JPH07203443A - Moving image coder - Google Patents

Abstract

PURPOSE:To obtain the moving image coder suppressing occurrence of block distortion even with respect to a picture in which block distortion appears remarkably. CONSTITUTION:An IDCT 11 applies inverse discrete cosine transformation to a transformation coefficient including a quantization error resulting from an image received timewise continuously quantized at a prescribed quantization step width to generate decoded data blocks, and they are given to an SVF 12, which detect positions MHP, MVP of a maximum frequency component based on the discrimination by a valid/invalid block discrimination section 22 and an output of an inverse quantization device Q<-1> 9. Then a low pass filter among plural low pass filters whose characteristics differ from each other is selected and applied to a border of notice blocks and inside of the notice block then production of block distortion is suppressed in a decoded image.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a compression encoding device for compressing digital moving image data, and is particularly suitable for a moving image within a screen such as an ultrasonic diagnostic image and a moving image whose moving direction is limited. The present invention relates to a moving picture coding device.

[0002]

2. Description of the Related Art The conventional compression encoding of moving images by a conventional moving image encoding apparatus is a mixture of a frame for intra-frame encoding and a frame for inter-frame encoding. This intra-frame encoding is performed by independently inputting the input image frame, for example, in D by 8 × 8 pixel block units.
After CT transformation is performed and the transform coefficient is quantized, variable length coding is performed on the quantized value.

On the other hand, the inter-frame coding is performed in order to reduce the deterioration of the redundancy in the time direction due to the motion between the input image frame and the decoded image of the previous frame.
A motion vector of a block in units of 6 × 16 pixels) is detected, and motion is compensated in units of blocks. The difference between the motion-compensated prediction image and the input image is calculated, the generated motion-compensated prediction error data is subjected to DCT transform in block units of 8 × 8 pixels, the transform coefficient is quantized, and the quantization is performed. The variable length coding is performed on the value.

Since such compression encoding enables efficient compression, the H.264 standard, which is an international standard for television radio waves, is used. 261 and MPEG which is an international standard for storage media.

[0005]

However, the compression coding by the moving picture coding apparatus described above basically uses the DCT in block units. For this reason, when the high compression is performed, the quantization step width of the DCT transform coefficient is controlled to be coarse, so that the quantization error increases, and as a result, the decoded image has a discontinuity at the block boundary noticeable. Like This deterioration is generally called block distortion.

In particular, when an image obtained from an ultrasonic diagnostic apparatus is compressed by such a compression method using DCT, block distortion appears more noticeably than a general image. Therefore, it is an object of the present invention to provide a moving picture coding apparatus that suppresses the occurrence of block distortion even in an image in which block distortion is prominent.

[0007]

In order to achieve the above-mentioned object, prediction is performed between an input image signal of a current frame and a decoded image of a previous frame with respect to an input image signal continuously input in time, A means for generating an inter-frame prediction error signal by subtracting each image, a prediction error signal or an input image signal is divided into blocks of n × m pixels, orthogonal transformation is performed in block units, and transform coefficients are quantized to obtain a variable length code. And a means for performing quantization, and means for performing inverse quantization on the quantized transform coefficient, performing inverse orthogonal transform, adding the result to the predicted image from the decoded image of the previous frame, and generating the decoded image of the current frame. In, the conversion coefficient of each of the blocks, or the information related to this conversion coefficient as a parameter,
A moving image coding including a filter selecting means for selecting one filter from a plurality of types of low-pass filters, and a filtering means for applying a signal after the filter selected by the selecting means is subjected to inverse orthogonal transform. Provide a device.

Further, with respect to an input image signal continuously input in time, prediction is performed between the current frame input image and the previous frame decoded image and subtraction is performed for each image to generate an inter-frame prediction error signal. Means, means for dividing the prediction error signal or the input image signal into blocks of n × m pixels, performing orthogonal transform in block units, quantizing transform coefficients to perform variable length coding, and means for quantizing transform coefficients. In a moving picture coding apparatus having means for generating a current frame decoded image by performing dequantization, performing inverse orthogonal transform and adding with a predicted image generated from a preceding frame decoded image, a motion vector for detecting a motion vector for each block is detected. Vector detection means and one block for each block are combined into a block line, and a block line for obtaining the frequency of occurrence of a motion vector in units of the block line Motion vector frequency calculation means, based on the frequency obtained by the motion vector frequency calculation means in the block line, representative motion vector generation means for generating a representative motion vector in the block line, the block line is divided into a plurality of, Motion compensation determination means for determining whether to permit or prohibit motion compensation in the divided block line based on the motion vector generation frequency distribution for each divided block line, and division for which motion compensation is permitted by the motion compensation determination means. There is provided a moving picture coding apparatus configured by means for determining and controlling whether or not to perform motion compensation prediction using the representative motion vector for each block in a block line.

[0009]

In the moving picture coding apparatus having the above-described structure, for example, the highest spatial frequency is determined by the DCT coefficient position in the block, and a low-pass filter having a plurality of types of characteristics is selected by the DCT coefficient position. To affect the data, the block distortion caused by the quantization error is erased,
Moreover, blurring is suppressed as much as possible. However, the filter has discontinuity in block boundaries other than quantization error (especially, when motion compensation is performed independently for each block)
At, on the contrary, it must remain discontinuous and the filter will not work.

Further, in order to suppress this inconvenience, one representative motion vector is obtained within the block line, and whether or not motion compensation using this representative motion vector is permitted is determined by a divided block line (for example, divided into two). Judged in units,
Only when it is permitted, the motion compensation is performed as much as possible on a block-by-block basis. On the contrary, when it is prohibited, the inter-frame difference (movement vector = 0) is performed. This makes it possible to avoid discontinuity at the block boundary in the motion-compensated prediction image as much as possible, and when discontinuity occurs due to motion-compensated prediction at the block boundary, the filter is not applied to the boundary and the quantum is discontinued. Block distortion due to digitization error can be effectively suppressed.

[0011]

Embodiments of the present invention will now be described in detail with reference to the drawings. FIG. 1 shows and describes a configuration of a moving picture coding apparatus as a first embodiment according to the present invention. In this moving image encoding apparatus, an input image is input in block units (for example, 8 × 8 pixels) to a switch 2 directly and in the form of difference data via a subtractor 1.

The subtractor 1 receives either the decoded image of the previous frame stored in the frame memory 15 or the original image of the previous frame stored in the frame memory 16 via the switch 17 as another input. One of them is selected by the control of the valid / invalid frame determination unit 22.
(When it is judged as a valid frame, the decoded image of the previous frame is selected as the original image of the previous frame when it is judged as an invalid frame)

In the in-frame / inter-frame determination section 4, the first frame of a time-determined period (for example, 1/2 second) or the signal from the valid / invalid frame determination section 22 means an invalid frame. If so, the input image signal is controlled to be output from the switch 2 so as to perform intra-frame encoding, and otherwise, the subtracter 1 is controlled to output the difference data from the switch 2.

The difference data output from the subtractor 1 is
Also, the sum of absolute differences of the difference data is calculated by inputting to the sum of absolute differences between blocks 20. This sum of absolute differences is
It is input to the valid / invalid block determination unit 5 and the valid / invalid frame determination unit 22.

In the valid / invalid block judgment section 5, the sum of the absolute value of the difference in the input block and the threshold value 1 determined in advance are set.
The output of switch 2 is DC when the threshold value is 1 or more.
The switch 3 is controlled so as to be input to T6, and when the threshold value is smaller than 1, the switch 3 is controlled so that the output of the switch 2 is not input to the DCT6. In addition, the valid / invalid frame determination unit 22 adds the sum of absolute differences between blocks for one frame, and when the sum of absolute differences between frames is equal to or more than a predetermined threshold value 2, or the code amount stored in the buffer 19. Is a threshold value of 3 or more, and when the rate control unit 24 further determines that the transmission buffer 25 overflows, the switch 21 is controlled to be disconnected as an invalid frame, and otherwise the switch 21 is connected as an invalid frame. Control.

Also, 1-bit information per block, which is the valid block or invalid block determined by the valid / invalid block determining section 5, is output to the MPX section 23. Further, the valid / invalid frame determination unit 22 outputs 1-bit information per frame indicating the determined valid frame or invalid frame to the MPX unit 23.

The 8 × 8 pixel original image or the inter-frame difference image output from the switch 3 is input to the DCT 6, subjected to discrete cosine transform, and input to the quantizer Q7. In the quantizer Q7, two types of quantization stored in advance in the quantization table 10 are determined by the intra-frame / inter-frame determination unit 4 (whether intra-frame coding or inter-frame coding is used). Select a matrix (the weights for each transform coefficient are arranged in a matrix) and input.
Further, the quantization scale from the rate control unit 24 is input. The quantization matrix is multiplied by the quantization scale, the quantization step width of each transform coefficient is calculated, and the transform coefficient which is the output of the DCT 6 is quantized using this quantization step width.

Then, the quantized transform coefficient is input to the variable length coder VLC8 and the inverse quantizer Q ^-1 9. The variable length encoder VLC8 outputs a Huffman code corresponding to the difference value between the DC component of the transform coefficient and the preceding block, and performs zigzag scanning from the low frequency component to the high frequency component for the AC component. A zero line and a two-dimensional Huffman code for a non-zero value immediately after the zero line are output for the one-dimensionalized data. Further, when zero continues to the end of the block, EOB (End of Bl
ock) code is output. The inverse quantizer Q ^-1 9 receives the same quantization matrix and quantization scale as those of the quantizer Q7, generates a quantization step width, and multiplies this quantization step width by the inverse quantization value. The transformed conversion coefficient is generated, and IDCT11 and SVF (S
Pace Variant Filter) 12.

In the IDCT 11, the inverse discrete cosine transform is performed on the inversely quantized transform coefficient to generate a decoded data block. This decoded data block is input to the SVF 12.

[0020] The In SVF12, position MHP and MVP of the highest frequency component not horizontal and vertical zero inversely quantized transform coefficient values are input from the inverse quantizer Q ^-1 9 as shown in FIG. 2 Is detected and MH as shown in FIG.
A plurality of low-pass filters having different characteristics are prepared in the horizontal and vertical directions using P and MVP as parameters, and the selected numbers are recorded in the memory.

The decoded image block input from the IDCT 11 is stored in the buffer memory in the SVF 12 (not shown), and the four adjacent upper, lower, left, and right blocks are provided.
When one block is stored in the buffer memory, the selected horizontal and vertical low-pass filters are respectively applied to the target block boundary and the inside thereof.

In the case of ultrasonic diagnostic images, this filter is sufficiently effective only in the horizontal direction. Further, it goes without saying that the characteristics of the filter may be changed from the block boundary toward the center.

The decoded data block after the low-pass filter processing is input to the switch 13 and is input to the adder 14 only when the valid / invalid block determination unit 5 determines that the block is a valid block. In the adder 14, zero is selected when intraframe coding is selected by the intraframe / interframe determination unit 4, and a decoded image of the previous frame controls the switch 18 when interframe coding is selected. It is input by doing.

The data input to the adder 14 are added, a decoded image of the currently encoded image is generated and stored in the frame memory 15. The original image data is stored in the frame memory 16 as it is.

When the valid / invalid frame determining unit 22 determines that the frame is a valid frame, the switch 17 is controlled so that the decoded image stored in the frame memory 15 is output to the subtracter 1 and the switch 18. If it is determined that the frame is invalid, the switch 17 is switched to output the original image stored in the frame memory 16 to the subtractor 1 and the switch 18.

The buffer 19 also stores variable-length coded data for one frame and outputs the data amount of the stored code to the valid / invalid frame determination section 22. The MPX unit 23 includes a quantization scale from the rate control unit 24, valid / invalid frame information from the valid / invalid frame determination unit 22, valid / invalid block information from the valid / invalid block determination unit 5, and The compressed data stored in the buffer 19 is input via the switch 21,
The data is rearranged into a predetermined bit stream and output to the transmission buffer 25 and the rate control unit 24.

The bit stream accumulated in the transmission buffer 25 is output at a constant rate. Further, the rate control unit 24 counts the code amount of the bit stream output from the MPX unit 23, and the quantization scale (quantization step of the next encoding frame is set so that the generated code amount falls within the set bit rate. Width). This prediction is
It is performed independently during interframe coding.

The prediction formula is as follows: Q _s (I) = (Next frame allocation code amount (I) / Current frame generated code amount (I)) ^1/2 × Q _s ′ (I) Q _s (P) = (Next frame allocation code amount (P) / Current frame generation code amount (P)) ^1/2 × Q _s ′ (P)

Here, Q _s ′ (I) and Q _s ′ (P) are the quantization scales used in the current intra-coded frame and the current inter-frame coded frame, and Q _s (I) and Q _s (P ) Represents the quantization scale used in the next intra-frame coded frame and the next inter-frame coded frame. The quantization scale is input quantizer Q7 to the inverse quantizer Q ^-1 9, the quantization step width used in the next frame is generated.

By repeating the processing for one frame as described above, the moving image is compressed. According to the present embodiment, the discontinuous distortion (block distortion) generated at the adjacent boundary of the decoded data block, which is the output result of the IDCT 11, can be eliminated, so that the decoded image at the output of the adder 14 is prevented from generating the block distortion. Can be suppressed. The first embodiment is a case where the inter-frame prediction is a simple inter-frame difference.

Next, FIG. 4 shows the configuration of a moving picture coding apparatus as a second embodiment according to the present invention. In this moving image coding apparatus, after the input image is written in the buffer 30 and a predetermined time has elapsed, the input image is written in block units (eg 8 × 8 pixels).
Each time, the difference data is input to the switch 32 directly and via the subtractor 31.

The subtractor 31 receives the decoded image of the previous frame stored in the frame memory 47 via the switch 54 as another input, or the frame memory 5
Either one of the previous frame input original images stored in No. 3 is selected by the control of the valid / invalid frame determination unit 60. However, when it is determined that the frame is valid, the decoded image of the previous frame is selected, and when it is determined that the frame is invalid, the original image of the previous frame is selected.

Then, the in-frame / inter-frame determination unit 34
A switch for performing intra-frame encoding when the first frame in a time-determined period (for example, 1/2 second) or the signal from the valid / invalid frame determination unit 60 means an invalid frame. The input image signal is controlled to be output from 32, and in other cases, the difference data from the subtractor 31 is switched to be output from the switch 32.

The difference data output from the subtractor 31 is also input to the intra-block difference absolute value sum calculation unit 59, and the absolute value sum of the difference data is calculated. This sum of absolute differences is input to the valid / invalid block determination unit 35 and the valid / invalid frame determination unit 60.

The valid / invalid block judgment unit 35 compares the input sum of absolute differences in blocks with a predetermined threshold value, and when the threshold value is 1 or more, the output of the switch 32 is input to the DCT 36. Control the switch 33 to
When the threshold value is smaller than 1, the output of the switch 32 is DCT3.
The switch 33 is controlled so as not to be input to the switch 6. Further, the valid / invalid frame determination unit 60 adds the sum of absolute differences between blocks for one frame, and when the sum of absolute differences between frames is equal to or greater than a preset threshold value 2, or the buffer 5
When the code amount stored in 7 is a threshold value of 3 or more, and when the rate control 62 determines that the transmission buffer 63 overflows, the switch 58 is disconnected as an invalid frame, and in other cases, the switch 58 is turned on. Switch to.

Further, the valid / invalid block determining section 35.
The 1-bit information per block, which is the valid block or the invalid block determined in step S1, is output to the MPX 61. Further, the valid / invalid frame determination unit 60 outputs to the MPX 61 1-bit information per frame indicating the determined valid frame or invalid frame.

The 8 × 8 pixel original image or the inter-frame difference image output from the switch 33 is input to the DCT 36, subjected to discrete cosine transform, and input to the quantizer Q37. The quantizer Q37 has two types of quantization matrices stored in advance in the quantization table 40 according to the determination in the intra-frame / inter-frame determination unit 34 (whether intraframe coding or interframe coding is performed). (The weight for each transform coefficient is arranged in a matrix) is selected and input, and the quantization scale from the rate control 62 is further input. The quantization matrix is multiplied by the quantization scale to calculate the quantization step width of each transform coefficient, and the DCT is calculated using this quantization step width.
The transform coefficient which is the output of 36 is quantized.

The quantized transform coefficient is the variable length coder V.
It is input to the LC 38 and the inverse quantizer Q ^-1 39. The variable length encoder VLC38 outputs the Huffman code corresponding to the difference value from the preceding block to the DC component of the transform coefficient to the buffer 57, and the AC component from the low frequency component to the high frequency component. Zigzag scanning is performed to output a zero line for the one-dimensionalized data and a two-dimensional Huffman code for the non-zero value immediately after that are output to the buffer 57. Furthermore, when zero continues to the end of the block, an EOB (End of Block) code is output. The inverse quantizer Q ^-1 39 receives the same quantization matrix and quantization scale as those of the quantizer Q37, generates a quantization step width, and multiplies the quantization step width by the quantization value to obtain the inverse value. The quantized transform coefficient is generated, and the IDCT 41 and the SVF (Space Va) are generated.
(Liant Filter) 44.

In the IDCT 41, the input transform coefficient is subjected to inverse discrete cosine transform to generate a decoded data block. This data is stored in the buffer 42 for the set number of blocks. The number of set blocks is the number of block data required by the SVF 44. For example, SVF4
In case of filtering only in the horizontal direction at 4, it is sufficient to store three consecutive blocks in the horizontal direction, and in case of filtering in the horizontal and vertical directions, 2 block lines + 1
It requires the accumulation of blocks. (“0” is stored as a filter in the invalid block position by the number of block pixels.)

Then, the processing block is input to the switch 43 from the decoded data accumulated in the buffer 42, and the filter O is controlled by the filter ON / OFF determination unit 56.
If it is determined to be N, this data is output to the SVF 44. When it is determined that the filter is off,
It is directly output to the switch 45.

In the SVF 44, the positions MHP and M of the highest frequency component in the horizontal and vertical directions of the inversely quantized transform coefficient value which is the input from the inverse quantizer Q ^-1 39.
VP is detected (shown in FIG. 2), a plurality of low-pass filters (shown in FIG. 3) with different characteristics prepared in the horizontal and vertical directions are selected using MHP and MVP as parameters, and this selection number is recorded in memory. To do. However, for the invalid block, (MHP = MVP = 0) is selected with the highest frequency component position as the DC component.

Next, for the decoded data block input from the buffer 42 via the switch 43, the low pass filter according to the selection number of the filter is applied to the block boundary and the inside thereof. The decoded data block after this filter processing is output to the switch 45, and is input to the adder 46 only when the valid / invalid block determination unit 35 determines a valid block. In the adder 46, “0” is selected when the intra-frame coding is selected by the intra-frame / inter-frame determining unit 34, and the decoded image of the previous frame is switched by the switch 55 when the inter-frame coding is selected. It is input by controlling. The data input to the adder 46 is added, a decoded image of the currently encoded image is generated, and the contents at a predetermined position of the frame memory 47 (decoded image data of the previous frame) are rewritten. The original image data is stored in the frame memory 53 as it is.

FIG. 7 shows a flow chart for motion detection and motion compensation, which will be described. First, ME48
Is input with the original image data and the decoded image data of the previous frame stored in the frame memory 47, and motion detection is performed (step S1). For motion detection, the original image data is within a search area (ultrasound diagnosis) within a range of ± d pixels around the same position as the original image data block of the decoded image data of the previous frame with respect to the n × m pixel block (for example, 8 × 8 pixels). When used for an image, the range is ± d pixels in the horizontal direction).

The criterion for determining the motion vector to be obtained is that the sum of absolute differences between the original image and the decoded image of the previous frame is minimum, and this sum of absolute differences is (sum of absolute differences at position 0 of motion vector). If the condition is not satisfied (NO), the motion vector is set to 0 (step S2, 3). When obtaining a position with an accuracy of one pixel or less, an interpolation image for that position is created from the decoded image of the previous frame, and the sum of absolute differences is calculated in the same manner. Here, as shown in FIG. 5, α is a constant greater than 1, and the frequency is calculated within one block line for the motion vector of each block, and the motion vector with the maximum frequency other than zero is represented in the block line. Vector (steps S4 to S8).

That is, this representative motion vector is MC5
0 and output to the MPX unit 61. Further, the difference absolute value sum storage memory 49 stores the difference absolute value sum at the motion vector 0 output from the ME 48.

Further, the block line is divided into a plurality of parts (for ultrasonic diagnostic images, it is divided into two), and the motion vector frequency is calculated for each of the divided block lines (steps S4 to
S6), and for this frequency distribution, [-Δ, Δ],
Recalculate the frequencies in the range of [representative, motion vector −Δ, representative motion vector + Δ]. Here, Δ ≦ | representative motion vector / 2 | is a constant and [-a, b] is -a
Represents an area that includes and does not include b (steps S9 to S
11).

The two frequencies in each divided block line area are defined as H _o (n) and H _mv (n). Here, n is an index representing the position of the divided block line shown in FIG. 6 (step S12). When these frequencies satisfy H _o (n) <H _mv (n), motion compensation in this divided block line is permitted, and the others are prohibited. This signal is output to the MC determination section 51.

Then, the MC 50 reads the previous frame decoded image in the frame memory 47 according to the representative motion vector from the ME 48, generates a motion compensation predicted image block, and outputs it to the switch 52. In addition, the sum of absolute differences (sum of absolute MC errors) between the original image and the motion-compensated prediction image block is calculated and output to the MC determination unit 51. Further, the difference absolute value sum (0) at the same block position is output from the difference absolute value sum storage memory 49 to the MC determination unit 51 (steps S13 and S14).

In the MC determining section 51, only when a signal for permitting motion compensation in the divided block line is received from the ME 48, the sum of MC error absolute values <sum of absolute difference values (0) × β
It is determined that the block to be subjected to the motion compensation is subjected to the motion compensation, and the signal with or without the motion compensation is output to the filter ON / OFF determination unit 56 and the MPX unit 61 as the control of the switch 52.

The switch 52 includes a frame memory 47.
If the previous frame decoded image of the above and the motion compensation prediction image from the MC 50 are input and the MC determination unit 51 determines that there is motion compensation, the motion compensation prediction image from the MC 50 is output to the switch 54, and the MC determination unit If it is determined in step 51 that there is no motion compensation, the previous frame decoded image in the frame memory 47 is output to the switch 54 (step S1).
6, S17).

The other input of the switch 54 contains the original image data of the previous frame stored in the frame memory 53, and the control signal from the valid / invalid frame determination unit 60 is a valid frame (for the previous frame). ), The input from the switch 52 is output, and when the frame is an invalid frame (relative to the previous frame), the input from the frame memory 53 is output. This output enters the subtractor 31 and the switch 55 (steps S18 and S1).
9).

The filter ON / OFF judging unit 56 can store the output from the MC judging unit 51 for several blocks. The MC judgment of the block one block before and the MC judgment of the current block are stored. When the states of and are the same, the switch 43 is controlled so as to filter at the two block boundaries. When the MC determination of the previous block and the MC determination of the current block are different, the switch 43 is controlled so as not to apply the filter.

The buffer 57 also stores the variable-length coded data for one frame and outputs the data amount of the stored code to the valid / invalid frame determination section 60. And MPX
The unit 61 includes a quantizer scale from the rate control unit 62, valid / invalid frame information from the valid / invalid frame determination unit 60, valid / invalid block information from the valid / invalid block determination unit 35, and a switch 58. Compressed data stored in the buffer 57 via
The block line representative motion vector from the ME 48 and the MC presence / absence block information from the MC determination unit 51 are input, and these data are rearranged in a predetermined bit stream syntax and output to the transmission buffer 63 and the rate control unit 62. To do.

The bit stream stored in the transmission buffer 63 is output at a constant rate. Further, the rate control unit 62 counts the code amount of the bit stream output from the MPX unit 61, and the quantization scale (quantization step width) of the next encoded frame is adjusted so that the generated code amount falls below the set bit rate. ) Predict. This prediction is performed independently during intraframe coding and interframe coding.

The prediction formula is as follows. Q _s (I) = (Next frame assigned code amount (I) / Current frame generated code amount (I)) ^1/2 × Q _s ′ (I) Q _s (P) = (Next frame assigned code amount (P) / Current frame generated code amount (P)) ^1/2 × Q _s ′ (P)

Here, Q _s ′ (I) and Q _s ′ (P) are the quantization scales used in the current intra-frame coded frame and the current inter-frame coded frame, and Q _s (I) and Q _s (P ) Represents the quantization scale used in the next intra-frame coded frame and the next inter-frame coded frame.

This quantization scale is input to the quantizer Q37 and the inverse quantizer Q ^-1 39 to generate the quantization step width used for the next frame. Also, in this embodiment, S
Whether or not to operate the VF44 was controlled by the switch 43, but the signal of the filter ON / OFF determination unit 56 is directly input to the SVF44, and when the filter is OFF, the characteristics of the filter are forcibly set as shown in FIG. MHP = MV
The same effect can be obtained by setting P = 7.

The moving image is compressed by repeating the above-described processing for one frame. According to this embodiment, it is possible to efficiently compensate for the movement of the graph portion that moves uniformly in the horizontal direction, which is often seen in, for example, an ultrasonic diagnostic image, and compression efficiency is improved.

In the SVF 44, the block boundary is originally discontinuous (not due to quantization error).
, The discontinuity may be erased, and if motion compensation is independently performed in block units, the block boundary of the motion compensation prediction error is very likely to be discontinuous. , SVF also causes deterioration at block boundaries. This defect is caused by the block line representative motion vector, the motion compensation permission / prohibition determination in the divided block line, and the motion compensation state of the block
At the boundary of adjacent blocks with different (none), the above-mentioned problem can be greatly reduced by performing the process of not applying the filter.

In this embodiment, the valid / invalid block determination and the motion compensation prediction presence / absence determination are performed separately, but the functions of the valid / invalid block determination unit 35 and the intra-block difference absolute value sum calculation unit 59 are provided. It is also possible to cause the MC determination unit 51 to carry. In that case, the MC determination unit 51 further outputs the valid / invalid block information and the intra-block difference absolute value sum.

Next, FIG. 8 shows a configuration of a moving picture coding apparatus as a third embodiment according to the present invention and will be described. In this moving image encoding apparatus, the input original image is stored in the buffer 70.
Is written in a block, and after a predetermined time has passed,
X8 pixels), and is input to the switch 72 in the form of difference data via the subtractor 71.

The subtractor 71 receives, as another input, the decoded image block of the previous frame or the input original image block of the previous frame via the switch 92 or the motion-compensated prediction image block using these images as reference images. Difference data is calculated. The difference data is output to the block difference absolute value sum calculation 99 and the switch 72.

Next, in the in-frame / inter-frame determination section 74, when the first frame of a predetermined period (for example, 1/2 hour) or the signal from the valid / invalid frame determination section 100 means an invalid frame The input original image block is controlled to be output from the switch 72 so that the intra-frame encoding is performed, and the difference data block from the subtracter 71 is controlled to be output from the switch 72 in other cases.

Then, the sum of absolute differences between blocks is calculated 99
Then, the absolute value sum of the input difference data is calculated, and this in-block difference absolute value sum is output to the valid / invalid block determination unit 75 and the valid / invalid frame determination unit 100. Further, the maximum absolute value of the difference within the block is output to the valid / invalid block determination unit 75. In the valid / invalid block determination unit 75, the input sum of absolute differences in block and maximum absolute difference in block are compared with the threshold 1 and the threshold 2, respectively.

That is, the switch 73 is switched so that the output of the switch 72 is input to the DCT 76 when the sum of absolute differences in block> tolerance 1 or the maximum absolute difference in block> tolerance 2 is set. At times other than the conditions,
The switch 73 is switched so that the output of the switch 72 is not input to the DCT 76.

Further, the valid / invalid frame determining section 100 is
When the sum of absolute differences between blocks is added for one frame and the sum of absolute differences between frames is equal to or greater than a preset threshold value 3 or when the code amount stored in the input buffer 97 is equal to or greater than 4 threshold values. Further, when the rate control unit 102 determines that the transmission buffer 103 overflows, the switch 98 is controlled to be disconnected as an invalid frame, and in other cases, the switch 98 is controlled to be connected.

Further, the valid / invalid block judging section 75
The MPX unit 101 provides 1-bit information per block indicating whether the block determined by
Output to. Further, the valid / invalid frame determination unit 100 outputs 1-bit information per frame, which is the determined valid frame or invalid frame, to the MPX unit 101.

The 8 × 8 pixel original image output from the switch 73 or the inter-frame prediction error image (the difference data) is input to the DCT 76, subjected to discrete cosine transform, and input to the quantizer Q77. It The quantizer Q77 has two types of quantization matrices (previously stored in the quantization table 80 according to the determination by the intra-frame / inter-frame determination unit 74 (whether intraframe coding or interframe coding is performed)). Weights for each transform coefficient are arranged in a matrix) are selected and input, and the quantization scale from the rate control unit 102 is also input. The quantization matrix is multiplied by the quantization scale, the quantization step width of each transform coefficient is calculated, and the transform coefficient which is the output of the DCT 76 is quantized using this quantization step width.

The quantized transform coefficient is the variable length encoder V
It is input to the LC 78 and the inverse quantizer Q ^-1 79. In the variable length encoder VLC78, the Huffman code corresponding to the difference value from the previous block is output to the buffer 97 for the DC component of the transform coefficient, and the AC component is changed from the low frequency component to the high frequency component. For one-dimensionalized data by performing zigzag scanning toward the two-dimensional Huffman code for zero run and non-zero value immediately after that, the buffer 97
Is output to. Further, when zero continues to the end of the block, an EOB (End of Block) code is output to the buffer 97.

In the inverse quantizer Q ^-1 79, the quantizer Q
The same quantization matrix and quantization scale as in 77 are input, and inversely quantized transform coefficients are generated, and IDCT
81 and SVF (Space Variant Fil)
ter) 84. In the IDCT 81, the input transform coefficient is inverse discrete cosine transformed to generate a decoded data block. This data is stored in the buffer 82 for the set number of blocks. This set block number is SV
It is the number of block data required in F84. For example, when the SVF 84 filters only in the horizontal direction, it is sufficient to store 3 consecutive blocks in the horizontal direction, and when filtering in the horizontal and vertical directions, it is necessary to store 2 block lines + 1 block. When the valid / invalid block determination unit 75 sends a symbol of invalid block, 0 is stored in the block position for the number of block pixels.

Then, the processing block is input to the switch 83 from the decoded data accumulated in the buffer 82, and the filter is turned on by the control of the filter ON / OFF determination unit 96.
If it is determined that this is the case, this data is output to the SVF 84. If it is determined that the filter is off, the output is directly output to the switch 85.

In the SVF 84, the non-zero highest frequency component positions MHP and MVP in the horizontal and vertical directions of the inversely quantized transform coefficient value which is the input from the inverse quantizer Q ^-1 79.
Is detected (shown in FIG. 2), a plurality of low-pass filters (shown in FIG. 3) having different characteristics are selected in the horizontal and vertical directions using MHP and MVP as parameters, and this selection number is stored in a memory (not shown). )).

For the invalid block, the filter is selected with the highest frequency component position as the DC component. For the decoded data block input from the buffer 82 via the switch 83, the low-pass filter according to the selection number of the filter is applied to the block boundary and the inside thereof. Here, it goes without saying that the filter characteristics used for the block boundary and the pixels inside the block may be changed. .

The decoded data block for which this filter processing has been completed is output to the switch 85, and the adder 8 is provided only when the valid / invalid block determination unit 75 determines that the block is a valid block.
6 is input. In the adder 86, zero is selected when intraframe coding is selected by the intraframe / interframe determination unit 74, and when interframe coding is selected,
The decoded image predicted between frames of the previous frame is input by controlling the switch 95. The data input to the adder 86 are added, a decoded image of the currently encoded image is generated, and the contents at a predetermined position of the frame memory 87 (decoded image data of the previous frame) are rewritten. The original image data is stored in the frame memory 93 as it is.

The data stored in the frame memories 87 and 93 is output via the switch 94.
The control of the switch 94 is such that the valid / invalid frame determination unit 100 is connected to the frame memory 93 when it is determined to be an invalid frame, and is connected to the frame memory 87 when it is determined to be a valid frame. The output of this switch 94 is the switch 92, ME88, MC90.
Entered in.

The ME 88 receives the input original image data and the previous frame decoded image or the previous frame input original image stored in the frame memory 87 or the frame memory 93, which is the output of the switch 94, and performs motion detection. . Here, in the motion detection, with respect to the n × m pixel block (for example, 8 × 8 pixels) of the original image data, the decoded image data of the previous frame or the surrounding image of the same position as the original image data block of the original image data of the previous frame is determined. The range of d pixels is the search area (horizontal direction ± when used for ultrasonic diagnostic images.
The range is d pixels. ) About.

The criterion for determining the motion vector to be obtained is that the sum of absolute differences between the original image and the decoded image of the previous frame is minimum, and this sum of absolute differences is (sum of absolute differences at position 0 of motion vector). ) × α, and the motion vector is 0 if this condition is not satisfied. here,
α is a constant. Further, when obtaining up to a position of one pixel precision or less, an interpolation image for the position is created from the decoded image of the previous frame or the original image of the previous frame, and the sum of absolute differences is calculated in the same manner.

Furthermore, 1 is set for the motion vector of each block.
The frequency is calculated in the block line, and the motion vector having the maximum frequency other than zero is set as the representative motion vector in the block line (shown in FIG. 5). This representative motion vector is MC90 and M
It is output to the PX unit 101. Further, the difference absolute value sum storage memory 89 stores the difference absolute value sum at the motion vector 0 output from the ME 88.

Next, the block line is divided into a plurality of parts (however, the ultrasonic diagnosis image is divided into two), the motion vector frequency is obtained for each of the divided block lines, and the frequency distribution is further calculated. Recalculate the frequency again within the range of [−Δ, Δ], [representative, motion vector −Δ, representative motion vector + Δ]. Where Δ ≦ | representative motion vector / 2
Is a constant defined as |, and [-a, b] represents a region that includes -a and does not include b.

These two frequencies of each divided block line area are defined as H _o (n) and H _mv (n). Here, n is an index indicating the position of the divided block line (shown in FIG. 5). When these frequencies satisfy H _o (n) <H _mv (n), motion compensation in this divided block line is permitted, and the others are prohibited. This signal is output to the MC determination unit 91.

In the MC 90, according to the representative motion vector from the ME 88, the previous frame decoded image in the frame memory 87 or the previous frame input original image in the frame memory 93 is read, a motion compensation prediction image block is generated and output to the switch 92. . Also, the sum of absolute differences (MC error absolute sum) is calculated between the original image and the motion-compensated prediction image block, and is output to the MC determination unit 91. Further, the sum of absolute difference values (0) at the same block position is output from the sum of absolute difference value storage memory 89 to the MC determination unit 91.

In the MC judgment section 91, only when a signal from the ME 88 for permitting motion compensation in the divided block line is received, the sum of MC error absolute values <sum of difference absolute values (0) ×
It is determined that the block to be β is subjected to motion compensation, and the signals with and without motion compensation control the switch 92 and are output to the filter ON / OFF determination unit 96 and the MPX unit 101. Where β is a constant.

The switch 92 has the previous frame decoded image of the frame memory 87 output from the switch 94,
Or the previous frame input original image of the frame memory 93,
When the motion compensation prediction image from the MC 90 is input and the MC determination unit 91 determines that there is motion compensation, the MC 9
The motion compensation prediction image from 0 is the subtractor 71 and the switch 95.
When it is determined that there is no motion guarantee, the output of the switch 94 is output to the subtractor 71 and the switch 95.

Next, the filter ON / OFF judgment unit 96 can store the output from the MC judgment unit 91 for several blocks. The MC judgment of the block one block before and the MC of the current block are stored. When the condition is the same as the judgment, the switch 83 is switched so as to filter at the boundary between the two blocks. When the MC determination of the previous block and the MC determination of the current block are different, the switch 83 is switched so as not to apply the filter.

The buffer 97 also stores the variable-length coded data for one frame and outputs the data amount of the stored code to the valid / invalid frame determination section 100. Next MPX
In the unit 101, the quantization scale from the rate control unit 102, the valid / invalid block information from the valid / invalid frame determination unit 75, the compressed data stored in the buffer 97 via the switch 98, and M
The block line representative motion vector from E88 and the MC from the MC determination unit 91. None block information is input and these data are rearranged in a predetermined bit stream syntax and output to the transmission buffer 103 and the rate control unit 102. To do.

The bit stream accumulated in the transmission buffer 103 is output at a constant rate. Further, the rate control unit 102 counts the code amount of the bit stream output from the MPX unit 101, and predicts the quantization scale of the encoded frame next so that the generated code amount falls within the set bit rate. This prediction is performed independently during intraframe coding and interframe coding.

The prediction formula is as follows. Q _s (I) = (Next frame assigned code amount (I) / Current frame generated code amount (I)) ^1/2 × Q _s ′ (I) Q _s (P) = (Next frame assigned code amount (P) / Current frame generated code amount (P)) ^1/2 × Q _s ′ (P)

Here, Q _s ′ (I) and Q _s ′ (P) are the quantization scales used in the current intra-frame coded frame and the current inter-frame coded frame, and Q _s (I) and Q _s (P ) Represents the quantization scale used in the next intra-frame coded frame and the next inter-frame coded frame.

This quantization scale is input to the quantizer Q77 and the inverse quantizer Q ^-1 79, and the quantization step width used for the next frame is generated. The moving image is compressed by repeating the above processing for one frame. In this embodiment, since motion compensation prediction is performed between the input original image of the previous frame and the input image of the current frame immediately after the invalid frame or at the time of intra-frame encoding that occurs at a constant time period, this frame is used even in the case of an image having a large moving area. This has the effect of preventing the absolute value of the inter-difference difference (that is, the sum of the absolute values of the motion interpolation prediction errors) from increasing, and prevents the frame from becoming an invalid frame unexpectedly.

Next, FIG. 9 shows the configuration of a moving picture coding apparatus as a fourth embodiment according to the present invention. Here, the fourth embodiment is similar to the configuration of the third embodiment, and only the characteristic part will be described. In this video encoding device,
The decoded data output from the IDCT 131 stored in the buffer 132 is input to the SVF 134 via the switch 133 when the filter ON / OFF determination unit 146 determines that the filter is ON.

Further, in the SVF 134, the inverse quantizer Q ^-1 1
The conversion coefficient output from 29, the motion vector output from the ME 138, and the motion compensation ON / OFF information output from the MC determination unit 141 are input.

In the SVF134, first, the inverse quantizer Q
^-1 Maximum non-zero frequencies MHP and MVP in the horizontal and vertical directions of the transform coefficient output from 129 (shown in FIG. 2)
To calculate. A low-pass filter that acts on the decoded data using the MHP and MVP as parameters is selected from a plurality of types (shown in FIG. 3). That is, when the MHP is large (the maximum frequency is high), a low-pass filter having a wide horizontal frequency band is selected, and conversely, when the MHP is small (the maximum frequency is low), the horizontal frequency band is narrow. Select a low pass filter.

The MVPs are similarly selected in the vertical direction. Further, it is determined whether to select a filter having a narrower frequency band than the selected low pass filter based on the magnitude of the motion vector from the ME 138.
When the absolute value of the motion vector of the block to be motion-compensated is larger than the threshold value X, it is switched to a low-pass filter having a narrower frequency band than the selected filter (this performs the same processing both horizontally and vertically). ).

As a result, the distortion can be reduced by obtaining a larger band limitation for a block having a large amount of motion in which a larger coding distortion occurs. Moreover, the characteristic that a portion with a large amount of movement is less noticeable even if there is a little blur is used. As described above, according to the moving picture coding apparatus of the present embodiment, when it is determined that the block distortion is promoted, the block boundary (block line) is not filtered and the generated distortion is generated. Does not get worse.

Further, motion compensation is realized by motion compensation to make many of the block boundaries of the predicted image continuous, and the number of block boundaries that can be low-pass filtered immediately after the inverse orthogonal transform is greatly increased. Block distortion can be reduced. Then, at the block boundary that has been motion-compensated with a large motion vector, the filter characteristic is processed in a direction that further impairs the distortion. It should be noted that if the amount of movement is large, even if the blurring increases a little, there are few visual problems.

Furthermore, since the reference image for valid / invalid frame determination immediately after the invalid frame is the motion-compensated predicted image of the original image of the previous frame, the difference error can be reduced for a large motion, and it is inadvertently determined to be an invalid frame. Can be prevented. Therefore, the moving picture coding apparatus of the present invention suppresses the occurrence of block distortion and performs efficient compression. Further, the present invention is not limited to the above-described embodiments, and it goes without saying that various modifications and applications can be made without departing from the scope of the invention.

[0097]

As described in detail above, according to the present invention, it is possible to provide a moving picture coding apparatus that suppresses the occurrence of block distortion even in an image in which block distortion is prominent.

[Brief description of drawings]

FIG. 1 is a diagram showing a configuration of a moving picture coding apparatus as a first embodiment according to the present invention.

FIG. 2 is a diagram showing the relationship between the positions MHP and MVP of the highest frequency component of the inversely quantized transform coefficient value.

FIG. 3 is a diagram showing filter characteristics of a low-pass filter.

FIG. 4 is a diagram showing a configuration of a moving picture coding apparatus as a second embodiment according to the present invention.

FIG. 5 is a diagram showing a relationship between a frequency and a motion vector within a block line.

FIG. 6 is a diagram showing a relationship between a frequency and a motion vector within a block line divided into two.

FIG. 7 is a flowchart for explaining motion detection and motion compensation.

FIG. 8 is a diagram showing the structure of a moving picture coding apparatus as a third embodiment according to the present invention.

FIG. 9 is a diagram showing the structure of a moving picture coding apparatus as a fourth embodiment according to the present invention.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Subtractor, 2, 3, 13, 17, 18, 21 ... Switch, 4 ... In-frame inter-frame determination section, 5 ... Valid / invalid block determination section, 6 ... DCT, 7 ... Quantizer Q, 8 ... Variable-length encoder VLC, 9 ... Inverse quantizer Q ^-1 , 10 ... Quantization table, 11 ... IDCT, 12 ... SVF (Space V)
aliant filter), 14 ... Adder, 15,
16 ... Frame memory, 19 ... Buffer, 20 ... In-block difference absolute value sum calculation unit, 22 ... Valid / invalid frame determination unit, 23 ... MPX unit, 24 ... Rate control unit,
25 ... Transmission buffer.

Claims (5)

[Claims] 1. An input image signal continuously input in time is predicted between an input image of a current frame and a decoded image of a previous frame and subtracted for each image to generate an inter-frame prediction error signal. Means, means for dividing the prediction error signal or the input image signal into blocks of n × m pixels, performing orthogonal transform in block units, quantizing transform coefficients to perform variable length coding, and means for quantizing transform coefficients. In a moving picture coding apparatus comprising means for performing inverse quantization and inverse orthogonal transformation, adding with a predicted image from a previous frame decoded image, and generating a current frame decoded image, the transform coefficient for each block or this transform A filter selecting unit that selects one filter from a plurality of types of low-pass filters, using the information related to the coefficient as a parameter, and a filter selected by the selecting unit. Video encoding apparatus characterized by comprising a filtering means for applying a signal after inverse orthogonal transform, the. 2. The moving picture coding apparatus according to claim 1, wherein the filter selected by the filter selecting means has different characteristics according to prediction information between a current frame input image and a previous frame decoded image. And a filter changing unit for changing the signal, and a determining unit for determining whether or not to apply the selected filter to the signal after the inverse orthogonal transform according to the prediction information. Device. 3. An inter-frame prediction error signal is generated by performing prediction between an input image signal of a current frame and a decoded image of a previous frame with respect to an input image signal input continuously in time, and subtracting each image. Means for dividing the prediction error signal or the input image signal into blocks of n × m pixels, performing orthogonal transform in block units, quantizing the transform coefficient to perform variable length coding, and means for quantizing the transform coefficient. In a moving picture coding apparatus including means for generating a current frame decoded image by performing dequantization, performing inverse orthogonal transform and adding with a predicted image generated from a preceding frame decoded image, a motion vector for detecting a motion vector for each block is detected. Vector detection means and block line internal motion for obtaining the frequency of occurrence of a motion vector for each block line, by combining each block for one line into a block line Cuttle frequency calculating means, representative motion vector generating means for generating a representative motion vector in the block line based on the frequency obtained by the motion vector frequency calculating means in the block line, the block line is divided into a plurality of, Motion compensation determination means for determining whether to permit or prohibit motion compensation in the divided block line based on the motion vector generation frequency distribution for each divided block line, and a divided block for which motion compensation is permitted by the motion compensation determination means A moving picture coding apparatus, comprising: means for judging and controlling whether or not to perform motion compensation prediction using the representative motion vector for each block in a line. 4. The moving picture coding apparatus according to claim 2, wherein the information of the motion compensation prediction is ON / OFF information of motion compensation prediction in an adjacent block. 5. The moving picture coding apparatus according to claim 2, wherein the motion-compensated predicted picture of the original picture of the previous frame serves as a reference picture for valid / invalid frame determination immediately after the invalid frame. apparatus.