2,..., H−1; H is a natural number of 2 or more) and outputs the result as an expanded signal U _k. is there. Upsampling may be, for example, increasing sample points by interpolation processing or increasing sampling points by super-resolution processing.

  The enlargement ratio in the upsampling of the signal enlargement means 31 is the reciprocal of the enlargement ratio at the time of generation of each reduced signal output from the hierarchical signal reduction means 20. That is, the signal enlarging means 31 performs upsampling with an enlargement ratio of K times for the reduced signal that has been reduced by the enlargement ratio of 1 / K in the hierarchical signal reduction means 20.

When the signal expansion means 31 upsamples the number of sampling points of I _k (t) by K _k times and outputs the result as U _k (t), up-sampling by linear hold shown below and linear interpolation Any of the upsampling method according to, or the upsampling method by cubic interpolation may be used.

<Upsampling with zero-order hold>
When upsampling is performed by zero-order hold, the signal expansion means 31 can be realized by the equation (9) or (10).

Here, Λ _k is a real constant that determines how to round t / K _k to an integer below the decimal point (method of rounding down, rounding up, rounding off, etc.). Preferably, 0 ≦ Λ _k <1.

<Upsampling by linear interpolation>
When upsampling is performed by linear interpolation, the signal expansion means 31 can be realized by the equation (11).

<Upsampling by cubic interpolation>
When upsampling is performed by cubic interpolation, the signal enlarging means 31 can be realized by equation (12).

The error calculating means 32 quantifies the difference between signals of the enlarged signal U _k and the frame I ₀ as an error for each block.
Specifically, the error calculation means 32 includes an enlarged signal U _k (t) (t = 0, 1,..., T−1) _obtained by up-sampling the reduced signal I _k by the signal enlargement means 31, and a frame I ₀ ( t) and an error signal F _k (t) with respect to t), and then summing up errors for each block to obtain an error signal E _k (i) (i = 0, 1,..., N−1; N is a natural number) Is output. Here, t is the number of sample points of frame I ₀ and i is a block number.

  For example, when the square error is used as the error evaluation standard, the error calculation means 32 executes the calculation of equation (13).

  For example, when an absolute value error is used as an error evaluation criterion, the error calculation means 32 executes the calculation of the equation (14).

Subsequently, the error calculation means 32 adds up the errors for each block i (i = 0, 1,..., N−1; N is a natural number). When the number of sample points of the frame I ₀ is T, and this is divided into N blocks, the error calculation means 32 calculates the error signal E _k (i) for the block i by the equation (15). . Here, B _i is a set of sample points belonging to the block i.

Furthermore, in general, the error calculation means 32 may use any error evaluation standard as long as the error between the two signals U _k and I ₀ can be quantified for each block i. For example, it is conceivable that the error calculation means 32 evaluates the error between the two signals U _k and I ₀ in the block i based on the difference in fractal nature of the waveform shape. In this case, for example, the error calculation means 32 obtains the fractal dimension of the signal U _{k and} the fractal dimension of the signal I _{0 in} the block i, and calculates the absolute value of the difference between these dimensions as the error signal E _k ( i).

Returning to FIG. 4, the sampling pattern database 50 stores a plurality of sampling patterns that describe which enlargement ratio is applied to which block.
Each sampling pattern satisfies all of the constraint conditions including the enlargement ratio of the entire frame, the number of blocks, and options of the enlargement ratio that can be assigned to each block. The sampling pattern database 50 may register all of the sampling patterns that satisfy this constraint condition, or may register a part of them.

For example, the sampling pattern database 50 stores, in a tabular form, a hierarchy (enlargement rate option) κ (p; i) used for the block i when the pattern number of the sampling pattern is p.
Specifically, for example, the sampling pattern database 50 has, for example, an enlargement ratio of 1/2 as a whole frame, the number of blocks of 6, and an enlargement ratio option to be assigned to each block {same size, 1/2 times , 1/4 times, 1/8 times}, the table shown in FIG. 7 is stored. This table is obtained by replacing the enlargement ratio display in the table shown in FIG. 3 with the layer number, and the same magnification (1/1 times) layer is κ = 0, the 1/2 layer is κ = 1, A quarter-fold hierarchy is represented by κ = 2, and a ８-fold hierarchy is represented by κ = 3.

The optimization means 70 uses the error signal E _k (i) to obtain a sampling pattern P that optimizes the error.
Specifically, the optimization unit 70 sends an error signal E _k (i) (k = 1, 2,..., H−1; H is a natural number equal to or greater than 2) from the hierarchy error calculation unit 30 ( i = 0, 1,..., N−1) are input, and the sampling pattern p stored in the sampling pattern database 50 is sequentially read out. Then, the optimization unit 70 obtains the sum of errors in the frame using the error signal E _k (i) for each sampling pattern p. The optimization unit 70 obtains a sampling pattern for optimizing (minimizing) the sum of the errors, and outputs the sampling pattern as the optimal sampling pattern P to the sample selection / placement unit 80 and the outside.

<The objective function is error only>
For example, the optimization unit 70 obtains the sampling pattern so as to minimize the error. Specifically, the calculation of equation (16) is performed.

Here, G (p) is the sum of errors in the entire frame when the sampling pattern of pattern number p is used, and has a meaning as an objective function for optimization. Further, since no error occurs in the case of hierarchy 0, that is, equal magnification, E ₀ (i) is defined to always have a value of 0 regardless of the block number i.

<Penal according to the relationship between error and layer number between adjacent blocks>
Further, the optimization unit 70 may operate so as to obtain the optimum sampling pattern P based on the error and the relationship between the sampling intervals between the blocks. For example, the objective function G (p) is defined on the basis of the error and the relationship between the hierarchical numbers between adjacent blocks, and the calculation of equation (17) is performed.

In (17), the function F _B is a function that gives a penalty in accordance with the hierarchical number between adjacent blocks. Preferably, the function F _B (κ ₀ , κ ₁ ) gives a larger penalty as the difference in the layer number (κ ₀ and κ ₁ ) between adjacent blocks increases. In this way, a rapid change in the sampling interval between adjacent blocks is suppressed, and the characteristics of the signal characteristics after sampling (for example, the change in amplitude value between adjacent sampling points, that is, the second floor) Difference value) Change is reduced.

This can prevent, for example, large prediction errors and conversion coefficients from occurring when prediction and conversion (for example, discrete cosine conversion, etc.) are performed on the output signal from this apparatus. The efficiency of encoding used can be improved.
For example, the expression (18) can be used as the penalty function F _B (κ ₀ , κ ₁ ).

  Γ in the equation (18) is a coefficient for adjusting the importance of this penalty with respect to the error, and is a positive real value.

<Error and code amount penalty according to sampling pattern>
Further, the optimization unit 70 may perform optimization based on the error and “the amount of code required to describe the sampling pattern”. This is a suitable method particularly when a variable-length code is used for describing the sampling pattern. In this case, optimization may be performed including a penalty F _P (p) corresponding to the sampling pattern p as shown in the equation (19).

The penalty F _P (p) is preferably a function that returns a larger value (or the same value) as the code amount allocated to the sampling pattern p increases.

<Penal according to the relationship between error and layer number between adjacent blocks, and penalty of code amount according to sampling pattern>
Further, the optimization unit 70, as shown in the equation (20), has a penalty F _B corresponding to the relationship of the layer number between adjacent blocks, and a penalty _FP (p) of the code amount corresponding to the sampling pattern p. It may be optimized in consideration of both.

Sample selection arrangement means 80, according to the optimum sampling pattern P, by connecting the frame I ₀ is a reduced signal I _k and the input signal while changing the hierarchy (magnification) for each block, which generates an output signal D It is.

Specifically, the sample selection / arrangement unit 80 receives the optimal sampling pattern P from the optimization unit 70, and also receives the frame I ₀ from the frame storage unit 10, and the reduced signal I _k (k from the hierarchical signal reduction unit 20. = 1, 2,..., H-1; H is a natural number of 2 or more). Then, according to the optimum sampling pattern P, the sample selection / placement means 80 changes the reduced signal I _k and the input signal frame I ₀ while changing the hierarchy (enlargement ratio) for each block as shown in the following equation (21). Concatenated to generate an output signal D, and outputs the output signal D to the outside.

Here, T _i ^(k) is the number of sample points included in the i-th block on the scale of the signal I _k (enlargement ratio K _k when the frame I ₀ is reduced). B _i is expressed by equation (15) (lower side), and | B _i | represents the number of reference sample points included in the block B _i in the frame I ₀ .

R _n ^(k) is the minimum value of t of sample points belonging to the _nth block (block B _n ) among the sample points on the signal I _k . Furthermore, S _n is the minimum value of t of sample points belonging to the _nth block (block B _n ) among the sample points on the output signal D. Here, S _N is the total number of sample points in the output signal D.

According to the adaptive sampling apparatus 1A, the frame storage means 10 sets the group of amplitude values sampled in a certain section of the input signal I as the frame I ₀ , and the hierarchical signal reduction means 20 converts the frame I ₀ stepwise. The reduced signal I _k is reduced to the reduced signal I _k , and the error signal with respect to the frame I ₀ is calculated for each reduced signal I _k by the hierarchized error calculation means 30, and this is aggregated for each block i to obtain the error signal E _k (I), an optimum sampling pattern P for optimizing the error in the frame is obtained from the sampling patterns read from the sampling pattern database 50 by the optimization means 70, and the optimum sample is obtained by the sample selection / placement means 80. In accordance with the conversion pattern P, the reduced signal I _k and the frame I ₀ are concatenated while changing the hierarchy for each block to generate the output signal D.

  For this reason, the adaptive sampling apparatus 1A samples the frame while changing the sampling interval for each block based on the optimal sampling pattern for optimizing the error in the frame, and outputs the sampling result. Can do.

  The configuration of the adaptive sampling device 1A is not limited to this, and can be changed without departing from the spirit of the present invention.

[Operation of Adaptive Sampling Apparatus 1A]
Next, the operation of the adaptive sampling apparatus 1A (1) according to the first embodiment of the present invention will be described with reference to FIG.

First, the adaptive sampling apparatus 1A uses the frame storage means 10 to convert the sampled amplitude values in a certain section of the input signal I into a frame I ₀ (t) (t = 0, 1,..., T−1; T Is stored as a natural number) (step S11). Then, the adaptive sampling apparatus 1A reduces the frame I ₀ stepwise by the hierarchized signal reduction means 20, and the reduced signal I _k (k = 1, 2,..., H−1; H is a natural number of 2 or more. ) Is output (step S12).

Then, the adaptive sampling device 1A calculates the error signal F _k (t) between the up-sampled enlarged signal U _k (t) and the frame I ₀ with respect to each reduced signal I _k by the hierarchization error calculation means 30. Then, the error signal E _k (i) is generated for each block i (i = 0, 1,..., N−1) (step S13).

Then, the adaptive sampling device 1A optimizes the sum of errors in the frame from the sampling patterns read from the sampling pattern database 50 by using the error signal E _k (i) by the optimization unit 70. An optimum sampling pattern P to be obtained is obtained (step S14).

Then, the adaptive sampling device 1A generates the output signal D by concatenating the reduced signal I _k and the frame I ₀ while changing the hierarchy (enlargement ratio) for each block according to the optimal sampling pattern by the sample selection / placement means 80. (Step S15).

(Second embodiment: approximate high-speed optimization)
As the number N of blocks into which a frame is divided increases, the number of sampling patterns increases rapidly. For example, a sample in which the block is N = 27, the magnification at the time of reduction is equal, 1/2, 1/4, or 1/8, or the magnification of the entire frame is 1/4. The conversion pattern is as shown in FIG. Among them, for example, there are 393693300 patterns just for the number of sampling patterns belonging to mode 3, and an enormous amount of calculation is required for optimization by full search.
Therefore, in the second embodiment, an adaptive sampling device that achieves high speed by approximating optimization will be described.

[Configuration of Adaptive Sampling Apparatus 1B]
With reference to FIG. 10, the configuration of the adaptive sampling apparatus 1B (1) according to the second embodiment of the present invention will be described.
When the adaptive sampling apparatus 1B divides the frame in the input signal into blocks and samples each block, the adaptive sampling apparatus 1B describes the number of blocks to which each enlargement ratio is applied when sampling stored in the sampling pattern database. From the recorded record and error signal, a mode-specific sampling pattern with a hierarchical number assigned to each block in each mode is generated, and the optimal sampling pattern with the least signal degradation is generated from this mode-specific sampling pattern. The optimum sampling pattern and an output signal obtained as a result of sampling using the optimum sampling pattern are output.

  The adaptive sampling apparatus 1B includes a frame storage unit 10, a hierarchical signal reduction unit 20, a hierarchical error calculation unit 30B, an ordering unit 40, a sampling pattern database 50B, an allocation unit 60, and an optimization unit. 70B and specimen selection / arrangement means 80B.

  The frame storage unit 10 and the hierarchized signal reduction unit 20 are the same as the units included in the adaptive sampling apparatus 1A according to the first embodiment, and thus description thereof is omitted.

The hierarchical error calculating means 30B outputs an error signal E _k (i) (k = 1, 2,..., H−1; H is a natural number of 2 or more) (i = 0, 1,. 1; N is the number of blocks) is the same as the hierarchization error calculation means 30 according to the first embodiment, except that the ordering means 40 and the optimization means 70B are output. Therefore, the description is omitted.

The ordering means 40 generates priority information by assigning priorities to the blocks in the frame based on the error signal.
Specifically, the ordering means 40 receives the error signal E _k (i) (k = 1, 2,..., H−1; H is a natural number of 2 or more) (i = 0) from the hierarchical error calculation means 30B. ,..., N-1). The ordering means 40 generates order information by assigning priorities to the blocks in the frame based on the error signal E _k (i) of one or more hierarchies k, and assigns the order information. 60.

It is preferable that the ordering unit 40 increases the priority of a block in which information loss due to a signal reduction operation is large.
For example, the ordering means 40 pays attention to the error signal E _k (i) of the hierarchy k and arranges the blocks in descending order of the error value (error). The jth block number (j = 0, 1,..., N−1) from the largest error value is set as β (j), and this is called rank information.

A specific operation of the ordering means 40 will be described with reference to FIG. In this example, the ordering means 40 performs ordering by paying attention to the error value E ₁ (i) (201 in FIG. 11) when k = 1, that is, 1/2 times. When the ordering means 40 assigns priorities in descending order of error values, the order becomes 202 as shown in FIG.
Accordingly, the rank information β (j) is expressed by the equation (22) (204 in FIG. 11).

Returning to FIG. 10, the sampling pattern database 50B stores a plurality of records describing the number of blocks to which each enlargement ratio is applied.
The number of blocks for each enlargement rate described in each record shall satisfy all constraints including the enlargement rate for the entire frame, the number of blocks, and the options for the enlargement rate that can be assigned to each block. . The sampling pattern database 50B may register all combinations of the number of blocks that satisfy this constraint condition, or may register some of them.

  The sampling pattern database 50B stores, for example, the number of blocks ν (m; κ) of the layer number (identifier for distinguishing the application magnification ratio) κ in the table format when the mode number is m.

Specifically, for example, the enlargement ratio of the entire frame is 1/4, the number of blocks is 27, and the enlargement ratio options that can be assigned to each block are {same size, 1/2 times, 1/4 times, 1 / If it is 8 times}, the table shown in the thick frame in FIG. 9 is stored.
Also, for example, the enlargement ratio of the entire frame is 1/2, the number of blocks is 6, and the enlargement ratio options that can be assigned to each block are {same size, 1/2 times, 1/4 times, 1/8 times} If it is, the table shown in the thick frame in FIG. 2 is stored.
Hereinafter, for the sake of simplicity, description will be made based on a specific example related to FIG.

The assigning means 60 assigns a hierarchical number to each block of each mode number.
Specifically, the assigning means 60 inputs the rank information β (j) (j = 0, 1,..., N−1) from the ordering means 40, and the mode number from the sampling pattern database 50B. The number of blocks ν (m; κ) to which the layer number κ for m is assigned is read. Then, the assigning means 60 uses the equation (23) to increase the enlargement magnification (preferably greater than 0 and less than or equal to 1) as the priority is higher for each block of each mode number m (to 1). Assign the near (hierarchical) number χ (m; i). Then, the assigning means 60 outputs this hierarchy number χ (m; i) to the optimizing means 70B.
This χ (m; i) is referred to as a mode-specific sampling pattern.

Note that η (m; j) in the equation (23) is a function that outputs a hierarchical number to be assigned to the block with the rank j when the mode number is m. This considers a sequence of N elements (η (m; j)) _{j = 0, 1,..., N−1,} and sets the value κ of each hierarchical number to ν (m; κ) can be generated by arranging them in ascending order with respect to κ.

  η (m; j) will be described by taking the mode number m = 3 in the case of FIG. 2 as an example. In the case of mode number m = 3 in FIG. 2, hierarchical number κ = 0 is assigned to 2 blocks, hierarchical number κ = 1 is assigned to 1 block, hierarchical number κ = 2 is assigned to 1 block, and hierarchical number κ = 3 is assigned to 3 blocks. It is done. That is, the equation (24) is obtained.

The assigning means 60 generates a numerical sequence (η (3; j)) _{j = 0, 1,..., 5} by the following procedure.
First, since there are 2 blocks in the hierarchy 0, (η (3; j)) _{j = 0, 1,.} Since layer 1 has one block, (η (3; j)) _{j = 0, 1,..., 5}
There is one term of value 1. Since layer 2 has one block, (η (3; j)) _{j = 0, 1,...} Since there are two blocks in the hierarchy 3, (η (3; j)) _{j = 0, 1,...} When the numerical sequence (η (3; j)) _{j = 0, 1,..., 5} is configured so that the values of the terms are in ascending order, equation (25) is obtained.

  For example, in FIG. 11 (203), when the mode number m = 0, all six blocks are halved, so the assigning means 60 assigns “½ times” (hierarchical number 1) to all blocks. Assign (equation (26)).

  Next, when the mode number m = 1, the same size (hierarchy number 0) is 1 block, 1/2 times (hierarchy number 1) is 3 blocks, 1/4 times (hierarchy number 2) is 2 blocks, 1 / 8 times (hierarchical number 3) is 0 block. Therefore, the assigning means 60 obtains equation (27).

  Similarly, when the mode number m = 2, the assigning means 60 obtains equation (28).

  Further, when the mode number m = 3, the assigning means 60 obtains the equation (29).

  Then, the assigning means 60 assigns a hierarchical number to each block according to the priority order of each block based on the equation (23), and sets the mode-specific sampling pattern χ (m; i) as shown in the equation (30). To get to. Note that reference numeral 203 in FIG. 11 represents the expression (30) at a specific magnification.

The optimization means 70B uses the error signal to obtain a mode-specific sampling pattern that minimizes the error.
Specifically, the optimization unit 70B receives the error signal E _k (i) (k = 1, 2,..., H−1; H is a natural number greater than or equal to 2) (i = 0) from the hierarchical error calculation unit 30B. ,..., N−1) and a mode-specific sampling pattern χ (m; i) (m is a mode number). Then, the optimization unit 70B uses the error signal E _k (i) to obtain the sum of errors in the frame by the mode-specific sampling pattern χ (m; i) for each mode m. Then, the optimization means 70B obtains a sampling pattern string (mode-specific sampling pattern) that optimizes (minimizes) the sum of the errors, and obtains an optimal sampling pattern string (Q _i ) _{i = 0, 1, ..., N-1} is output to the specimen selection / arrangement means 80B and the outside.
The optimization unit 70B performs the calculation of equation (31).

Here, Γ (m) is the sum of errors in the entire frame when the mode-specific sampling pattern χ (m; i) is used. Further, since no error occurs in the case of the hierarchy 0, that is, equal magnification, E ₀ (i) is always defined as 0 regardless of the block number i.

  As shown in FIG. 12, the optimization means 70B calculates the error for each block for each mode number m, and calculates the sum of errors in the frame. Then, the optimization unit 70B searches for a mode that minimizes the sum of errors in the frame (mode number 3 in the example of FIG. 12), and uses the sampling pattern sequence in this mode as the optimal sampling pattern sequence. Output as.

  Note that the optimization unit 70B uses the hierarchical number relationship between adjacent blocks as in the case of the equations (17), (19), and (20) of the optimization unit 70 according to the first embodiment. It is also possible to consider a penalty corresponding to the code amount and a code amount penalty corresponding to the sampling pattern sequence.

For example, the optimization unit 70B can perform the optimization according to the equation (32) when considering the penalty F _B according to the relationship of the layer number between adjacent blocks. This corresponds to equation (17).

  Further, for example, the optimization unit 70B can perform the optimization according to the expression (33) when considering the penalty of the code amount according to the sampling pattern sequence χ (m; i). This corresponds to the equation (19).

Here, F _Q (κ ₀ , κ ₁ ,..., Κ _N-1 ) is set according to the code amount assigned to the sampling pattern sequence (κ ₀ , κ ₁ ,..., Κ _N-1 ). Preferably, the penalty is set to be larger (or equal) as the assigned code amount is larger.

Further, the optimization unit 70B may consider both the penalty F _B corresponding to the relationship of the layer numbers between adjacent blocks and the code amount penalty F _Q corresponding to the sampling pattern sequence. In this case, the optimization unit 70B can perform the optimization using the equation (34). This corresponds to the equation (20).

The sample selection / arrangement means 80B is a frame that is the reduced signal _Ik and the current signal while changing the hierarchy (enlargement ratio) for each block in accordance with the optimum sampling pattern sequence (Q _i ) _{i = 0, 1,.} by connecting the I _0, and generates an output signal D.

Specifically, the sample selection / arrangement unit 80B receives the optimal sampling pattern sequence (Q _i ) _{i = 0, 1,..., N−1} from the optimization unit 70B, and the frame I ₀ from the frame storage unit 10. , The reduced signal I _k (k = 1, 2,..., H−1; H is a natural number of 2 or more) is input from the hierarchical signal reducing means 20. Then, according to the optimum sampling pattern sequence (Q _i ), the sample selection / arranging means 80B is the reduced signal I _k and the current signal while changing the hierarchy (enlargement ratio) for each block as in the following equation (35). The frame I ₀ is concatenated to generate an output signal D, and this output signal D is output to the outside.

Here, the expression (35) is obtained by replacing κ (P; i) with Q _i in the expression (21) of the sample selection / arranging means 80 according to the first embodiment (i = 0, 1, ..., N), the meanings of the variables other than the replaced variable are the same.

According to the adaptive sampling apparatus 1B, the frame storage means 10 sets the amplitude value group sampled in a certain section of the input signal I as the frame I ₀ , and the hierarchical signal reduction means 20 converts the frame I ₀ stepwise. The reduced signal I _k is reduced to a reduced signal I _k , and the error signal with respect to the frame I ₀ is calculated for each reduced signal I _k by the layered error calculation means 30B. The order means 40 prioritizes the blocks in the frame based on the error signal and generates rank information. The assignment means 60 generates the rank information and ν (m; κ) read from the sampling pattern database 50B. The hierarchical number (mode-specific sampling pattern) is assigned to each block i of each mode number m, and the error signal is transmitted by the optimization means 70B. Is used to obtain an optimum sampling pattern for optimizing the sum of errors in the frame from the mode-specific sampling patterns, and the sample selection and placement means 80B changes the hierarchy for each block according to the optimum sampling pattern. The reduced signal I _k and the frame I ₀ are concatenated to generate the output signal D.

  For this reason, the adaptive sampling apparatus 1B does not have to obtain the sum of errors in the frame for all sampling patterns that satisfy the constraint conditions, and therefore can perform optimization approximately and increase the speed. Can be achieved.

  The configuration of the adaptive sampling device 1B is not limited to this, and can be changed without departing from the spirit of the present invention.

[Operation of Adaptive Sampling Apparatus 1B]
Next, the operation of the adaptive sampling device 1B (1) according to the second embodiment of the present invention will be described with reference to FIG.

First, the adaptive sampling device 1B uses the frame storage means 10 to convert the sampled amplitude values in a certain section of the input signal I into a frame I ₀ (t) (t = 0, 1,..., T−1; T Is stored as a natural number) (step S21). Then, the adaptive sampling device 1B reduces the frame I ₀ stepwise by the hierarchized signal reduction means 20, and the reduced signal I _k (k = 1, 2,..., H−1; H is a natural number of 2 or more. ) Is output (step S22).

Then, the adaptive sampling device 1B calculates the error signal F _k (t) between the up-sampled enlarged signal U _k (t) and the frame I ₀ for each reduced signal I _k by the hierarchization error calculation means 30. Then, this is totaled for each block i (i = 0, 1,..., N−1) to generate an error signal E _k (i) (step S23).

Then, the adaptive sampling device 1B uses the ordering means 40 to prioritize the block groups in the frame based on the error signal E _k (i) and generate rank information (step Ss24).

  Then, the adaptive sampling device 1B uses the allocation unit 60 to assign the rank information β (j) and the block number ν (m; κ) to which the hierarchical number κ when the mode number read from the sampling pattern database is m. Are used to assign a hierarchical number χ (m; i) to each block i of each mode number m (step S25).

Then, the adaptive sampling device 1B optimizes the sum of errors in the frame from the mode-specific sampling pattern χ (m; i) using the error signal E _k (i) by the optimization unit 70B. The optimum sampling pattern Q _i to be obtained is obtained (step S26).

Then, the adaptive sampling device 1B concatenates the reduced signal I _k and the frame I ₀ while changing the hierarchy (enlargement ratio) for each block according to the optimal sampling pattern Q _i by the sample selection / arrangement means 80B, and outputs the output signal D. Is generated (step S27).

(Third embodiment)
In the first embodiment and the second embodiment, the case where the layered signal reduction unit 20 outputs all reduced signal groups having different sampling intervals has been described. It may be configured to output the same including reduced signals having different sampling phases.
Therefore, in the third embodiment, an adaptive sampling device in the case where the hierarchical signal reduction means outputs “including reduced signals having the same sampling interval and different sampling phases” will be described.

[Configuration of Adaptive Sampling Device 1C]
The configuration of the adaptive sampling apparatus 1C (1) according to the third embodiment of the present invention will be described (refer to FIG. 4 as appropriate).
The adaptive sampling device 1C divides the frame in the input signal into blocks and samples each block, and then selects from all sampling patterns distinguished from the sampling interval and phase stored in the sampling pattern database. Then, an optimum sampling pattern with the least signal degradation is obtained, and this optimum sampling pattern and an output signal obtained as a result of sampling using the optimum sampling pattern are output.

  The adaptive sampling apparatus 1C includes a frame storage unit 10, a layered signal reduction unit 20C, a layered error calculation unit 30C, a sampling pattern database 50C, an optimization unit 70, and a sample selection / placement unit 80. .

  The frame storage unit 10, the optimization unit 70, and the sample selection / arrangement unit 80 are the same as the units included in the adaptive sampling apparatus 1A according to the first embodiment, and thus description thereof is omitted.

Layered signal reduction means 20C is out of phase with reducing the frame I ₀ in stages, different reduced signal I _{kappa phases,} and outputs the _L.
Specifically, the hierarchical signal reduction unit 20C receives the frame I ₀ from the frame storage unit 10, reduces the frame I ₀ to a reduced signal with a small number of sample points, and shifts the phase. Different reduced signals I _{κ, L} (κ = 1, 2,..., H−1; H is a natural number of 2 or more) (L is a phase corresponding to κ) and hierarchical error calculation means 30C and sample selection arrangement means 80 Output to.

Layered signal reduction means 20C (as a hierarchical 0 frames I _0, the hierarchy in order magnification is large 1,2, ... and, this is called hierarchical number) layer corresponding to the enlargement ratio for the frame I ₀ and, each layer The signal reduction means 21C is provided according to the number of phases to be shifted.
Signals output by the signal reduction means 21C when the layer number is κ and the phase is L are written as I _{κ and L.}

A configuration example of the hierarchized signal reduction means 20C is shown in FIG. In FIG. 14, the hierarchized signal reduction means 20C reduces the frame I ₀ to three hierarchies, and generates a reduced signal by shifting the phase by 2 ^kappa for each hierarchy k. For example, the sample points of the reduced signals I _{κ and L} output from the signal reduction means 21C-κL for the block having eight sample points of the input frame I ₀ (upper right in FIG. 14) are as shown on the right side. become.

  Similarly to the hierarchized signal reduction unit 20 of the first embodiment, the signal reduction unit 21C applies each of the formulas (3) based on the hierarchy number κ and the phase L when performing reduction by simple decimation. Generate a reduced signal. Note that the sampling interval K in equation (3) is given by equation (36).

Similarly to the hierarchized signal reduction unit 20 of the first embodiment, the signal reduction unit 21C performs the reduction after moving average or the reduction after convolution (6) or (7 The parameters L ₀ and L ₁ in the formula are determined according to the phase L. For example, it is determined using a constant Δ ₀ and a constant Δ ₁ as shown in equation (37). At this time, Δ ₀ <Δ ₁ is satisfied.

Layered error calculating unit 30C is reduced signals are one or more input I _{kappa, L} is intended to quantify each block degradation component which receives the frame I ₀ before reduction as an error.

Specifically, the hierarchization error calculation means 30C receives the pre-reduction frame I ₀ from the frame storage means 10, and the reduction signal I _{κ, L} (κ = 1, 2,...) From the hierarchization signal reduction means 20C. , H-1; H is a natural number of 2 or more) (L is a phase corresponding to κ). Then, the layered error calculation means 30C obtains error signals F _{κ and L} (t) between the enlarged signals U _{κ and L} (t) obtained by up-sampling the reduced signals I _{κ and L} and the frame I ₀ (t). calculate. Subsequently, the hierarchization error calculation means 30C aggregates F _{κ, L} (t) for each block i (i = 0, 1,..., N−1) to obtain error signals E _{κ, L} (i), This E _{κ, L} (i) is output to the optimization means 70. Here, N is the number of divisions of the frame into blocks.

The hierarchization error calculation means 30C includes a signal expansion means 31C and an error calculation means 32C.
The configuration of the hierarchization error calculation unit 30C is the same as the configuration of the hierarchization error calculation unit 30 according to the first embodiment (see FIG. 6).

The signal expanding means 31C upsamples the input reduced signals I _{κ, L} (κ = 1, 2,..., H−1; H is a natural number of 2 or more) (L is the number of phases corresponding to κ). The result is output as the enlarged signals U _{κ and L.} The signal expansion unit 31C may perform signal expansion in consideration of the phase, or may perform signal expansion like the signal expansion unit 31 according to the first embodiment without considering the phase (fixing the phase). May be.

  When the signal expansion means 31C performs signal expansion in consideration of the phase, for example, based on the phase L shown in each block of the signal reduction means 21C shown in FIG. 14, for example, Expression (9), Expression (10) Instead of the equations (11) and (12), each function is expressed in the t direction as in the equations (38), (39), (40), and (41), respectively. What moved by L may be used.

The error calculation means 32C quantifies the difference between the signals of the enlarged signals U _{κ and L} and the frame I ₀ as an error for each block.
Specifically, the error calculating means 32C includes the enlarged signals U _{κ, L} (t) (t = 0, 1,..., T−1) obtained by up-sampling the reduced signals I _{κ and L} by the signal expanding means 31C. An error signal F _{κ, L} (t) with _respect to the frame I ₀ (t) is generated, and then the error for each block is totaled to obtain an error signal E _{κ, L} (i) (i = 0, 1,... N-1; N is a natural number). Here, t is the number of sample points of frame I ₀ and i is a block number.

  An example of the operation of the error calculation means 32C is an example based on the expressions (13) to (15) and the fractal dimension immediately after that in the error calculation means 32 according to the first embodiment (when the sampling phase is fixed). The same as described. However, the subscript “k” in each mathematical expression is replaced with the subscript “κ, L”.

  The sampling pattern database 50C stores a plurality of sampling patterns describing which enlargement ratio and which phase are applied to which block. That is, the sampling pattern database 50C stores sampling patterns in which phases are different and sampling intervals are different from each other.

  Specifically, the sampling pattern database 50C has one phase (L = 0) for the block with hierarchical number 0 and two phases (L = 0) for the block with hierarchical number 1. 1), four phases (L = 0, 1, 2, 3) for the block of layer number 2, and eight phases (L = 0) for the block of layer number 3 , 1, 2, 3, 4, 5, 6, 7) or a combination of all of them is stored as a sampling pattern.

  For example, the enlargement ratio of the entire frame is 1/2, the number of blocks is 6, and the enlargement ratio options that can be assigned to each block are {same size, 1/2, 1/4, 1/8} (in order of layer number). In the case of {0, 1, 2, 3}), when the sampling pattern database 50C stores all combinations of hierarchy numbers and phases, the sampling pattern database 50C is as shown in FIG. Store the sampling pattern.

  In FIG. 15, for example, pattern numbers 64 to 191 correspond to pattern number 1 (0, 1, 1, 1, 2, 2) in FIG. 7 (sampling pattern that does not distinguish phases), Since there are one phase in layer 0, two phases in layer 1, and four phases in layer 2, the result is 1 × 2 × 2 × 2 × 4 × 4 = 128.

According to the adaptive sampling apparatus 1C, the frame storage means 10 sets the amplitude value group sampled in a certain section of the input signal I as the frame I ₀ , and the hierarchical signal reduction means 20C converts the frame I ₀ stepwise. And the phase is shifted to obtain reduced signals I _{κ and L} , and the layered error calculation means 30C calculates an error signal with respect to the frame I ₀ for each reduced signal I _{κ and L} , and this is calculated for each block i. The error signal is E _{κ, L} (i), and an optimum sampling pattern P for optimizing the error in the frame from the sampling patterns read from the sampling pattern database 50C by the optimization means 70 is obtained. The sample selection and arrangement means 80 _{obtains the} reduced signals I _{κ, L,} and F while changing the hierarchy and phase for each block according to the optimum sampling pattern P. The output signal D is generated by concatenating the frames I ₀ .

  For this reason, the adaptive sampling apparatus 1C changes the sampling interval and phase for each block based on the optimum sampling pattern that further minimizes the error in the frame as compared with the case where only the sampling interval is changed. , Sampling a frame, and outputting the sample result.

  The configuration of the adaptive sampling device 1C is not limited to this, and can be changed without departing from the spirit of the present invention.

[Operation of Adaptive Sampling Apparatus 1C]
Next, the operation of the adaptive sampling device 1C (1) according to the third embodiment of the present invention will be described with reference to FIG.

First, the adaptive sampling apparatus 1C uses the frame storage means 10 to convert the sampled amplitude values in a certain section of the input signal I into a frame I ₀ (t) (t = 0, 1,..., T−1; T Is stored as a natural number) (step S31). Then, the adaptive sampling apparatus 1C reduces the frame I ₀ stepwise and shifts the phase by the hierarchical signal reduction means 20C, and reduces the reduced signals I _{κ, L} (κ = 1, 2,..., H−1; H is a natural number of 2 or more (L is a phase corresponding to κ) (step S32).

Then, the adaptive sampling apparatus 1C uses the layered error calculation means 30C to perform error signal F _κ between the up-sampled enlarged signals U _{κ and L} (t) and the frame I ₀ for each reduced signal I _{κ and L. L} (t) is calculated and summed up for each block i (i = 0, 1,..., N−1) to generate error signals E _{κ, L} (i) (step S33).

Then, the adaptive sampling apparatus 1C uses the error signal E _{κ, L} (i) by the optimization unit 70 to calculate the sum of errors in the frame from the sampling patterns read from the sampling pattern database 50C. An optimum sampling pattern P to be optimized is obtained (step S34).

Then, the adaptive sampling apparatus 1C concatenates the reduced signals I _{κ and L} and the frame I ₀ while changing the layer (enlargement ratio) and the phase for each block according to the optimal sampling pattern by the sample selection / arrangement means 80, and outputs it. A signal D is generated (step S35).

  As mentioned above, although each apparatus was demonstrated according to embodiment about this invention, this invention is not limited to these embodiment, In the range which does not deviate from the meaning of this invention, it can change.

  For example, instead of the hierarchized signal reduction unit 20 of the adaptive sampling device 1B according to the second embodiment, the hierarchized signal reduction unit 20C described in the third embodiment can be applied.

  In this case, the sampling pattern database 50B stores the mode-specific sampling patterns without distinguishing those having different sampling intervals from each other. The assigning means 60 operates without distinguishing those having different sampling intervals.

Instead, the optimization unit 70B performs optimization for all or a part of the phase combinations.
That is, the optimization unit 70B first determines the mode-specific sampling pattern χ (m; i) for each mode number m and the error signal E _k (i) obtained by the hierarchization error calculation unit 30B. Find the sum of errors in the frame. At this time, the optimization means 70B can take a plurality of sampling phases when the hierarchy number χ (m; i) of the block number i of the mode number m (for example, for the hierarchy number 3, the integer value 0). In other words, the sum of errors is calculated for all or some of the cases.

Then, the optimization unit 70B obtains a sampling pattern sequence that minimizes the sum of the errors, and outputs it as an optimal sampling pattern sequence (Q _i ) _{i = 0, 1,.} Q _i is a pair (hierarchy number, phase number). Specifically, the calculation of equation (42) is performed.

In the case of hierarchy 0, ie, equal magnification, no error occurs, so E _0,0 (i) is always defined as 0 regardless of the block number.

  Further, for example, the adaptive sampling apparatus can also realize each means as a function program in a computer, and can also operate the adaptive sampling program by combining the function programs.

  Since the adaptive sampling apparatus according to the present invention can perform sampling at variable intervals according to the local complexity of the signal, it is possible to express a complex waveform with a small number of sampling points. By using this apparatus for sampling audio and images, the transmission band can be narrowed. The apparatus can also be regarded as a local time-direction equalizer. For example, when performing waveform analysis, the waveform can be deformed by this device, so that detailed waveforms can be observed in detail and coarse waveforms can be observed globally, facilitating the understanding of waveform trends. There is a possibility.

1 (1A, 1B, 1C) Adaptive sampling apparatus 10 Frame storage means 20 (20C) Hierarchical signal reduction means 21 (21C) Signal reduction means 30 (30B, 30C) Hierarchization error calculation means 31 Signal expansion means 32 Error calculation Means 40 Ordering means 50 (50B, 50C) Sampling pattern database 60 Allocation means 70 (70B) Optimization means 80 (80B) Specimen selection arrangement means

Claims (8)

An adaptive sampling device that divides a frame, which is a constant section of an input signal, into a plurality of blocks, and performs sampling at different sampling intervals or for each of the blocks,
Hierarchical signal reduction means for sampling the frame at a plurality of different sampling intervals with a reduced number of sampling points to obtain a reduced signal, and outputting the reduced signal group;
The reduced signal is returned to the same sampling interval as that of the input signal to be an enlarged signal, and an error between each enlarged signal and the frame is quantified for each block and output as an error signal;
Of the sampling pattern which is a method of assigning the sampling interval to be applied to each block of the frame when the frame is divided into the plurality of blocks, the enlargement ratio as the whole frame, A sampling pattern database that stores a plurality of sampling patterns that satisfy all of the constraint conditions consisting of the number of blocks and an enlargement factor that can be assigned to each block , and
Optimization means for searching for the sampling pattern for optimizing the error based on each error signal output from the layered error calculation means and the sampling pattern;
Sample selection arrangement means for sampling the frame of the input signal based on the sampling pattern obtained by the optimization means and outputting as an output signal;
An adaptive sampling device comprising: An adaptive sampling device that divides a frame, which is a constant section of an input signal, into a plurality of blocks, and performs sampling at different sampling intervals or for each of the blocks,
Hierarchical signal reduction means for sampling the frame at a plurality of different sampling intervals with a reduced number of sampling points to obtain a reduced signal, and outputting the reduced signal group;
The reduced signal is returned to the same sampling interval as that of the input signal to be an enlarged signal, and an error between each enlarged signal and the frame is quantified for each block and output as an error signal;
A method of assigning the number of blocks for each enlargement ratio that satisfies all of the constraints consisting of the enlargement ratio for the entire frame, the number of blocks in the frame, and the choice of enlargement ratio that can be assigned for each block is defined as one mode. A sampling pattern database that stores a record that records the sampling interval of the input signal and the number of blocks to which the plurality of different sampling intervals are assigned for each mode;
Based on the error signal output from the hierarchization error calculation means, ordering means for prioritizing the blocks in descending order of the error;
Based on the record for each mode and the priority order, a sampling pattern for each mode is generated by allocating the sampling interval having a higher number of sampling points to a block with higher priority for each block of the frame. Assigning means;
Optimization means for searching for the sampling pattern for optimizing the error based on each error signal output from the layered error calculation means and the sampling pattern;
An adaptive sampling apparatus comprising: a sample selection / arrangement unit that samples the frame of the input signal based on the sampling pattern obtained by the optimization unit and outputs the sample as an output signal.   2. The optimization unit according to claim 1, wherein, for each sampling pattern, a total error in the entire frame is obtained from each error signal, and the sampling pattern that minimizes the total error is searched. Or the adaptive sampling apparatus of Claim 2.   The optimization means obtains an error in the entire frame from each error signal for each sampling pattern, and this error is penalized according to a combination of sampling intervals between blocks included in the sampling pattern. The adaptive sampling apparatus according to claim 1, wherein the sampling pattern that minimizes the total error is searched for. The optimization means obtains an error in the entire frame from each error signal for each sampling pattern, and adds a penalty corresponding to the amount of code necessary to describe the sampling pattern to this error. The adaptive sampling apparatus according to claim 1, wherein the sampling pattern that minimizes the total error is searched for. The reduced signal generated by the hierarchical signal reduction means includes reduced signals having the same sampling interval and different sampling phases,
The adaptive sampling apparatus according to claim 1, wherein the optimization unit optimizes a sampling interval and a phase of the reduced signal for each of the blocks. In order to divide a frame, which is a constant section of an input signal, into a plurality of blocks and perform sampling at different sampling intervals or different sampling intervals for each block,
Hierarchical signal reduction means for sampling the frame at a plurality of different sampling intervals with a reduced number of sampling points to obtain a reduced signal, and outputting the reduced signal group,
Hierarchical error calculation means for returning each reduced signal to the same sampling interval as the input signal and making it an enlarged signal, quantifying the error between each enlarged signal and the frame for each block, and outputting it as an error signal,
Of the error signals output by the layered error calculation means and the sampling pattern that is the method of assigning the sampling interval to be applied to each block of the frame, the enlargement ratio of the entire frame, and the frame The error is optimized based on the sampling pattern read from the sampling pattern database that stores only a plurality of sampling patterns that satisfy all the constraint conditions consisting of the number of blocks in the block and the enlargement ratio options that can be assigned to each block. Optimization means for searching for the sampling pattern to be converted,
Sample selection arrangement means for sampling the frame of the input signal based on the sampling pattern obtained by the optimization means and outputting as an output signal;
Adaptive sampling program to function as In order to divide a frame, which is a constant section of an input signal, into a plurality of blocks and perform sampling at different sampling intervals or different sampling intervals for each block,
Hierarchical signal reduction means for sampling the frame at a plurality of different sampling intervals with a reduced number of sampling points to obtain a reduced signal, and outputting the reduced signal group,
Hierarchical error calculation means for returning each reduced signal to the same sampling interval as the input signal and making it an enlarged signal, quantifying the error between each enlarged signal and the frame for each block, and outputting it as an error signal,
Based on the error signal output from the hierarchization error calculation means, ordering means for prioritizing the blocks in descending order of the error,
A method of assigning the number of blocks for each enlargement ratio that satisfies all of the constraints consisting of the enlargement ratio for the entire frame, the number of blocks in the frame, and the choice of enlargement ratio that can be assigned for each block is defined as one mode. Based on the record read from the sampling pattern database storing the record in which the sampling interval of the input signal and the number of blocks to which the plurality of different sampling intervals are assigned for each mode, and the priority order. Assigning means for assigning the sampling interval having a higher number of sampling points to a block with higher priority for each block of the frame, and generating a sampling pattern for each mode;
Optimization means for searching for the sampling pattern that optimizes the error based on each error signal output by the hierarchization error calculation means and the sampling pattern;
Sample selection arrangement means for sampling the frame of the input signal based on the sampling pattern obtained by the optimization means and outputting as an output signal;
Adaptive sampling program to function as

Images