JP3218220B2 - Image information compression method and image information compression system - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image information compression system, and more particularly to a three-dimensional pseudo-Hilbert scanning method for a rectangular parallelepiped region in a three-dimensional space and an image information compression system thereof.

[0002]

2. Description of the Related Art As an example of a curve group called a space filling curve, D. Hilbert in 1891
There is Hilbert curve shown by, image information compression method using a two-dimensional piano curve scanning similar thereto has been proposed (see Kokoku No. 7-22345).

FIG. 5 is a diagram showing an example of such a conventional two-dimensional Hilbert curve. FIG. 5 (a) shows a case where pixels are 2 × 2, FIG. 5 (b) shows a case where pixels are 4 × 4, FIG. 5C shows a case where the number of pixels is 8 × 8. This curve has been applied to various researches such as spectral image classification, database information search, image compression, and computer generated hologram in recent years because of its good neighborhood preservation. In such applications, the Hilbert curve is used to process data distributed in a two- or three-dimensional space. In general, an N-dimensional space (N = 2, 3,
..) Has a one-to-one correspondence between grid points and one-dimensional points, and scanning along a Hilbert curve is called Hilbert scanning.

Further, there are the following prior art documents relating to the image information compression method. (1) Takeshi Yasui, Takanori Nagae, Masayuki Nakajima, "Scanning Arrays of Arbitrary Size by Generalizing Peano Scanning" Television Technical Report, Vol. 14, No. 37 pp. 25-30,
Jul. 1990. (2) Yukihiro Bando and Seiichiro Kamada, "Pseudo-Hilbert Scanning Method for Filling Rectangular Regions", IEICE (D-II), Vol. J
80-D-II, No. 10, pp. 2864-286
7, Oct. 1997. (3) Seiichiro Kamata, Perez Arnolfo, Eiji Kawaguchi,
"Method for calculating Hilbert curve in two- and three-dimensional space", IEICE (D-II), Vol. J74-D-II, No. 9,
pp. 1217-1226, Sep. 1991. (4) X. Liu, G .; F. Schrack, “An
Algorithmism Encoding and
Decoding the 3-D Hilbert
Order, "IEEE Trans. Image"
Processing. , IP-6, No. 9, pp.
1333-1337. Sep. 1997.

[0005]

However, the problem when employing Hilbert scanning is that its application range is limited.

[0006]

(Equation 1)

[0007] This is limited to the region of the N-dimensional space that satisfies. On the other hand, in a general image, a scanning target is a rectangle, and therefore, ordinary two-dimensional Hilbert scanning cannot be applied in many cases. Therefore, as a generalization of the two-dimensional Hilbert scanning, a scanning method for a rectangular area has been proposed [see Prior Art Document (1) and Prior Art Document (2)]. Since the method of the above-mentioned prior art document (1) is a method based on recursive processing, there remains a problem in terms of calculation time and hardware.

On the other hand, the method of the prior art document (2) proposed by the inventors of the present application is a high-speed scanning method which does not include recursive processing and is an extension of the above prior art document (3). In these prior art documents (2) and (3), the rules for generating the Hilbert curve are tabulated in advance, and recursion is removed by sequentially referencing the tables.

[0009] Such generalization of Hilbert scanning has not been studied in the case of three dimensions due to the computational complexity. The method of the prior art document (1) is a generalization of two-dimensional Hilbert scanning, and does not describe a specific method for three-dimensional scanning. The method of the prior art (4) is based on the calculation method in the cubic region (2 ^m × 2 ^m × 2 ^m ), but does not describe the case of the rectangular parallelepiped region. For this reason, the application of Hilbert scanning in a three-dimensional space has been very limited.

However, based on the technique of the above-mentioned prior art document (2), generalization of Hilbert scanning in a three-dimensional space can be performed in the same manner as in the two-dimensional case. The present invention
In view of the above situation, as an extension of the three-dimensional Hilbert scan, an image information compression method and an image information compression system capable of performing a three-dimensional pseudo-Hilbert scan on a rectangular parallelepiped region to improve throughput and reduce cost are provided. The purpose is to provide.

[0011]

According to the present invention, in order to achieve the above object, [1] In the image information compression method, a three-dimensional pseudo-Hilbert scan corresponding to a rectangular parallelepiped region of a three-dimensional space is performed by capturing a captured moving image. Thus, sequential processing is performed using a lookup table, three-dimensional information is converted into one-dimensional series information, and the converted data is subjected to data compression.

[2] In the image information compression method according to the above [1], the division rule for each side of the recursive division of the rectangular parallelepiped block of the minimum unit in the three-dimensional pseudo-Hilbert scan is as follows. The even-length and even-length sides are divided so that they are closest to the midpoint, and the odd-length sides are divided by odd- and even-length sides so that the division point is closest to the midpoint. Addresses are allocated in a rectangular parallelepiped area.

[3] In the image information compression method according to the above [1], the data compression is such that a one-dimensional data developed by three-dimensional pseudo-Hilbert scanning is subjected to a dividing process by a threshold value. . [4] In the image information compression method according to the above [3], in the division processing, the one-dimensional data is divided into sections to minimize a square error with the one-dimensional data, and a linearly approximated compression for each section is performed. It is designed to generate data.

[5] In the image information compression system, an image pickup device for picking up a moving image, and a three-dimensional pseudo for scanning the picked-up moving image and converting three-dimensional information corresponding to a rectangular parallelepiped region into one-dimensional series information. Hilbert scanning means;
The moving image storage device includes a moving image storage device that stores dimension sequence information, and a data compression unit that compresses data read from the moving image storage device.

[6] In the image information compression system, an imaging device for capturing a moving image and a three-dimensional pseudo-Hilbert for converting three-dimensional information corresponding to a rectangular parallelepiped region for scanning the captured moving image into one-dimensional sequence information Scanning means;
A first moving image storage device for storing dimension series information, data compression means for compressing data read from the moving image storage device, a lossless encoding unit for the compressed data,
A transmission path for the compressed data from the lossless encoding unit, a lossless decoding unit for the transmitted compressed data, a data restoration unit from the lossless decoding unit, and a second moving image storing data from the data restoration unit An image storage device, three-dimensional pseudo-Hilbert scanning means corresponding to a rectangular parallelepiped region for scanning compressed data read from the moving image storage device, and a display device for displaying an output moving image are provided. .

[0016]

Embodiments of the present invention will be described below in detail. FIG. 1 is a block diagram of an image information compression system using a three-dimensional pseudo-Hilbert scanning method for a rectangular parallelepiped region of the present invention, and FIG. 2 is a schematic diagram showing a three-dimensional pseudo-Hilbert scanning (part 1) of the present invention. 5x6
Is shown. FIG. 3 is a schematic diagram showing a three-dimensional pseudo-Hilbert scan (part 2) of the present invention.
14 is shown. FIG. 4 is a schematic diagram showing a three-dimensional pseudo-Hilbert scan (part 3) of the present invention.
The case of 10 × 14 is shown.

In FIG. 1, 1 is a video camera, 2 is a moving image input unit, 3 is a cuboid filling scanning unit, 4 is a moving image memory, 5 is a control circuit, 6 is a data compression unit, 7 is a cumulative error calculation unit, 8 is a division section detection section, 9 is a lossless encoding section, 10 is a transmission path, 11 is a lossless decoding section, 12 is a data restoration section, 13 is a division section restoration section, 14 is a filter processing section, 15 is a control circuit, and 16 is a control circuit. A moving image memory, 17 a rectangular solid filling scanning unit, 1
8 is a moving image output unit, and 19 is a display.

In the image information compression system of the present invention based on three-dimensional pseudo-Hilbert scanning of a rectangular parallelepiped region, a moving image input from a video camera 1 is transmitted to a moving image input unit 2 under the overall control of a control circuit 5. The three-dimensional information is converted into one-dimensional sequence information by three-dimensional pseudo-Hilbert scanning from the cuboid filling scanning unit 3 and stored in the moving image memory 4. The one-dimensional sequence information from the moving image memory 4 is sent to the lossless encoding unit 9 by the data compression unit 6 in cooperation with the accumulated error calculation unit 7 and the divided section detection unit 8.

The compressed data from the lossless encoding unit 9 is sent to a lossless decoding unit 11 via a transmission line 10,
The data is restored in the data restoration unit 12 by the divided section restoration unit 13 and the filter processing unit 14 under the overall control by the control circuit 15, and is stored in the moving image memory 16, and is stored. The data is obtained by three-dimensional pseudo-Hilbert scanning from the cuboid filling scanning unit 17.
The image is displayed as a moving image on the display 19 via the moving image output unit 18.

The three-dimensional pseudo-Hilbert scan from the cuboid filling scanning unit 3 is performed, for example, in the case of 4 × 5 × 6 pixels as shown in FIG.
As shown in FIG. 3, in the case of the pixel 8 × 8 × 14, the operation is performed as shown in FIG. 3, and in the case of the pixel 8 × 10 × 14, the operation is performed as shown in FIG. Hereinafter, the gist of the present invention will be sequentially described in detail. First, generalization of Hilbert scanning in a three-dimensional space will be described.

(1) Hilbert Scan Filling Cubic Region Here, a Hilbert scanning method using two types of look-up tables (terminal table, induction table) for a cubic region in a three-dimensional space is described. )]]. The target of this three-dimensional Hilbert scan is 0 in the XYZ rectangular coordinate system.
≦ X <2 ^m , 0 ≦ Y <2 ^m , 0 ≦ Z <2 ^m , (m = 1,
2, ...), and this area is 2 ^m ×
Includes 2 ^m × 2 ^m grid points.

The assignment of the address of each pixel is performed as follows. FIG. 7 is a diagram showing address assignment of the eight divided areas shown in FIG. 6. As shown in FIG.
The area is divided, and an address as shown in FIG. 6B is assigned to each area after the division. Next, the same address assignment is performed for each area obtained by further dividing each area after division into eight, and each area is represented by six bits with three lower bits added.

Therefore, the size 1 × 1 after m division operations
The address Z of the × 1 area can be represented by the following 3m-bit binary representation.

[0024]

(Equation 2)

Where z_i(I = 1, 2,..., M) is 3
A binary number of bits, x_1,1... x _{m, 3}Is 0 or 1
Take the value of (For reference, address assignment in two-dimensional case
The assignment is shown in FIG. 5) For each of the eight divided areas in FIG. 6, each area is represented by one representative point.
Consider how to follow when you do. In this case, the 8 starting points
In contrast, there are three types of end points, and the start point and end point are switched
12 (= 8 × 3/2) ways if you omit the combinations
Can be considered as a combination of the start point and the end point.

Therefore, there are 12 types of Hilbert curves used in the present invention, as shown in FIG. Here, each alphabet represents the type of curve. The method of tracing each divided area into eight divided areas is also described in this 1
One of two types of curves A to M. For example, in the case of the curve A, the order of tracing the eight regions shown in FIG. 6A is 000 → 010 → 011 → 001 → 101 → 111 → 1
10 → 100, and the type of the Hilbert curve in each region changes in the order of B → C → C → J → J → I → I → E as shown in FIG.

Therefore, for each type of curve and the order of tracing, the two
Type information, that is, address information and curve type information, are stored in a terminal table T _trm [γ _j ] [ _ij ] and an induction table T _ind [γ _j ] [ _ij ], respectively. Here, γ _j (= A, B,..., K, M) is the type of the curve at the time of j-times division, and i _j (= 1, 2,. It is order. For example, γ ₀
= A for the region where the curve of A scans eighth: z ₁ = T
_trm [A] [8] = 100, γ ₁ = T _ind [A] [8]
= E.

The following _shows a terminal table (T _trm ) for address reference and an induction table (T _ind ) for curve type reference. T _trm [A] = [000,010,011,001,1
01,111,110,100], T _trm [B] = [000,001,101,100,1
10,111,011,010], T _trm [C] = [000,100,110,010,0
11, 111, 101, 001], T _trm [D] = [110, 100, 101, 111, 0]
11,001,010,010], T _trm [E] = [ _{110,111,011,010,0}
00,001,101,100], T _trm [F] = [110,010,000,100,1
01,001,011,111], T _trm [G] = [ _{101,111,110,100,0}
00,010,011,001], T _trm [H] = [101,100,000,001,0
11,010,110,111], T _trm [I] = [101,001,011,111,1
10,010,000,100], T _trm [J] = [011,0011,000,010,1
10, 100, ₁₀₁ , ₁₁₁ ], T _trm [K] = [011, 010, 110, 111, 1
01,100,000,001], T _trm [M] = [011,111,101,001,0
00, 100, 110, 010]. _Tind [A] = [B, C, C, J, J, I, I, E], _Tind [B] = [C, A, A, H, H, D, D, M], _Tind [C] = [A, B, B, F, F, K, K, G], T _ind [D] = [E, F, F, G, G, M, M, B], T _ind [E ] = [F, D, D, K, K, A, A, I], T _ind [F] = [D, E, E, C, C, H, H, J], T _ind [G] = [H, I, I, D, D, C, C, K], _Tind [H] = [I, G, G, B, B, J, J, F], _Tind [I] = [G , H, H, M, M, E, E, A], T _ind [J] = [K, M, M, A, A, F, F, H], T _ind [K] = [M, J , J, E, E, G, G, C], T _ind [M] = [J, K, K, I, I, B, B, D]. The specific algorithm is as follows:
_Are removed, _{iM + 1 is set} to ₁ to 8, and the address reference table is replaced with _PTtrm [p] by _Ttrm .

(2) Hilbert Scan Filling a Rectangular Area The present invention extends the above-described three-dimensional Hilbert scan for a cube (2 ^m × 2 ^m × 2 ^m ) to a scanning method capable of handling a rectangular parallelepiped area. I do. Here, the case where the Hilbert curve is used for scanning the rectangular parallelepiped region is referred to as pseudo-Hilbert scanning. Normally, the Hilbert scanning is performed on a cubic region, and its applicable range is limited. However, in the present invention, a three-dimensional pseudo-Hilbert scanning method for a rectangular parallelepiped region is employed, and this method uses a lookup table. Using the sequential processing method described above. For this reason, there are many advantages over a calculation method based on recursive processing in terms of calculation time and hardware.

The extension from the cube to the rectangular parallelepiped consists of two points, that is, a change in the rule for dividing edges in recursive division into eight and a change in the assignment of the least significant three bits of the address. In the recursive division into eight, each side follows the following division rule. (1) An even-length side is divided by an even-length side and an even-length side so that the division point is closest to the midpoint.

(2) The odd-length side is divided by odd-length and even-length sides so that the division point is closest to the middle point. According to this division rule, a line segment having a length a · 2 ^m is m times 2
The division is performed into 2 ^m line segments having a length a. From this, if the line segment length l is 2 × 2 ^m ≦ l <4 × 2 ^m , as a result of m divisions into two, l is the line segment length 2, 3, 4
Are divided into 2 ^m segments. Therefore,
The three side lengths l _x , l _y , l _z are 2 × 2 ^m ≦ l _x <4 × 2 ^m , 2 × 2 ^m ≦ l _y <4 × 2 ^m , 2 × 2 ^m ≦ l _z <4 × 2 ^{In a} rectangular parallelepiped region satisfying ^m ,... (1), a rectangular parallelepiped having a combination of side lengths of 2, 3, and 4 is generated as a result of m divisions into eight (hereinafter, a rectangular parallelepiped of these minimum units is referred to as a block). ).

In the case of a rectangular parallelepiped satisfying the above equation (1), the scanning conditions are as follows: (A) The starting point and the ending point are on the same side in the block. (B) A hamming distance between addresses moves to a lattice point of 1. Considering that it is the combination of side length satisfying this condition block is expressed by _{(l x × l y × l} z), (2
× 2 × 2), (2 × 3 × 2), (2 × 2 × 4), (2 ×
4 × 2), (4 × 2 × 2), (2 × 3 × 4), (4 × 3
× 2), (2 × 4 × 4), (4 × 2 × 4), (4 × 4 ×
2), (4 × 3 × 4) and (4 × 4 × 4).

On the other hand, in the case of the cube (2 ^m × 2 ^m × 2 ^m ), only blocks (2 × 2 × 2) generated by the division were obtained. As described above, in the case of the rectangular parallelepiped, since the block resulting from the division is different from the case of the cube, the assignment of the least significant bit of the address needs to be changed. Therefore, the address is stored in the terminal table PT for each block.
_trm [p] [γ] [i] [p (= 1, 2, ..., 12) is a number assigned to the block. γ and i are T _trm
Is the same as in the above case), for example, (2 × 2 × 4)
For the block _No. , PT _trm [3] [A] [16]
= 1000.

Here, 3 which is the first argument of the table is a number indicating a (2 × 2 × 4) block. Therefore,
For a rectangular parallelepiped satisfying the above expression (1), the same eight-division scanning as in the case of the cube is performed m times, and thereafter, by referring to the address from the terminal table PT _trm corresponding to each block, the address allocation in the rectangular parallelepiped region is performed. Becomes possible.

Based on this idea, the above equation (1) is satisfied.
Rectangular parallelepiped region (l_x ⁰× l_y ⁰× l _z ⁰)
The algorithm for calculating the dress non-recursively is as follows:
Swell. Where l_x ^k× l_y ^k× l_z ^kIs the k-th
This is the length of each side after division. Next, about algorithm 1
To explain, Hilbert-Address (γ₁, L_x ⁰, L_y
⁰, L_z ⁰, M) for i₁= 1,2, ..., 8 z₁= T_trm[Γ₁] [I₁]; Γ_Two= T_ind[Γ₁]
[I₁]; ｛L_x ⁰, L_y ⁰, L_z ⁰Area is divided according to the division rule
Then z₁Length l of the divided area corresponding to_x ¹, L_y ¹, L
_z ¹・｝ For i_m= 1,2, ..., 8 z_m= T_trm[Γ_m] [I_m]; Γ_{m + 1}= T
_ind[Γ_m] [I_m]; ｛L_x ^m-1, L_y ^m-1, L_z ^m-1Area according to the division rule
Divided, z_mLength l of the divided area corresponding to_x ^m, L_y
^m, L_z ^m｝; ｛L_x ^m, L_y ^m, L_z ^mBlock (p) for
｝; For i_{m + 1}= 1, ..., l_x ^m× l_y ^m× l_z ^m z_{m + 1}= PT_trm[P] [γ_{m + 1}] [I_{m + 1}]; Output <z₁... z_mz_{m + 1}> Where l is the total number of grid points in the scanning area_x ⁰× l_y ⁰× l
_z ⁰= LN, the amount of calculation of this algorithm is LN
And only O (LN).

On the other hand, the generalization of Hilbert scanning by recursive processing [prior art document (1)] does not show a specific method for the three-dimensional case, so that comparison cannot be performed. However, as in the case of the two-dimensional case, the processing becomes complicated including recursive processing, so that the calculation time is much longer than that of the method of the present invention. Also, there is a one-to-one correspondence between the order in a one-dimensional sequence obtained by Hilbert scanning and the address of a lattice point in a three-dimensional space, and one can guide the other. The specific method is the same as that of the above-mentioned prior art document (2), and is omitted here.

As described above, the constraints in the method of the present invention are as follows. (1) (minimum side length) ≦ (maximum side length) ≦ 2 × (minimum side length) (2) Length of side where start point and end point exist, and modulus of surface not including start point and end point The length of the side in the line direction is an odd length.

For a size outside the above range (1), almost all the range can be covered by arranging the rectangular parallelepipeds satisfying the above (1) in series. Regarding the above (2), a problem that needs to be studied in the future in order to expand to an arbitrary size without restriction can be dealt with by adding one-line scanning to the end of the image.

The method of the present invention uses a look-up table to eliminate recursive processing. Table 1 summarizes the storage capacity of this table.

[0040]

[Table 1]

The table required for the cubic filling scan is described in the above-mentioned "Hilbert scan for filling the cubic region".
Are two types (terminal table and induction table). In the case of three-dimensional pseudo-Hilbert scanning, it was necessary to add a table of address information for each block. However, if the symmetry of the curve shown in FIG. 7 is used, the storage capacity of this table can be significantly reduced. Can be. It can be seen from Table 1 that when the symmetry is used, the storage capacity can be increased by about 1/5. Thus, the storage capacity of the table can be made extremely compact also in the case of a rectangular parallelepiped.

Next, the information compression method of the present invention will be described. In a moving image, if there are images of X × Y pixels for Z frames, the image becomes a X × Y × Z rectangular parallelepiped when viewed three-dimensionally. Here, a high-speed and high-quality moving image compression method using three-dimensional pseudo-Hilbert scanning by the rectangular parallelepiped filling scanning unit 6 is realized. In this method, a one-dimensional data developed by X-Y-Z three-dimensional pseudo-Hilbert scanning is subjected to a dividing process using a threshold.

Hereinafter, this division processing will be described.

[0044]

(Equation 3)

The section up to the (N-1) th pixel in which e _i exceeds a predetermined threshold value THRE is defined as one section. This process is 1
It is performed in order from the top of the dimension data. The compression algorithm is as follows. Here, the subprogram GROPI
NG determines the division point based on the evaluation value. Here, SIZE is the total number of pixels, and THRE is a threshold value for the evaluation scale e _i . [Compression algorithm]

[0046]

(Equation 4)

When the compressed data is restored, the processing of embedding the information of {l _k } and {c _k } in the moving image regarded as a rectangular parallelepiped along the three-dimensional pseudo-Hilbert scanning of the rectangular parallelepiped filling scanning unit 17 is sufficient. . The algorithm for restoring moving image data is as follows. Here, the filter processing in the filter processing unit 14 is a conventional method, for example,
Use a King filter or the like. [Restoration Algorithm] (1) Decode the main {{l _k } and {c _k } (k = 1, 2,...,
K); {c _k } is developed with a length of {l _k } along the three-dimensional pseudo-Hilbert scan of the cuboid filling scanning unit 17; filter processing is performed. {Divided section detection unit of the data compression unit 6} 8 will be described with reference to FIG.

It is assumed that THRE = 10. First, in order to determine the first section, the cumulative error is examined sequentially from point number 1. (1) Since x ₁ = 5, when j = 1, that is, x ₂
Calculating the mean x and the cumulative error e including the point of = 7,
x = 6 and e = 2. At this time, since e <THRE, the next point is checked.

(2) When j = 2, that is, when the average x and the accumulated error e are calculated including the point of x ₃ = 9, x = 7,
e = 8. At this time, since e <THRE, the next point is checked. (3) When j = 3, that is, including the point of x ₄ = 7,
When the average x and the accumulated error e are calculated, x = 7 and e = 8. At this time, since e <THRE, the next point is checked.

(4) When j = 4, that is, when the average x and the accumulated error e are calculated including the point of x ₅ = 2, x =
6, e = 28. At this time, since it is e> THRE x ₁ and x ₂ and x ₃ and x ₄ are the same interval (N = 4).
Since the first section has been determined, the second section is obtained from point number 5 next. (1) Since x ₅ = 2, when j = 1, that is, x ₆
When the average x and the accumulated error e are calculated including the point of = 4,
x = 3 and e = 2. At this time, since e <THRE, the next point is checked.

(2) When j = 2, that is, when the average x and the accumulated error e are calculated including the point of x ₇ = 3, x = 3
e = 2. At this time, since e <THRE, the next point is checked. (3) Similarly, the average x for j = 3, 4,.
And THRE while calculating the accumulated error e, e>
If it is THRE, the section up to that point is the same section.
x ₅ and x ₆ and x ₇ and x ₈ are the same segment.

Thus, the second section can be determined.
The above processing is performed up to the data size SIZE. Next, a description will be given of high image quality utilizing discrete cosine transform, vector quantization, and the like. In FIG. 9, the difference between the original one-dimensional data and the average value of each section is determined.

This is approximated error data {y _i , i = 1, 2,..., SIZE} as shown in FIG. For this data, compression efficiency is increased by using a conventional method such as one-dimensional discrete cosine transform or vector quantization. This makes it possible to improve the quality of a moving image. This conventional method is incorporated in the output part of the data compression part 6 and the input part of the data decompression part 12.

According to the three-dimensional pseudo-Hilbert scanning method for the rectangular parallelepiped region of the present invention as described above, it is possible to improve the efficiency of image information compression as shown in FIG. Here, the image is a Miss Ameri
ca (360 × 288 pixels / frame), compression ratio = 0.
19 bits / pixel (only luminance). That is, FIG.
In FIG. 1, the horizontal axis indicates the frame number, and the vertical axis indicates the processing time (second). Graph a is the case of the image information compression according to the present invention, and graph b is the case of the conventional image information compression (MPEG: Moving). Picture Expe
rtsGroup: Moving Picture I
image Coding Experts Grou
p: International standard method of moving image compression).

As is apparent from this figure, in the case of the present invention, the processing time is about 2.5 seconds, and in the conventional case, the processing time is about 28 seconds, and the processing time is reduced by about 10 times. Can be. It should be noted that the present invention is not limited to the above embodiment, and various modifications can be made based on the gist of the present invention, and these are not excluded from the scope of the present invention.

[0056]

As described above, according to the present invention, the following effects can be obtained. By using a look-up table and a sequential processing method by a three-dimensional pseudo-Hilbert scan corresponding to a rectangular parallelepiped region in a three-dimensional space, a reduction in throughput and an increase in cost in the case of general recursive calculation are performed. Can be avoided.

In particular, in moving image processing, a high-speed and high-quality moving image compression technique can be provided at low cost. Also, the moving image information has an enormous amount of information, and the enormous amount of information is converted into a one-dimensional series with good neighborhood preservation by an accurate three-dimensional cuboid filling scan. Moving image compression.

Further, the redundancy of the moving image can be accurately removed. Thus, it is possible to establish the compression processing of video information for cuboid region of the three-dimensional space more practical, in particular, its short compression-restoration process, the conversion process utilizing hierarchy In addition, compression parameters with little image dependency can be used, which greatly contributes to the development of digital image compression technology.

[Brief description of the drawings]

FIG. 1 is a block diagram of an image information compression system based on a three-dimensional pseudo-Hilbert scanning method for a rectangular parallelepiped region according to the present invention.

FIG. 2 is a schematic diagram illustrating pseudo-Hilbert scanning of a three-dimensional rectangular parallelepiped (part 1) according to the present invention.

FIG. 3 is a schematic diagram showing pseudo-Hilbert scanning of a three-dimensional rectangular parallelepiped (part 2) of the present invention.

FIG. 4 is a schematic diagram showing pseudo-Hilbert scanning of a three-dimensional rectangular parallelepiped (part 3) of the present invention.

FIG. 5 is a diagram showing an example of a conventional two-dimensional Hilbert curve.

FIG. 6 is a diagram showing address assignment of an eight-part area.

FIG. 7 is a diagram illustrating a Hilbert curve in a three-dimensional space.

FIG. 8 is a diagram showing a scanning pattern of a divided area.

FIG. 9 is an explanatory diagram of divided section detection by the data compression unit.

FIG. 10 is a diagram showing approximate error data.

FIG. 11 is a diagram showing processing time characteristics of the present invention and a conventional processing time.

[Explanation of symbols]

1 the video camera 2 moving image input unit 3, 17 rectangular solid filling scan portion 4,16 moving image memory 5 and 15 control circuit 6 the data compression unit 7 accumulated error calculating unit 8 divided section detection unit 9 reversible sign-unit 10 the transmission line 11 reversible Decoding unit 12 data restoration unit 13 divided section restoration unit 14 filter processing unit 18 moving image output unit 19 display

Claims (6)

(57) [Claims] 1. A method for compressing image information, comprising sequentially processing a captured moving image by a three-dimensional pseudo-Hilbert scan corresponding to a rectangular parallelepiped region in a three-dimensional space using a look-up table to convert the three-dimensional information into one-dimensional data. A method for compressing image information, comprising converting the data into sequence information and compressing the converted data. 2. The image information compression method according to claim 1, wherein the dividing rule of each side of the recursive division of the rectangular parallelepiped block of the minimum unit in the three-dimensional pseudo-Hilbert scanning is such that a dividing point of an even-length side is a midpoint. Is divided by the even-length and even-length sides so as to be closest to, and the odd-length side is divided by the odd-length and even-length sides so that the division point is closest to the midpoint,
An image information compression method, wherein an address is assigned in a rectangular parallelepiped area corresponding to the method. 3. The image information compression method according to claim 1, wherein the data compression is performed by a threshold process on the one-dimensional data expanded by three-dimensional pseudo-Hilbert scanning. Method. 4. The image information compression method according to claim 3, wherein the dividing process is performed by dividing a section of the one-dimensional data that minimizes a square error with the one-dimensional data, and performing linear approximation for each section. An image information compression method comprising generating compressed data. 5. In an image information compression system, (a)
An imaging device that captures a moving image; and (b) three-dimensional pseudo-Hilbert scanning means that scans the captured moving image and converts three-dimensional information corresponding to a rectangular parallelepiped region into one-dimensional sequence information;
An image information compression system comprising: (c) a moving image storage device that stores the one-dimensional sequence information; and (d) a data compression unit that compresses data read from the moving image storage device. 6. In an image information compression system, (a)
An imaging device for capturing a moving image; and (b) three-dimensional pseudo-Hilbert scanning means for converting three-dimensional information corresponding to a rectangular parallelepiped region for scanning the captured moving image into one-dimensional sequence information;
(C) a first moving image storage device for storing the one-dimensional sequence information, (d) data compression means for compressing data read from the moving image storage device, and (e) reversibility of the compressed data. An encoding unit, (f) a transmission path for compressed data from the lossless encoding unit, (g) a lossless decoding unit for the transmitted compressed data, and (h) a data restoring unit from the lossless decoding unit.
(I) a second moving image storage device for storing data from the data restoring unit; and (j) three-dimensional pseudo-Hilbert scanning means corresponding to a rectangular parallelepiped region for scanning compressed data read from the moving image storage device. , (K) a display device for displaying a moving image to be output.