JP2002232295A - Method and device for quantizing vector - Google Patents

Abstract

PROBLEM TO BE SOLVED: To select the same representative vector as full search at high speed only by searching for the representative vectors of one part. SOLUTION: This method has arithmetic steps (F105 and F110) for calculating the distortions of input vector and representative vector and calculating the address of the representative vector to be next searched, decision steps (F106 and F111) for comparing two distortions and deciding the candidates of the representative vector to be encoded and storage steps (F109 and F114) for storing a sorted code book and by repeatedly performing these steps, the vector is encoded to a representative vector wmin of a number min having minimum distortion dcurmin.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a vector quantization technique, and more particularly to a vector quantization method and apparatus for pattern recognition, computational geometry, image signals, audio signals, and the like.

[0002]

2. Description of the Related Art Vector quantization is a technique for encoding an input space with a finite number of representative vectors. Application examples of vector quantization include feature extraction of pattern recognition, the latest problem of computational geometry, information compression, and quantization of images and sounds.

FIG. 21 is a conceptual diagram of vector quantization.
Vector quantization has two aspects: (1) encoding an input vector and (2) generating a codebook.

[0004] In the vector quantization (1) procedure of encoding an input vector, a code consisting of N representative vectors (sometimes called a code vector) given in advance to an input vector x (1050) is used. book ^{_{W = {w i ∈R K:}} i = 1, ···, N} (1040)
, The closest representative vector w _min is selected (1
060), the representative vector number min or the value of the representative vector is output (1070).

[0005] At that time, as a measure of when choosing a closest representative vectors, non-negative strain measure d (x, w _i) utilize. That is, the expression for outputting the number of the representative vector can be expressed by the following expression (1).

[0006]

(Equation 1)

An expression for outputting the value of the representative vector can be expressed by the following expression (1).

[0008]

(Equation 2)

In (2) codebook generation of vector quantization, an initial codebook (1010) is set, and an input vector set (103) for codebook generation is set.
0) and repeatedly modifying the value of the representative vector to match the input vector set to optimize (102)
0), finally obtaining the optimal codebook (1040). The input vector for generating the codebook may be different from the input vector input in (1) encoding of the input vector.

The above (1) encoding of an input vector is based on a full search for calculating the distortion with respect to one representative vector and all representative vectors and selecting the nearest representative vector. The problem is that it takes much time.

As one of the techniques for solving this problem, there is a vector quantization method and a vector quantization apparatus disclosed in Japanese Patent Application Laid-Open No. 4-156181. This vector quantization technique is described in the literature [Keiichi Nakano, Hironori Kasahara: "Fast vector quantization using sorted codebooks (Fast Vec
tor Quantization Using Sorted Codebook (VQ-SC) ”
IEICE Transactions D-II Vol.J77-D-II No.10 pp.1
984-1992, 1994], and VQ-SC
Is roughly the following algorithm: (1) The representative vectors are sorted for each dimension, and the order is stored. (2) Clear the flag indicating whether or not the searched representative vector has been reached. (3) Calculate the strain that is the lower limit of the strain. (4) A search for a representative vector that does not improve the accuracy of the solution by using the upper limit of the distortion based on the principle of the branch and bound method is omitted. The searched representative vector sets a flag. When the search for the representative vector used for the lower limit of the distortion is omitted, the lower limit of the distortion is updated. The search is continued until the upper limit of the strain and the lower limit of the strain match. The method of determining the next representative vector to be searched is heuristic.

By using the upper limit value of the strain and the lower limit value of the strain, if a part of the codebook is examined, the same representative vector as in the case of performing a full search can be searched at high speed.

[0013]

However, the above method has the following problems. (1) It is necessary to sort the representative vectors for each dimension, and the higher the dimension, the longer the time required for sorting. (2) It is necessary to store the order of the representative vectors sorted for each dimension, and the memory capacity to store becomes enormous. (3) It is necessary to clear a flag indicating which representative vector has been searched for each input vector,
If there are many input vectors, the time to clear becomes enormous.
Also, a memory capacity for flags is required for the number of representative vectors. (4) Since the next representative vector to be searched may be determined heuristically, it is necessary to consider an efficient determination method.

An object of the present invention is to provide a vector quantization method and a vector quantization apparatus which can solve the above-mentioned problems.

[0015]

Here, for a K-dimensional input vector x, a representative vector w having the minimum distortion is set.
When searching for a _min, for a dimension k, it is not necessary to search for the representative vector w _i such that the following equation (3).

[0016]

(Equation 3)

However, equation (3) shows that the encoded w
_This is a condition when _min is obtained, and the equation (3) cannot be calculated. In addition, the values of the representative vectors for the k-dimension are not arranged regularly, and the order is in a random state, so that it is not possible to determine the search termination condition.

Therefore, the present invention employs the following method.
To use. (1) The smallest representative vector found by the search so far
W^cur _minAnd That is, when the search ends
If finally obtained d^cur _minRepresentative vector
w^cur _minIs the representative vector w with the least distortion
_min = W^cur _minBecomes (2) Before starting the search, the representative
Sort the vectors in ascending (or descending) order.
If there is a representative vector satisfying equation (3), sort
The representative vector beyond that has the minimum distortion
There is no possibility to be a representative vector with. Sand
In other words, by sorting the representative vectors, the search
The rate can be greatly increased.

Considering the calculation of the distortion in equation (3), the range in which the value of the representative vector of each dimension can be generally different. Therefore, choosing the one with the widest range should reduce the number of searches. In the gradation conversion of an image, R, G, and B take values of 0 to 255 in any dimension. Therefore, it is expected that the same number of times will be obtained even if any dimension is selected.

The method of terminating the search is determined by sorting, but it is desirable to terminate the search with a minimum number of times. For that purpose, it is important to determine the minimum representative vector w ^cur _min set first. It is expected that the search efficiency will greatly change depending on how to determine the initial representative vector.

It is considered that a method of determining an initial representative vector to be searched for in the following two cases is an efficient choice by setting as follows.

When an input vector is encoded into a predetermined representative vector, a representative vector obtained by encoding adjacent input data is generally used as an initial value.

In the case of a still image, since there is a high possibility that adjacent pixels have the same color, the representative vector of the adjacent pixel is set as an initial value.

In the case of a moving image, two methods are conceivable.
One is to set a representative vector of an adjacent pixel as an initial value as in the case of a still image. The other is to use the similarity between pictures between frames, and set a representative vector at the same position in the previous frame as an initial value.

A vector quantization method according to the present invention calculates a distortion of an input vector and a representative vector, and calculates an address of a representative vector to be searched next,
It has a determining step of comparing two distortions to determine a candidate of a representative vector to be encoded, and a storing step of storing a sorted codebook, and the above steps are repeatedly performed.

The vector quantization method according to the present invention calculates a distortion between an input vector and a plurality of representative vectors in parallel, calculates an address of a representative vector to be searched next, and calculates a plurality of distortions. The method includes a determining step of determining a candidate of a representative vector to be encoded by comparison, and a storing step of storing a sorted codebook, wherein the above steps are repeatedly performed.

In the vector quantization method according to the present invention, a sorting step of rearranging a codebook, an operation step of calculating distortion of an input vector and a representative vector, and calculating an address of a representative vector to be searched next, A determining step of comparing two distortions to determine a candidate of a representative vector to be encoded; a storing step of storing a sorted codebook; and a storing step of storing a sorted order of the codebook. ,
Each of the above steps is repeatedly performed.

Further, in the vector quantization method according to the present invention, the sorting step of rearranging the codebook, the distortion between the input vector and the plurality of representative vectors are calculated in parallel, and the address of the next representative vector to be searched is calculated. And a determining step of comparing a plurality of distortions to determine a candidate of a representative vector to be encoded, and a storing step of storing a sorted codebook, wherein the above steps are repeatedly performed. And

The vector quantization apparatus according to the present invention calculates the distortion of the input vector and the representative vector, and calculates the address of the representative vector to be searched next. It is characterized by comprising a determination means for determining a candidate of a representative vector, a storage means for storing a sorted codebook, and a control means for controlling a processing flow of the calculation means and the determination means.

Further, the vector quantization apparatus according to the present invention calculates in parallel the distortion between the input vector and the plurality of representative vectors, calculates the address of the next representative vector to be searched, and calculates the plurality of distortions. Determining means for determining a candidate of a representative vector to be encoded by comparison, storage means for storing a sorted codebook, and control means for controlling a processing flow of the calculating means and the determining means. I do.

Further, the vector quantization apparatus according to the present invention includes a sorting means for rearranging the codebook, an arithmetic means for calculating the distortion of the input vector and the representative vector, and calculating the address of the representative vector to be searched next. 2
Determining means for determining a candidate of a representative vector to be encoded by comparing two distortions, codebook storage means for storing a codebook, storage means in which the sorted order of sorted codebooks is described, A control unit for controlling a processing flow of the calculation unit and the determination unit.

Further, the vector quantization apparatus according to the present invention calculates sorting of the codebook in parallel, calculates the distortion between the input vector and the plurality of representative vectors in parallel, and calculates the address of the next representative vector to be searched. Calculating means for comparing a plurality of distortions to determine a candidate of a representative vector to be encoded; storage means for storing a sorted codebook; and processing flow of the sorting means, the calculating means, and the determining means And control means for controlling

[0033]

Embodiments of the present invention will be described below in detail with reference to the drawings.

The present invention is applied to, for example, a vector quantizer 100 having a configuration as shown in FIG.

This vector quantizer 100 has an input unit 1
10, sorting device 120, codebook memory 1
30, arithmetic device 140, determination device 150, and control device 1
60.

In the vector quantizer 100, the input unit 110 receives a new codebook from the input terminal 101. Then, the input unit 110 gives an address to write the codebook and a new codebook to the codebook memory 130 that stores the codebook.

The sorting device 120 includes the control device 16
Upon receiving a sorting instruction from 0 and an instruction on which dimension k is to be sorted, the code book is read from the code book memory 130 and the code book is sorted based on the specified dimension. Also, the sorting device 12
0 is a codebook memory 1 for storing sorted codebooks.
Write to the specified 30 addresses.

The codebook memory 130 stores the input unit 11
Receives the address to write from 0 and the new codebook and stores the codebook at the specified address. Further, the codebook memory 130 transfers the codebook to the sorting device 120. Further, the codebook memory 130 receives the codebook sorted by the sorting device 120 and the address to be written, and stores the codebook at the specified address. further,
The codebook memory 130 receives an address from the arithmetic unit 140 that performs strain calculation and address calculation, and passes representative vector data (representative vector value w _i , number i, etc.).

The arithmetic unit 140 receives an instruction for calculating a distortion / address from the control unit 160, and specifies the address of the representative vector to be read into the codebook memory 130 (the initial representative vector is specified by the control circuit). And the corresponding data (representative vector number and value). The arithmetic unit 140 receives the input vector x from an input terminal 102, calculates the distortion d (x, _{w i)} of the input vector the representative vector. The calculation unit 140 is passed to the strain d (x, _{w i)} a determination device data (number and value of the representative vector) 150. Further, the arithmetic unit 140 receives an instruction of the dimension k serving as a sorting criterion from the control unit 160, and receives a k-dimensional distortion d (x _k ,
w _ik ) is also passed to the determination device 150. Further, the arithmetic unit 1
40 receives up / down information indicating the direction of the representative vector to be searched based on the determination result from the determination device 150, and calculates the address of the next representative vector to be searched.
Note that the arithmetic device 140 ends the search for the representative vector when receiving the information finish indicating the search end.

The determination device 150 determines the current minimum distortion d.
^cur _min and the number min of the representative vector are held, and mi having the current minimum distortion d ^cur _min
Strain representative vector of the calculated i-th by the representative vector and the arithmetic unit 140 of the n-th comparing d (x, _{w i),} distortion smaller new min number, the current minimum distortion ^d _{cur min} . Also, the determination device 150
Outputs the number min (index) of the representative vector having the current minimum distortion or the value of the representative vector to the output terminal 103 in accordance with an instruction from the control device 160. Further, the determination device 150 receives a determination instruction from the control device 160, and receives a k-dimensional distortion d (x _k , w _ik ).
And the current minimum distortion d ^cur _min , and d (x
_{If k} , w _ik ) is larger, the information of up / down indicating the direction of the next representative vector to be searched is changed. If both the up and down directions have been checked, the information finish indicating search end is changed. Passed to the arithmetic unit 140 and d
If (x _k , w _ik ) is smaller, information for examining the next representative vector in the same direction is passed to the arithmetic unit 140.

The control device 160 controls the sorting device 12
0, the processing flow of the arithmetic unit 140 and the determination unit 150 is controlled. Further, the control device 160 designates a dimension serving as a sorting reference to the sorting device 120. Further, the control device 160 specifies the address (or number) of the initial representative vector to the arithmetic device 140, and also indicates the dimension used as the reference for sorting. Further, the control device 160 instructs the determination device 150 whether the output to the output terminal 103 is the number of the representative vector or the value of the representative vector.

The vector quantizer 100 having the above configuration
In the present invention, high-speed vector quantization (Vector Quantization) using representative vectors sorted according to the encoding algorithm shown in the flowchart of FIG.
by Searching in Sirted Codebook (VQ-SSC).

That is, in the vector quantizer 100, first, in step F101, a reference dimension k for sorting representative vectors is designated.

In the next step F102, the representative vectors are sorted based on the value of the representative vector of dimension k.

Regarding the information of the sorted order, the following two methods are conceivable.

(1) The order of the representative vectors is directly changed. If it is not desired to change the number of the representative vector, the structure is replaced with a structure having information on the number of the representative vector.

Since the numbers are directly changed, the lower number of the representative vector lower than the own is the number of the array whose number is −1 of the own array, and the upper number of the representative vector higher than the own is the upper number of the own array. It becomes the number of the array with the number + 1.

(2) When an array indicating the order of sorting is prepared, the information that each representative vector should hold is the lower number of the representative vector lower than itself and the upper number of the representative vector higher than itself. is there.

Here, the steps F101 and F102 need to be performed only once unless the codebook is changed.

In the next step F103, the input vector
x and initial representative vector w_i And calculate the strain with
This strain d (x, w_i) Is the current minimum strain d
^cur _minAnd store the number i as a candidate for encoding.
Keep it.

[0051] ^{_{d cur min = d (x,}} w i) (4) min = i (5) In the next step F 104, the number of the representative vector on both sides of the initial representative vector _{w i,} i.e., lower = lower direction The number of the next representative vector upper = the number of the next representative vector in the upper direction is stored.

In the next step F105, the input vector
x and the representative vector w of the lower number _lowerWith the strain
Only d (x, w_lower) And calculate d
(X _k, W_lowerk) Is saved.

In the next step F106, step F1
05 strain was calculated _{d (x, w l ower)} whether the current minimum distortion ^d _{cur min} smaller, i.e., _d whether _{^{(x, w lower) <d}} cur min (6) is satisfied judge. If this equation (6) is satisfied, the process proceeds to the next step F107, and otherwise proceeds to step F108.

In step F107, d ^cur _min = d (x, w _lower ) (7) min = lower (8)

In the next step F108, it is determined whether to continue the search in the lower direction. In this step F108, if d (x _k , w _lowerk )> d ^cur _min (9) is satisfied, the search in the lower direction is completed, and step F
Go to 110, otherwise go to the next step F109.

Step F109 is a step of updating the representative vector number lower. The process returns to step F105 by setting the number of the next lower representative vector of the current lower representative vector as the new lower.

[0057] In step F110, the strain of the representative vector _{w up per} input vector x and the upper number d
(X, w _upper ) is calculated, and d (x
_k , w _upperk ) are saved.

In the next step F111, step F1
Strain was calculated by _{10 d (x, w u pper} ) whether the current minimum distortion ^d _{cur min} smaller, i.e., _d whether _{^{(x, w upper) <d}} cur min (10) is satisfied judge. If this expression (10) is satisfied, the process proceeds to the next step F112, and otherwise proceeds to step F113.

In step F112, d ^cur _min = d (x, w _upper ) (11) min = upper (12)

In the next step F113, it is determined whether or not the search in the upper direction is to be continued. In this step F113, if d (x _k , w _upperk )> d ^cur _min (13) is satisfied, the search in the upper direction is completed, and step F
Go to 115, otherwise go to the next step F114.

In step F114, the representative vector number u
This is a step of updating the upper, in which the number of the representative vector next to the current upper-numbered representative vector in the upper direction is set as a new upper and the process returns to step F110.

In the next step F115, the input vector x
Is the representative vector w_minTo be encoded. This step F
At 115, the input vector x has the minimum distortion d
^cur _minRepresentative vector w of number min with_min
Is encoded.

Here, an encoding algorithm (VQ-SS) for vector quantization in the vector quantizer 100
The operation of C) will be described using an example.

In this example, a square error distortion measure is used as the distortion measure.

D (x, w) = (x−w) ² (14) Naturally, the present invention is also effective when other strain measures are used. The present invention is also effective when a different strain measure is used for each dimension.

A three-dimensional input vector x = (12,16,
FIG. 3 shows an example in which the distortion is calculated for all the representative vectors as in the conventional method, and the representative vectors to be subjected to vector quantization (encoding) are obtained. In this case, 12
If distortion is not calculated for the representative vectors, the representative vector to be encoded cannot be determined. In this example,
The 9th representative vector w ₉ = (14,
18, 3).

On the other hand, in the coding algorithm for vector quantization (VQ-SSC), representative vectors are sorted for k = 1. The representative vectors are numbered in the sorted order for easy understanding.
Actually, it is not necessary to renumber the representative vectors, and the program may be programmed to save the numbers of the representative vectors on both sides (upper direction / lower direction) of the own representative vector.

If the representative vector to be encoded is known in advance, since the minimum distortion is 8, the maximum integer less than or equal to √8, that is, the range of 2 or less may be searched. From FIG. 3, only the three representative vectors (7), (8), and (9) need to be examined for the representative vectors that become 12 ± 2 in one dimension.

However, since the representative vector to be encoded cannot be known in advance, the search is performed using the following algorithm (search 0 to search 6).

That is, in search 0, as shown in FIG. 4, the initial representative vector is set as the (7) th representative vector, and the distortion d (x, w ₇ ) is calculated.

D (x, w ₇ ) = 17 lower = 6 upper = 8 Then, the minimum distortion d ^cur _min = 17 and its representative vector number min = 7 are stored.

In search 1, as shown in FIG.
The distortion d (x, w) of the representative vector of r (6)
_lower ).

D (x ₁ , w _lower1 ) = 9 d (x, w _lower ) = 189 Then, when d (x, w _lower ) is compared with the current minimum distortion d ^cur _min , d (x, w _lower )
Since w _lower ) is greater than d ^cur _min , the minimum distortion d ^cur _min retains its value, that is, d ^cur _min = 17 min = 7. _{Also, d (x 1, w lower1} ) because the current minimum of less than the strain ^{d cu} _{r min,} moves to explore 2 of the next lower = 5.

In the next search 2, as shown in FIG.
The distortion d (x, w) of the representative vector of the wer (5)
_lower ).

D (x ₁ , w _lower1 ) = 9 d (x, w _lower ) = 14 Then, when d (x, w _lower ) is compared with the current minimum distortion d ^cur _min , d (x, w _{lower 1} )
Since w _lower ) is smaller than d ^cur _min , the minimum distortion d ^cur _min is updated to d ^cur _min = 14 min = 5. D (x ₁ , w _lower1 ) is d
Since it is smaller than ^cur _min = 14, the next lower
Move to search 3 where = 4.

In search 3, as shown in FIG.
The distortion of the representative vector of r (4) is calculated.

D (x ₁ , w _lower1 ) = 16 d (x, w _lower ) = 186 Then, when d (x, w _lower ) is compared with the current minimum distortion d ^cur _min , d (x, w _lower )
Since w _lower ) is greater than d ^cur _min , the minimum distortion d ^cur _min retains its value, that is, d ^cur _min = 14 min = 5. D (x ₁ , w _lower1 ) is d
Since it is larger than ^cur _min = 14, the search in the lower direction ends, and the process then proceeds to search 4 in the upper direction with upper = 8.

In search 4, as shown in FIG.
The distortion of the representative vector of r (8) is calculated.

[0079] _{d (x 1, w upper1)} = 1 d (x, w upper) and _{= 46, d (x, w} upper) if the comparing the current minimum distortion ^d _{cur min,} ^d (x,
Since w _lower ) is greater than d ^cur _min , the minimum distortion d ^cur _min retains its value, that is, d ^cur _min = 14 min = 5. D (x ₁ , w _upper1 ) is d
^{Since cur} _min = 14, the next upper
Move to search 5 where = 9.

In search 5, as shown in FIG.
The distortion of the r (9) -th representative vector is calculated.

D (x ₁ , w _{upper 1} ) = 4 d (x, w _upper ) = 8 Then, when d (x, w _upper ) is compared with the current minimum distortion d ^cur _min , d (x, w _upper )
Since w _lower ) is smaller than d ^cur _min , the minimum distortion d ^cur _min is updated so that d ^cur _min = 8 min = 9. D (x ₁ , w _upper1 ) is d ^cur
Since it is smaller than _min = 9, the process moves to the next search 6 of upper = 10.

In search 6, as shown in FIG.
The distortion of the representative vector of er (10) is calculated.

D (x ₁ , w _{upper 1} ) = 9 d (x, w _upper ) = 41 Then, when d (x, w _upper ) is compared with the current minimum distortion d ^cur _min , d (x, w _upper )
Since w _lower ) is greater than d ^cur _min , the minimum distortion d ^cur _min retains its value, that is, d ^cur _min = 8 min = 9. (X ₁ , w _upper1 ) is d
Since it is larger than ^cur _min = 9, the search in the upper direction ends.

The search is completed in both the lower direction and the upper direction in this way, and the finally obtained minimum distortion d
^The min = 9 representative vector with ^cur _min = 8 is the representative vector to be encoded for the input vector x = (12,16,13).

Here, from FIG. 11, if the representative vector to be encoded is known in advance, it is sufficient to search for three representative vectors. However, since it is generally impossible, it is necessary to determine and search for the initial representative vector. There is. In this case, since the search was performed while updating the minimum distortion, a representative vector having the minimum distortion could be obtained by searching for seven representative vectors. In the gradation conversion, since the number of representative vectors is 256, it can be expected that the search range is greatly limited.

As described above, the encoding algorithm (VQ-SS) for vector quantization in the vector quantizer 100
The operation of C) has been described using an example.
Simulation results of encoding a still image by the vector quantizer 100 (FIGS. 12 and 13) and simulation results of encoding a moving image (FIGS. 14 and 1)
5) will be described.

As an example of the encoding of a still image by the vector quantizer 100, gradation conversion for reducing the color of a full-color image of 8 bits (R, G, B) to 256 colors will be described. Codebooks are created for each image. The sort indicates the number of searches for the representative vector when each of R, G, and B is selected.

As a test image, the 24 BMP files shown in FIGS. 17, 18, 19 and 20 included in Kodak PHOTOCD are also subjected to the following two methods.
The gradation was converted to 56 colors. (1) It can be seen from FIG. 12 that a representative vector to be coded has been found by searching about 20 or less on average regardless of which color is selected. In other words, it is not necessary to care much about the colors used for sorting, and the same number of searches is performed regardless of which color is selected. (2) 24 BMPs from Kodock PHOTOCD
From the application of the file, the color with the least number of searches differs for each picture. However, in the case of Kodak PHOTOCD, it can be seen from FIG. 12 and FIG. 13 that if sorting is performed on the basis of red, a representative vector that can be encoded faster as an average can be found.

Also, when used for encoding of a moving image or the like, if the representative vector of the previous frame is selected as the initial representative vector or the representative vector of an adjacent pixel is selected as the initial representative vector, the representative vector to be searched is determined. Will be reduced.

In the simulation, Flower G
For 10 consecutive frames of Arden, 25
Encode into 6 color representative vectors (no scene change). The 256-color representative vectors are designed for the first frame. There are two ways to select an initial representative vector: (1) when the representative vector of the previous frame is selected, and (2) when the representative vector of the adjacent pixel is selected. FIG. 14 shows the number of searches and the search time. FIG. 15 shows the relationship between each frame and the number of searches.
However, the time was measured in the following environment.

CPU: Pentium II 400MH
z Memory: 64 Mbytes OS: Windows NT 4.0 As described above, when the pixels of the previous frame are used as the initial representative vector, the average number of searched representative vectors is about 44.
, And about 27 when adjacent pixels are used. In each case, a representative vector to be coded is found in 1 / 5.82 and 1 / 9.48 of the full search. In Flower Garden this time, better results could be obtained by using neighboring pixels for the initial representative vector.

Further, the present invention is more effective when a high-speed technique using a partial sum of distortions is combined. The high-speed technique using the partial sum of distortion searches for the next representative vector when the partial sum of distortion represented by the following equation is larger than the current minimum distortion d ^cur _min .

[0093]

(Equation 4)

That is, this is a method of increasing the efficiency of the search by comparing the distortion in the middle of calculation with the current minimum distortion, instead of calculating the distortion to the end. A high-speed method using this partial sum of distortions is described in the literature [Bei CD. & Gray,
RM. “An improvement of minimum distortion enco
ding algorithm for vector quantization. ”IEEE Tra
nsactions on Communications, COM-33, pp.1132-1133,
1985.].

Here, only a difference will be described with reference to FIG. 16 as a modification of the above-described modification of the vector quantizer 100 shown in FIG.

The code book memory 130 passes the code book and the address to the sorting device 120.

[0097] Sorting device 120 is controlled by control device 16.
Upon receiving a sort instruction from 0 and an instruction on which dimension k is to be sorted, the code book address and the code book are read from the code book memory 130, and the code book is sorted based on the designated dimension. Then, the addresses of the representative vectors are rearranged in the order of the sorted representative vectors, and the addresses arranged in the sorted order are written in the codebook sort table 170.

The arithmetic unit 140 determines from the determination unit 150
Up indicating the direction of the representative vector to be searched based on the determination result
In response to the information of / down, the address of the next representative vector to be searched is referred to the codebook sort table 170, and the address is received. Then, the address is specified and the corresponding data (representative vector number and value) is read into the codebook memory 130.

Then, the codebook sort table 170 is
The address received from the sorting device 120 is stored. Further, the codebook sort table 170 stores the address of the representative vector in accordance with the instruction from the arithmetic unit 140 to refer to the next representative vector to be searched.
Pass to.

[0100]

As described above, according to the present invention,
By searching only a part of the representative vectors, the same representative vector as in the full search can be selected at high speed. Further, according to the present invention, the number of memories to be stored can be reduced as compared with the conventional method. Further, according to the present invention, as compared with the conventional method, since the sorting only needs to be performed for any one dimension, the time can be reduced. Further, according to the present invention, as compared with the conventional method, there is no need for a memory for storing the order of the sorted representative vectors. According to the present invention,
Compared with the conventional method, a memory for a flag indicating whether or not the search for the representative vector has been completed is not required, and no time is required to clear the flag for each input vector. Further, according to the present invention, the method of determining the next representative vector to be searched is uniquely determined.

[Brief description of the drawings]

FIG. 1 is a block diagram illustrating a configuration of a vector quantizer to which the present invention has been applied.

FIG. 2 is a flowchart showing an encoding algorithm (VQ-SSC) for vector quantization in the vector quantizer.

FIG. 3 is a diagram showing an example of searching for a representative vector by full search.

FIG. 4 is a diagram illustrating a search example (search 0) of a representative vector by the encoding algorithm (VQ-SSC).

FIG. 5 is a diagram showing an example of searching for a representative vector (search 1).

FIG. 6 is a diagram showing a search example (search 2) of a representative vector.

FIG. 7 is a diagram showing an example of searching for a representative vector (search 3).

FIG. 8 is a diagram showing an example of searching for a representative vector (search 4).

FIG. 9 is a diagram showing an example of searching for a representative vector (search 5).

FIG. 10 is a diagram showing an example of searching for a representative vector (search 6).

FIG. 11 is a diagram showing a search range of the encoding algorithm (VQ-SSC).

FIG. 12 is a diagram showing a simulation result of the encoding of a still image by the above-mentioned encoding algorithm (VQ-SSC) (comparison of the number of searches for a representative vector when each of R, G, and B is selected for sorting).

FIG. 13 is a diagram showing the number of searches for a representative vector of gradation conversion for Kodak PHOTOCD.

FIG. 14 is a diagram illustrating encoding of a moving image (Flower Garden).

FIG. 15 is a diagram showing the number of searches when a moving image (Flower Garden) is subjected to gradation conversion using the codebook of the first frame.

FIG. 16 is a block diagram showing a modification of the vector quantizer to which the present invention is applied.

FIG. 17 is a diagram showing six Kodak PHOTOCDs (24).

FIG. 18 is a diagram showing six Kodak PHOTOCD (24).

FIG. 19 is a diagram showing six Kodak PHOTOCD (24).

FIG. 20 is a diagram showing six Kodak PHOTOCDs (24).

FIG. 21 is a conceptual diagram of vector quantization.

[Explanation of symbols]

100, 200 Vector quantizer, 101, 102
Input terminal, 103 output terminal, 110 input section, 120
Sorting device, 130 codebook memory, 1
40 arithmetic unit, 150 judgment unit, 160 control unit, 170 codebook sort table

 ──────────────────────────────────────────────────続 き Continued on the front page F term (reference) 5C059 KK06 MC18 MD04 MD07 PP04 UA02 5J064 AA03 BA13 BB14 BC01 BC02 BC14 BC25 BD01

Claims (32)

[Claims] An arithmetic step of calculating a distortion between an input vector and a representative vector and calculating an address of a representative vector to be searched next; and a determining step of comparing two distortions to determine a candidate of a representative vector to be encoded. And a storage step of storing the sorted codebook, wherein the above steps are repeatedly performed. 2. The vector quantization method according to claim 1, further comprising sorting for rearranging the codebook. 3. The vector quantization method according to claim 1, wherein in the determining step, a partial sum of distortion is compared. 4. The vector quantization method according to claim 1, wherein a representative vector in which a previous input vector is encoded is designated as an initial representative vector. 5. The vector quantization according to claim 4, wherein a vector of a moving image is quantized by designating a representative vector in which an input vector at the same position in the previous frame is encoded as an initial representative vector. Method. 6. A computing step of calculating distortions of an input vector and a plurality of representative vectors in parallel, calculating an address of a representative vector to be searched next, and comparing a plurality of distortions with a representative vector to be encoded. A vector quantization method, comprising: a determination step of determining a candidate; and a storage step of storing a sorted codebook, wherein the above steps are repeatedly performed. 7. The vector quantization method according to claim 6, further comprising sorting for rearranging the codebook. 8. The vector quantization method according to claim 6, wherein a representative vector in which a previous input vector is encoded is designated as an initial representative vector. 9. The vector quantization according to claim 8, wherein a vector of a moving image is vector-quantized by designating a representative vector in which an input vector at the same position in the previous frame is encoded as an initial representative vector. Method. 10. A sorting step of rearranging a codebook, a calculating step of calculating a distortion of an input vector and a representative vector, and calculating an address of a representative vector to be searched next. A determination step of determining a candidate of a representative vector to be stored, a storage step of storing a sorted codebook, and a step of storing an arrangement order of the sorted codebook, wherein the above steps are repeatedly performed. Vector quantization method. 11. The method according to claim 11, wherein:
11. The vector quantization method according to claim 10, wherein a comparison is made between partial sums of distortions. 12. The vector quantization method according to claim 10, wherein a representative vector in which a previous input vector is encoded is designated as an initial representative vector. 13. The vector quantization according to claim 12, wherein a vector image of a moving image is vector-quantized by designating a representative vector in which an input vector at the same position in the previous frame is encoded as an initial representative vector. Method. 14. A sorting step of rearranging a code book, an operation step of calculating distortion of an input vector and a plurality of representative vectors in parallel, and calculating an address of a representative vector to be searched next, and comparing the plurality of distortions. A vector quantization method, comprising: a determination step of determining a candidate of a representative vector to be encoded by encoding; and a storage step of storing a sorted codebook, wherein the above steps are repeatedly performed. 15. The vector quantization method according to claim 14, wherein a representative vector in which a previous input vector is encoded is designated as an initial representative vector. 16. The vector quantization according to claim 14, wherein a vector of a moving image is vector-quantized by designating a representative vector in which an input vector at the same position in the previous frame is encoded as an initial representative vector. Method. 17. A calculating means for calculating a distortion between an input vector and a representative vector and calculating an address of a representative vector to be searched next, and a judging means for comparing two distortions to determine a candidate for a representative vector to be encoded. And a storage means for storing the sorted codebook; and a control means for controlling a processing flow of the calculation means and the determination means. 18. The vector quantization apparatus according to claim 17, further comprising sorting means for rearranging the codebook. 19. The vector quantization apparatus according to claim 17, wherein said determination means further performs a comparison of partial sums of distortions. 20. The vector quantization apparatus according to claim 17, wherein a representative vector in which a previous input vector is encoded is designated as an initial representative vector. 21. Vector quantization according to claim 20, wherein a moving image is vector-quantized by designating a representative vector in which an input vector at the same position in the previous frame is encoded as an initial representative vector. Device. 22. An arithmetic unit for calculating a distortion between an input vector and a plurality of representative vectors in parallel, calculating an address of a representative vector to be searched next, and a representative vector encoded by comparing the plurality of distortions. A vector quantization apparatus comprising: a determination unit that determines a candidate; a storage unit that stores a sorted codebook; and a control unit that controls a processing flow of the calculation unit and the determination unit. 23. The vector quantization apparatus according to claim 22, further comprising a sorting apparatus for rearranging codebooks. 24. The vector quantization apparatus according to claim 22, wherein a representative vector in which a previous input vector is encoded is designated as an initial representative vector. 25. The vector quantization according to claim 24, wherein a moving image is vector-quantized by designating a representative vector in which an input vector at the same position in the previous frame is encoded as an initial representative vector. Device. 26. A sorting means for rearranging a codebook, a calculating means for calculating distortion of an input vector and a representative vector, and calculating an address of a representative vector to be searched next. Determining means for determining a representative vector candidate to be stored, a codebook storing means for storing a codebook, a storing means in which the sorted order of the codebook is described, and a processing flow of the sorting means, the calculating means and the determining means. A vector quantization device, comprising: control means for controlling. 27. The vector quantization apparatus according to claim 26, wherein said determination means further performs comparison of partial sums of distortions. 28. The vector quantization apparatus according to claim 26, wherein a representative vector in which a previous input vector is encoded is designated as an initial representative vector. 29. The vector quantization according to claim 28, wherein a moving image is vector-quantized by designating a representative vector in which an input vector at the same position in the previous frame is encoded as an initial representative vector. Device. 30. A sorting means for rearranging a codebook, an arithmetic means for calculating a distortion between an input vector and a plurality of representative vectors in parallel and calculating an address of a representative vector to be searched next, and comparing the plurality of distortions. Determining means for determining a candidate of a representative vector to be coded by encoding, storage means for storing a sorted codebook, and control means for controlling a processing flow of the sorting means, the calculating means and the determining means. Characteristic vector quantization device. 31. The vector quantization apparatus according to claim 30, wherein a representative vector in which a previous input vector is encoded is designated as an initial representative vector. 32. Vector quantization according to claim 31, wherein a moving image is vector-quantized by designating a representative vector in which an input vector at the same position in the previous frame is encoded as an initial representative vector. Device.