JPH0775109A - Image information compressing method - Google Patents

Abstract

PURPOSE:To reduce the code amount of a high band signal in the case of sub band encoding by using an output value at the time of being multiplied to the signal value of an integral picture element around any prescribed filter as the signal value of a 1/2 picture element. CONSTITUTION:Sub band division is respectively performed into the LL signal of a vertival low band at a horizontal low band, the HL signal of the vertical low band at a horizontal high band, the LH signal of a vertical high band at the horizontal low band, and the HH signal of the vertical high band at the horizontal high band. Next, a present frame is divided into macro blocks composed of (m) pieces of horizontal picture elements X (n) pieces of vertical picture elements for each of respective LL, HL, LH and HH band signals. Motion compensating prediction is performed between the same band signals of the present frame and a reference frame. In order to perform the motion compensating prediction with 1/2 picture element precision the 1/2 picture element value is calculated from the integral picture element value concerning the motion compensated block and the peirpheral blocks. While including the block composed of this 1/2 picture element value, the final motion compensated block and motion vector are selected.

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 method for compressing moving image information of a high definition television (hereinafter referred to as "HD-TV"), and more specifically, it relates to an image information compression method. The present invention relates to an improvement of an image information compression method suitable when performing compression processing by sub-band division.

[0002]

2. Description of the Related Art For digital transmission of HD-TV image information and its recording, a study of high-efficiency coding having a hierarchical property is being made (Tsunashima et al., "HD-TV hierarchy considering interlaced structure". On encoding ", 199
3rd year, IEICE Spring Conference, D-253).
This is performed by extracting a part of the bit stream in which the image information of HD-TV is encoded and decoding it to obtain S
The image information of D-TV is obtained.
According to this, use of subband coding is considered as one of the coding methods.

For example, an HD- having an interlaced structure
Consider hierarchical coding of a TV image signal and an SD-TV image signal having an interlaced structure as an example. First, HD-
The image signal of the TV1 field is divided into two band signals of a low band and a high band in the horizontal direction. Next, the band signals are divided into two band signals, a low band and a high band in the vertical direction. FIG. 6 shows the situation, and HD-TV
The image signal is subband-divided into a total of four band signals LL, HL, LH, and HH. Then, the band signal LL having a low band in both the horizontal and vertical directions is used as the video signal of the SD-TV1 field. Then, two SD-TV1 field signals are collected to form an SD-TV1 frame video signal.

By the way, in encoding moving image information such as a television image signal, it is known that the code amount compression efficiency is improved by utilizing the correlation between frames. MPEG (Motion Picture Image Coding Expert)
s Group) and MPEG2 also perform coding with high compression efficiency by combining interframe prediction and DCT (Watanabe, “MPEG2 interframe prediction method”, Technical Report of the Institute of Television Engineers, ITEJ Technical Report Vol.16, N).
o.61, pp.37-42, ICS'92-73 (Oct.1992)).

From this point of view, a method of compressing the HD-TV video signal divided into sub-bands as shown in FIG. 6 by utilizing the correlation between frames can be considered. That is, the horizontal and vertical low band signals LL (or 2
For signals that gathered fields into one frame),
Interframe prediction may be performed in the same manner as the conventional methods such as MPEG and MPEG2. That is, the picture type of the frame is I picture (intra-frame coded screen), P
It is divided into a picture (inter-frame coding screen) and a B picture (bidirectional prediction / interpolation screen), and the P-picture and B-picture are subjected to motion compensation prediction coding with the I-picture or P-picture which is the reference frame.

Specifically, the current frame (P or B picture) is divided into macroblocks of horizontal m pixels × vertical n pixels. On the other hand, an approximate image in the reference frame, that is, a motion-compensated block of horizontal m pixels × vertical n pixels is selected. Then, either the difference value for each pixel of both the macroblock of the current frame and the block of the reference frame or the original pixel value of the macroblock in the current frame is adaptively selected for each block.

That is, when there is no difference between the two images in the block, the difference value is selected, when there is a difference, the original pixel value is selected, and the selected difference value or pixel value is encoded. Also, as the motion compensation with 1/2 pixel accuracy, a difference may be taken with respect to the interpolated value of the pixel of the reference frame. It is assumed that the reference frame is also subband-divided as shown in FIG. 6 and the block of horizontal m pixels × vertical n pixels is selected from the LL signals in the horizontal and vertical low bands.

Other bands HL, LH, which are sub-band divided,
Similarly for HH, motion compensation interframe prediction is performed. Similarly, the reference frame is subband-divided, and a block of horizontal m pixels × vertical n pixels at a predetermined position is selected from the corresponding HL, LH, and HH band signals to obtain the difference value.

[0009]

By the way, when such motion-compensated inter-frame prediction is performed, the code amount of the current frame can be compressed with high efficiency for the horizontal and vertical low band LL signals. This can also be explained by the fact that the statistical properties of the LL signal are similar to the statistical properties of the original image signal before subband division (Takishima et al., "Image quality evaluation method and statistics of subband coding". On the physical properties ", IEICE Technical Report IE91-4).

However, HL, LH, H of bands other than LL
For the H signal, the code amount of the current frame does not decrease so much compared to before the motion-compensated interframe prediction. As a result, the code amount of the entire image of LL to HH increases, and the transmission at the time of transmitting and recording the image information of HD-TV,
The recording capacity becomes large. Further, when the code amount of the entire image is limited to a certain fixed amount, it is necessary to reduce the code amount by requantization, and as a result, the image quality at the time of decoding deteriorates.

The present invention focuses on these points,
It is an object of the present invention to provide an image information compression method capable of reducing the code amount of a high band signal in sub-band coding such as an HD-TV signal.

[0012]

To achieve the above object, the first invention is to subdivide an image signal of an image of a predetermined unit into a low band signal and a high band signal, and for each band signal. When performing inter-frame coding by motion-compensated prediction coding, at least for high-band signals, in the image information compression method for performing motion-compensated prediction with high accuracy using 1/2 pixel between integer pixels, As the signal value of 2 pixels,
It is characterized in that an output value obtained by applying a filter whose sum of filter coefficients is negative to signal values of peripheral integer pixels is used.

According to a second aspect of the present invention, when an image signal of an image of a predetermined unit is subband-divided into a low band signal and a high band signal and inter-frame coding is performed by motion compensation predictive coding for each band signal. For at least a high-bandwidth signal, in the image information compression method for performing highly accurate motion-compensated prediction using 1/2 pixel between integer pixels and 1/4 pixel or more pixels, As a signal value of a pixel, an output value obtained by applying a filter having a negative sum of filter coefficients to a signal value of a peripheral integer pixel is used as a signal value of a pixel having accuracy of ¼ pixel or more. It is characterized in that an output value obtained by applying a filter having a positive sum of coefficients to signal values of peripheral integer pixels and half pixels is used.

A third invention is characterized in that, in the above invention, a motion vector of the high band signal is encoded by a difference vector from a vector corresponding to the motion vector of the low band signal.

[0015]

According to the present invention, at least for high-band signals, when performing motion compensation prediction with 1/2 pixel accuracy or 1/4 pixel accuracy or more, a filter whose sum of filter coefficients is negative can be used as a peripheral filter. The output value when multiplied by the signal value of the integer pixel is used as the signal value of the 1/2 pixel. For high-band signals, signal values obtained by a filter having a negative sum of filter coefficients are used, rather than using a conventional technique without subband division or an interpolated value such as LL signals obtained by subband division. In this case, as described in the embodiment below, it is possible to perform encoding with a smaller amount of information.

[0016]

Embodiments of the image information compression method according to the present invention will be described below in detail with reference to the accompanying drawings.

<First Embodiment> First, a first embodiment of the present invention will be described with reference to FIGS. FIG. 1 shows a processing procedure of the first embodiment as a flowchart. The processing shown in FIG. 1 is executed in a series of image information compression processing. In the figure, the image signals of the current frame and the reference frame to be compressed are, as shown in FIG. 6, the LL signal of the horizontal low band and the vertical low band, and the horizontal high band of the vertical low band HL signal. The signal is sub-band-divided into an LH signal having a low band in the horizontal direction and a high band in the vertical direction, and an HH signal having a high band in the horizontal direction and a high band in the vertical direction (step S1).

Next, for each band signal of LL, HL, LH and HH, the current frame is divided into macroblocks of horizontal m pixels × vertical n pixels (step S2). Then, motion compensation prediction is performed between the same band signals of the current frame and the reference frame.

[In case of LL band signal (Y in step S3
In this case), first, a block of horizontal m pixels × vertical n pixels of an image (motion compensation image) approximated from the LL signal of the reference frame in macroblock units of the LL band signal of the current frame is selected (step S4). Then, the pixel value (integer pixel value) of the macroblock of the current frame,
The pixel values (integer pixel values) of the motion compensation blocks are compared, and the positional shift between the two blocks is used as the motion vector.

Next, in order to make the motion-compensated prediction accurate to 1/2 pixel, the 1/2 pixel value is obtained from the integer pixel value for the selected motion compensation block and its periphery (step S5). The 1/2 pixel value is obtained by interpolation from the peripheral integer pixel values. The final motion compensation block and motion vector including the block consisting of this 1/2 pixel value are selected (step S6).

Next, both blocks are compared, and any one of the difference value for each pixel value of both blocks and the original pixel value of the macroblock of the current frame is adaptively selected (step S7).

Then, any one of the selected values is encoded as compression information of the macroblock (steps S8 and S9). Specifically, (a) When the difference value is selected, the macro block is set as an inter block (inter-frame coded block). Then, the code representing the inter block, the code representing the difference value for each pixel, and the code representing the motion vector representing the amount of deviation between the position of the macro block and the position of the block in the reference frame are defined by the macro. Output as the compression code of the block (step S
8).

(B) When the original pixel value is selected, the macro block is defined as an intra block (intra-frame coded block). Then, the code representing the intra block and the code representing the original pixel value are output as the compressed code of the macro block (step S9).

[In the case of band signals other than LL (step S
3 N)] Next, HL, LH, H other than the LL signal
For the H band signal, the motion vector obtained in the macro block of the LL signal is used in macro block units,
A motion compensation block is selected from the reference frame (step S10). Here, of the motion vector obtained from the LL signal, only a portion in units of integer pixels is used. Then, in order to make the motion compensation prediction with 1/2 pixel accuracy, the 1/2 pixel value is obtained from the integer pixel value for the selected motion compensation block and its periphery (step S11).

However, in this case, the interpolated value used in the LL signal is not used. Instead, a signal value obtained by applying a digital filter having a negative sum of filter coefficients to peripheral integer pixel values is used as the 1/2 pixel value (step S11). Then, in addition to the integer pixel value, the 1/2 pixel value is also used to select the motion compensation block and detect the motion vector (step S1).
2). After that, the processes of steps S7 to S9 are performed.

As described above, in this embodiment, the integer pixel value and the 1/2 pixel value obtained by the interpolation calculation are used for the LL signal as in the prior art, but the integer pixel value and the 1/2 pixel value for the high band signal are used. It is characterized in that the motion compensation is performed by using a 1/2 pixel value which is a digital filter calculation value for which the sum of the filter coefficients is negative. For other parts, well-known techniques such as MPEG and MPEG2 may be appropriately combined as needed.

Next, the above embodiment will be described more specifically with reference to FIGS. For easy understanding, a horizontal one-dimensional image model as shown in FIG. 2A is used. In FIG. 2, the horizontal axis represents the horizontal position Y of the pixel, and the vertical axis represents the signal level L of the pixel. Further, the vertical line shows the digitized pixel signal (sampling value), and the dotted line shows the corresponding analog signal.
When the image signal is a luminance signal, an 8-bit positive value (0 to 2
Although it may be digitized to take 55), in such a case, the central value 128 is subtracted from the figure.

When such an image signal is separated into a low band signal and a high band signal, they become as shown in FIGS. 2 (C) and 2 (E), respectively. In both cases, the sampling frequency is 1/2 that of the signal of FIG. The low band signal corresponds to, for example, the LL signal in FIG. 6, and the high band signal is HL.
Corresponding to the signal.

On the other hand, it is assumed that the reference frame is an image slightly moved with respect to the current frame, as shown in FIG. 2 (B). When the image signal of the reference frame shown in FIG. 2 (B) is similarly separated into a low band signal and a high band signal, they are respectively shown in FIGS. 2 (D) and (F).

First, regarding the low band signal, as shown in FIG.
The motion compensation block and the motion vector are obtained by comparing (C) and (D) to obtain an approximate portion. For example,
Assuming that the macroblock of the current frame has four pixels indicated by WA in FIG. 2C, the motion compensation block is obtained as WB, and the motion vector having a size of Δ in the left direction is obtained. That is, 4 included in the motion compensation block WB
Encoding processing by motion compensation prediction is performed using pixels.

At this time, as in the prior art, if the difference value between each pixel value of the macro block WA of the current frame and each pixel value of the motion compensation block of the reference frame is encoded, a sufficiently good code is obtained. Is possible. For example, focusing on the pixel PA, the level is LA and the level of the corresponding pixel PB of the motion compensation block is LB, so the difference value is LA-LB, and this is encoded.

Next, for the high band signal, FIG.
2E corresponding to the macroblock WA of FIG. 2E, the motion compensation block is obtained by using the motion vector Δ obtained by the low band signal. In FIG. 2F, which is a high-band signal of the reference frame, the block corresponding to the motion vector Δ is 4 pixels shown by WB in the figure.

Further, in this embodiment, as shown in steps S11 and S12 in FIG. 1, the motion vector search is performed with 1/2 pixel precision around the motion compensation block WB. This operation will be described with reference to FIG. In addition, in FIGS. 3 (A) and 3 (B), FIG.
(E) and (F) are shown in enlarged form, respectively.
(C) is obtained by adding 1/2 pixel to each integer pixel in FIG. 2 (F).

The blocks WA and WB are shown in FIG.
As shown in FIG. The four pixel levels included in the macroblock WA of the current frame shown in FIG. 3A are converted into integer pixel motion compensation blocks W shown in FIG. 3B.
Compared with the pixel level included in B, the positive and negative levels are reversed in three.

As described above, in the high band signal, the positive and negative of the signal level may be reversed depending on the pixel. this is,
In the high-band signals of FIGS. 3A and 3B, FIG.
This is because the sampling frequency is 1/2 that of the signal before subband division in (B). For example, FIG. 3 (A)
The pixel PC of the level LC in the current frame of the above corresponds to the pixel P3 of the level L3 in the reference frame of FIG. When these levels are compared, LC≈−L3 and the difference value becomes L3−LC≈L3 + L3.

This cannot be solved even if the 1/2 pixel value is obtained by the interpolation operation as in the prior art. For example, if the 1/2 pixel value between the pixels P2 and P3 is interpolated, it is approximately equal to L3, and if the 1/2 pixel value between the pixels P3 and P4 is interpolated, it becomes approximately 0, and Level L
It cannot be a valid predictor of C. For this reason, even if motion-compensated prediction is performed on a high-band signal in the conventional manner and the difference value is encoded, the information amount is not compressed so much.

Therefore, in this embodiment, the pixel value of the intermediate value of the integer pixels, that is, the 1/2 pixel value is obtained for the motion compensation block WB of the reference frame and its periphery.
At this time, the interpolation value (average value of adjacent integer pixel values) as in the related art is not used as the 1/2 pixel value. Instead, a value obtained by applying a digital filter having a negative sum of coefficients to integer pixels is used.

More specifically, x is a sequence of integer pixel signals.
When expressed by (k), as a 1/2 pixel value between x (k) and x (k + 1), (1/8) · x (k−1) + (− 5/8) · x (k ) + (− 5/8) · x (k + 1) + (1/8) · x (k + 2) …………………… (1) is used.

In FIG. 3C, the 1/2 pixel values PQ to PT obtained for the motion compensation block WB and its surroundings are x.
Indicated by dots. Note that the integer pixel values P1 to P7 are indicated by a circle. Then, the signal value of each of these pixels is
LQ to LT and L1 to L7, respectively.

For example, the signal value LQ of the 1/2 pixel PQ is
x (k-1) = L1, x (k) = L2, x (k + 1) =
Since L3, x (k + 2) = L8, substituting them into the equation (1), LQ = (1/8) .L1 + (-5/8) .L2 + (-5/8) .L3 + (1/8) ・ L4 …………………… (2).

Similarly, the signal value LR of the 1/2 pixel PR
Is LR = (1/8) .L2 + (-5/8) .L3 + (-5/8) .L4 + (1/8) .L5 .................... (3)

The signal value LS of the 1/2 pixel PS is LS = (1/8) .L3 + (-5/8) .L4 + (-5/8) .L5 + (1/8) .L6. ………………… (4)

The signal value LT of the 1/2 pixel PT is LT = (1/8) .L4 + (-5/8) .L5 + (-5/8) .L6 + (1/8) .L7 ... ………………… (5) In this way, from the motion compensation block WB of the reference frame and the integer pixel values L1 to L7 around it, 1
/ 2 pixel values LQ to LT, ... Are obtained.

Next, the integer pixel value and the 1/2 pixel value of the macro block WA of the current frame and the motion compensation block WB are compared to obtain the motion compensation block and the motion vector of 1/2 pixel precision. Macroblock W of FIG.
When the pixel pattern of A is compared with the pixel pattern of ½ pixel value indicated by X in FIG. 7C, the portion indicated by WC is the closest, so this is designated as a motion compensation block. Further, the motion vector is ΔC based on the moving direction and the moving amount of the macro block WA with respect to the motion compensation block WC.
In this way, the motion compensation block and the motion vector with 1/2 pixel accuracy are obtained respectively.

Next, the macroblock WA of the current frame
Then, the difference value for each pixel value of both blocks of the motion compensation block WC is obtained. For example, for the pixel PC in the WA, the difference value LC-L from the pixel PQ in the WC
Q is encoded.

Next, the filter shown in the equation (1) will be further described. Reexpressing the equation (1) by Z1 ^-1 representing a delay of one unit time, (1/8) + (-5/8) .Z1 ^-1 + (-5/8) .Z1 ^-2 + (1/8) ・ Z1 ^-3 …………………… (6).

This is the sampling frequency f of the image signal.
with return to the previous sub-band dividing the s, the unit of the delay time as a half Z2 ^-1 = Z ^-1 / 2 of ^{Z -1, {(1/8) +} (- 5/8) · Z2 -2 + Z2 ^-3 + (-5/8) ・ Z2 ^-4 + (1/8) ・ Z2 ^-6 } / 2 …………………… (7), 0-fs / 4 is blocked Area, fs / 4 to fs /
In order to be close to the characteristics of the ideal HPF with 2 as the pass band,
The coefficient is decided.

In other words, 0 to fs / 4 is the stop band, fs /
A filter coefficient {a- _{(2i + 1)} · Z- ^{(2i + 1)} + a- _(2i-1)・ which is determined so as to have a frequency characteristic close to that of an ideal HPF having a pass band of 4 to fs / 2. Z- ^(2i-1) ＋ …… ＋ a _-1・ Z ^-1 +1 + a ₁・ Z ¹ ＋ …… ＋ a _(2i-1)・ Z ^(2i-1) + a _{(2i + 1)}・ Z ^{(2i + 1)}} / 2 ........................ in _{(8) a - (2i +} 1), a - (2i-1), ......, a -1, a 1, ......,
This means that the signal value of 1/2 pixel is obtained by using a _(2i-1) and a _{(2i + 1)} .

The above description has been made with reference to the case where a one-dimensional image model in the horizontal direction is taken as an example, but a case where this is applied to two dimensions will be described with reference to FIG. At first,
The case of the LH signal having a high band in the horizontal direction and a low band in the vertical direction will be described. In FIG. 4, for the arrangement mode of the 1/2 pixel of the X mark with respect to the integer pixel of the two-dimensional array of the O mark,
There are three possible ways shown in (A) to (C). Same figure (A)
Shows a case of 1/2 movement in the horizontal direction, FIG. 7B shows a case of 1/2 movement in the vertical direction, and FIG. 7C shows a case of 1/2 movement in both the horizontal and vertical directions.

Of these, 1/2 shown in FIG.
For the pixel QA, as in the one-dimensional case described above,
A value obtained by applying a digital filter having a negative sum of filter coefficients to integer pixel values located on the left and right may be used as the signal value. On the other hand, the 1/2 pixel QB shown in FIG.
As for LH, since LH has a low band in the vertical direction, it is necessary to obtain a signal value by an interpolation operation using upper and lower pixel values as in the conventional technique.

Next, for the 1/2 pixel QC shown in FIG. 4C, the signal value is obtained by interpolation in the vertical direction as in the case of (B), and the filter output value is obtained in the horizontal direction. The signal value. For example, for 1/2 pixel QCS, LQCS = (1/8). (G1 + G5) / 2 + (-5/8). (G2 + G6) / 2 + (-5/8). (G3 + G7) / 2 + (1/8) ・ (G4 + G8) / 2 …………………… (9).

Next, the horizontal and vertical directions of the HL signal of the low band in the horizontal direction and the high band in the vertical direction may be reversed in the horizontal and vertical directions. For the HH signal in the high band in the horizontal direction and the high band in the vertical direction, an output value obtained by applying a digital filter having a negative sum of filter coefficients to peripheral pixel values in both the horizontal and vertical directions may be used as a signal value.

The calculation for obtaining the interpolated value described for the LL signal and the like in the prior art and this embodiment can be regarded as a digital filter calculation. However, the sum of the coefficients of the digital filter used to obtain the 1/2 pixel value in the present embodiment is negative, whereas the sum of the filter coefficients is positive in the interpolation calculation. .

A method of using an output value obtained by applying a digital filter to peripheral pixel values as a 1/2 pixel value instead of the simple interpolation operation by averaging two pixels has been conventionally proposed. However, the sum of the filter coefficients is positive. As described above, if the filter coefficients are expressed as in the above equation (8), their filter characteristics are ideal L
It is close to the characteristics of PF. This difference is due to the property peculiar to the subband-divided high-band signal and does not occur unless subband division is performed.

As described above, according to the present embodiment, the HL signal, L, which is the high band component of the current frame, is compared with the conventional technique.
It is possible to reduce the post-motion prediction inter-frame prediction code amount of the H signal and the HH signal. As a result, it is possible to satisfactorily compress the code amount of the entire image, and it is possible to reduce the transmission capacity and recording capacity when transmitting or recording the image information. Moreover, when the code amount of the entire image is limited to a certain fixed amount, the image quality at the time of decoding is improved.

<Second Embodiment> Next, a second embodiment of the present invention will be described. Although the motion vector search accuracy is set to 1/2 pixel accuracy in the above embodiment, this embodiment is further divided into 1/4.
It is intended to be highly accurate with pixels.

The signal value of 1/4 pixel is equal to integer pixel and 1/2
The signal value of the intermediate position of the pixel, the signal value of the 1/4 pixel is obtained by interpolation from the signal value of the 1/2 pixel and the signal value of the integer pixel obtained as described above. For example, in FIG. 3C, the signal value of the 1/4 pixel at the midpoint between the 1/2 pixel PQ and the integer pixel P3 is (1/2) (L
It is calculated by Q + L3). It should be noted that instead of a simple interpolation operation by averaging two pixels, an output value obtained when a digital filter having a positive sum of filter coefficients is applied to peripheral integer pixels and 1/2 pixels may be used. Is natural.

<Third Embodiment> Next, a third embodiment of the present invention will be described with reference to FIG. In the first embodiment, the image signal is equally divided into four LL, HL, LH, and HH, but the motion compensation predictive coding may be performed after further division (Nosezawa et al., " Frequency domain realization of motion-compensated interframe prediction in non-uniform subband / wavelet transform coding ", IEICE Research Report IE91-83).

According to this, the LL signal, which is divided into a low band and a high band in each of the horizontal direction and the vertical direction, and has a low band in both the horizontal direction and the vertical direction, is recursively divided into motion compensations for each band. A method of performing interframe predictive coding has been pointed out. FIG. 5 shows the state of this band division. The divided signal according to FIG. 6 is divided into LL0, HL0,
It is represented by LH0 and HH0. According to this second embodiment, the portion of the LL0 signal is further in the low band in the horizontal and vertical directions,
It is divided into high band LL1, HL1, LH1 and HH1. Further, the portion of the LL1 signal is further divided horizontally and vertically into low band and high band LL2, HL2, LH2 and HH2.

In such a case, the same processing as that in the first embodiment is performed for the signals in the low and horizontal bands, that is, the signals in the other bands except the LL2 signal in the lowest band. To do. That is, the 1/2 pixel is obtained between the block of the reference frame corresponding to the macro block of the current frame and its periphery, and the block of the reference frame including this is compared with the macro block in the current frame to determine the motion compensation block. Try to find the motion vector. Then, either the difference value for each pixel value or the original pixel value of the macroblock is adaptively selected and encoded.

It is to be noted that HL1 and HL2 have a low band in the vertical direction like HL0, and LH1 and LH2 have a low band in the horizontal direction like LH0.
As for HH1 and HH2, note that both HH1 and HH2 have a high bandwidth in both the vertical and horizontal directions.
It is natural to obtain the pixel value (see FIG. 4). Further, at this time, the point that the size of the macroblock is changed for each band is as described in the above document.

<Fourth Embodiment> Next, a fourth embodiment of the present invention will be described. As shown in FIG. 3C,
The motion vector of the macro block of the high band signal is obtained by searching the surroundings using the motion vector obtained for the macro block of the low band signal corresponding to the macro block. Obviously, the search range may be considerably narrower than it would be without the use of low band signals.
Moreover, if the motion vector is encoded only in the narrow range, the amount of information (the number of bits) after encoding can be obviously reduced.

In the present embodiment, attention is paid to such a point, and a motion vector of a macroblock of a high band signal, a vector corresponding to a motion vector obtained for a macroblock of a low band signal corresponding to the macroblock. And the difference vector between and is encoded and transmitted. Figure 3
In the example of (C), a vector Δ corresponding to the motion vector obtained for the macroblock of the low band signal (FIG. 3B).
(Reference) and the difference ΔC−Δ between the final motion vector ΔC obtained by searching for the surrounding area) and is encoded and transmitted.

By doing so, it is possible to reduce the number of bits after the coding of the motion vector of the high-band signal, and it is possible to satisfactorily compress the code amount of the entire image, and thus the image information. It is possible to reduce the transmission capacity and the recording capacity when transmitting or recording.

When sub-band division is performed as shown in FIG. 5, LL2 which is a low band signal and HL which is a high band signal.
As described in the third embodiment, it is necessary to change the size of the corresponding macroblock for 1, LH0, etc. That is, when the macroblock of LL2 is m vertical pixels x n horizontal pixels, the macroblock of HL1 is vertical 2 m.
Pixels × 2 × n pixels horizontally, and LH0 macroblocks need to be 4 × m pixels vertically × 4 × n pixels horizontally. This also applies to the motion vector, and the vector corresponding to the motion vector obtained in LL2 must be doubled in the horizontal and vertical directions in HL1 and in the horizontal and vertical directions in LH0. Both need to be quadrupled.

<Other Embodiments> The present invention is not limited to the above embodiments, and includes, for example, the following. (1) As the reference frame when the current frame is a B picture, any one of a front frame, a rear frame, and an interpolated value obtained from the front frame and the rear frame can be selected.

(2) In the above embodiment, motion compensation is performed between frames, but field / frame adaptive motion compensation may be performed. (3) Further, either the difference value selected in the above embodiment or the original pixel value may be simply subjected to DCT or may be subjected to field / frame adaptive DCT.

(4) Various processes may be performed as necessary, for example, subband coding may be further performed on each band signal of the current frame predicted between motion-compensated frames in the above embodiment. .

(5) The filter coefficient shown in the above embodiment may be changed as appropriate. As a very simple method, for example, the positive / negative inversion value of the interpolation value obtained by averaging the pixel values of two simple pixels is set as the signal value of 1/2 pixel. (6) As one of preferred application examples of the present invention, a code is recorded in a medium for each sub-band shown in FIG. 6, and only the code of the LL signal is read out from these to read, for example, an image of NTSC system. In some cases, reproduction is performed, the entire code is read, and, for example, high-definition video reproduction is performed. But,
Subband Coding Generally, the present invention is applicable.

[0070]

As described above, the image information compression method according to the present invention has the following effects. (1) When encoding by motion compensation for each band signal divided into sub-bands, for high band signals, 1/2 of the signal value obtained by applying a filter having a negative sum of coefficients to peripheral integer pixels Since the motion-compensated prediction is performed using pixels, it is possible to reduce the code amount of the high band signal.

(2) When performing coding by motion compensation for each band signal obtained by sub-band division, for high band signals, a filter whose negative sum of coefficients is negative is applied to peripheral integer pixels to obtain signal values. In addition to the 1/2 pixel, the 1/4 pixel obtained by further performing the interpolation is used to perform the motion compensation prediction, so that the code amount of the high band signal can be further reduced.

(3) With respect to the motion vector of the high band signal, since the difference vector from the vector corresponding to the motion vector of the low band signal is coded, the code amount can be further compressed.

[Brief description of drawings]

FIG. 1 is a flowchart showing a first embodiment of an image information compression method according to the present invention.

FIG. 2 is an explanatory view showing the operation of the first embodiment.

FIG. 3 is an explanatory diagram showing an operation of the first embodiment.

FIG. 4 is an explanatory diagram showing a position mode of intermediate pixels when the first embodiment is applied two-dimensionally.

FIG. 5 is an explanatory diagram showing a state of subband division according to the third embodiment of the present invention.

FIG. 6 is an explanatory diagram showing an example of subband division of an HD-TV signal.

[Explanation of symbols]

LL, HL, LH, HH, LL0, HL0, LH0, HH
0, LL1, HL1, LH1, HH1, LL2, HL2, LH
2, HH2 ... Band signals divided into subbands, L1 to L
7, LA to LC, LQ to LT ... Signal level, P1 to P7
... integer pixels, PA, PB, PC ... integer pixels, PQ to PT
... 1/2 pixel, QA, QB, QC ... two-dimensional intermediate pixel,
WA ... Macro block, WB ... Motion compensation block obtained with LL signal, WC ... Motion compensation block obtained with 1/2 pixel precision, Δ ... Current frame and reference frame obtained with macro block with low band signal Motion vector between,
ΔC: Motion vector obtained with 1/2 pixel accuracy.

Claims (3)

[Claims] 1. When an image signal of an image of a predetermined unit is sub-band divided into a low band signal and a high band signal and inter-frame coding by motion compensation predictive coding is performed for each band signal, at least a high band signal is generated. For band signals, 1 between integer pixels
In the image information compression method for performing highly accurate motion-compensated prediction using / 2 pixels, when a filter whose sum of filter coefficients is negative is applied to the signal values of peripheral integer pixels as the signal value of the 1/2 pixel. An image information compression method characterized by using the output value of 2. When an image signal of an image of a predetermined unit is subband-divided into a low band signal and a high band signal, and at the time of performing inter-frame coding by motion compensation predictive coding for each band signal, at least For band signals, 1 between integer pixels
In the image information compression method for performing highly accurate motion-compensated prediction using pixels of 1/2 pixel and 1/4 pixel or more, the sum of the filter coefficients is negative as the signal value of 1/2 pixel. Using the output value when the filter is applied to the signal values of the peripheral integer pixels, the filter whose sum of the filter coefficients is positive is used as the signal value of the pixel having the accuracy of 1/4 pixel or more. And an image information compression method using an output value obtained by multiplying a signal value of 1/2 pixel. 3. The image information compression method according to claim 1, wherein the motion vector of the high band signal is encoded by a difference vector from a vector corresponding to the motion vector of the low band signal. Image information compression method.