JPH0541861A - Moving picture encoding equipment - Google Patents

Abstract

PURPOSE:To realize the moving picture encoding equipment in which encoding is implemented without implementing motion detection with respect to a refer ence picture and moving compensation prediction to an interlace picture is implemented with high accuracy and less arithmetic operation quantity. CONSTITUTION:The coder is provided with a motion vector candidate generating circuit 21 which generates selectively plural motion vector candidates to designate a reference picture among encoded pictures and pictures generated by using the encoded pictures, an interpolation value generating circuit 19 generating an interpolation value between picture elements or in-pattern picture elements by using the picture element of the encoded picture of the plural patterns with the generating method in response to the motion vector candidate, a motion vector retrieval circuit 17 retrieving an optimum motion vector whose correlation between the reference picture designated by the motion vector candidate and an encoded object picture is maximized from the motion vector candidate and outputting the retrieved vector, and an encoding section 14 encoding the encoding object picture by using the optimum motion vector.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a moving picture coding apparatus for coding a moving picture signal such as a video signal in image communication, transmission, storage and broadcasting, and more particularly to motion compensation predictive coding. The present invention relates to a moving picture coding device.

[0002]

2. Description of the Related Art Video telephones, video conferences, optical disk devices,
Devices such as VTRs and CATVs require a technique for encoding moving image signals. As a coding method for such a moving image, a pixel value of an image to be coded (hereinafter referred to as a coded image) is used by using a pixel value of an already coded reference image specified by a motion vector. There is known a so-called motion compensation predictive coding which predicts and codes the prediction error and the motion vector.

When this motion compensation predictive coding is applied to an interlaced image (field image) such as an ordinary television video signal, a motion accuracy of 1/2 line accuracy or more in a field (motion of 1 line accuracy or more in a frame). There is a problem in that the accuracy cannot be met because the corresponding pixel value does not exist in the reference image.

Therefore, there has been proposed a method of performing motion compensation by interpolating the pixel values of pixels having no corresponding pixels on the reference image by using images of two adjacent fields (for example, 1990 Image Coding Symposium). (PCSJ9
0), 8-1, "Consideration of adaptive line interpolation inter-field motion compensation method"). In this motion compensation method, the encoding target image signal is encoded using the reference image signal and the optimum motion vector. The reference image signal is created by interpolation from the signal at the position designated by the motion vector obtained by the motion vector search circuit, out of the coded image signals of the past two fields. Specifically, three field memories are prepared, and the signal generated by the intra-field interpolation from the output of the first field memory and the output of the second field memory are mixed at a mixing ratio km: 1-km. Thus, a reference image signal is created. km varies depending on the amount of motion detected by the motion detection circuit based on the outputs of the first field memory and the third field memory.

FIG. 2 is a diagram showing the relationship between the image signals according to this conventional technique. The relationship between the image signals 41, 42, 43 and the encoding target image signal 44 accumulated in the three field memories is as shown in the figure, and the vertical component of the motion vector is n + 1/2 lines (n
Is an integer), the corresponding pixel value in the image signal 41 is output as it is as the reference image signal. For example, when n = 0, the pixel value of the reference image signal is 46 with respect to the pixel value 45 of the encoding target image signal 44. If the vertical component of the motion vector is n lines in the field, the image signal 4
Interpolation value △ using the signal ○ adjacent to the interpolation value △ of 2,43
To create. That is, for example, the interpolated value 49 which is the reference image signal for the pixel value 45 in the encoding target image signal 44 is
The value obtained by inter-field interpolation is the sum of the pixel value 46 and the pixel value 47 multiplied by km, and the pixel value 48 multiplied by (1-km). At this time,
If the motion detection circuit determines that the motion is large,
Since it is appropriate to interpolate the interpolated value Δ mainly by using the signal adjacent to the interpolated value Δ in the image signal of the same field as the interpolated value Δ, it is appropriate to set km to be large, and when it is determined that the motion is small, the interpolated value Δ is interpolated. Since it is appropriate to interpolate the value Δ mainly by using the signal adjacent to the interpolation value Δ in the image signal of the field adjacent thereto, km is set to be small.

As described above, the interpolated value is created by adapting the motion amount using the image signals of the adjacent two fields, so that even in the field image, it is 1/2 in the field.
It becomes possible to generate a reference image signal suitable for motion accuracy of line accuracy or more (motion accuracy of one line in a frame or more), and it is possible to perform accurate motion compensation prediction.

However, in this method, it is necessary to detect the motion in the reference image as described above, and the motion detection circuit for that is required. Also, in order to be able to detect motion, three adjacent fields must already be coded. Therefore, if the three adjacent fields are not already coded, there is a problem that motion detection cannot be performed.

FIG. 3 shows an example in which three adjacent fields have not been coded. In this example,
Special playback on VTR (random access, fast forward playback,
In consideration of adaptation to a storage system that requires reverse reproduction, etc., two adjacent fields (fields 1, 2, 7, in the example in the figure) for every multiple fields (every 6 fields in the figure)
8) is pre-encoded, and the remaining 4 fields (3 to 6 in the example in the figure) are pre-encoded. The neighboring 2 fields are used as a reference image for motion compensation predictive encoding. In such a case, the motion detection cannot be performed because the three adjacent fields have not been encoded.

On the other hand, in the moving picture coding apparatus using the motion compensation predictive coding method as described above, in the conventional case, the reference picture for searching the motion vector is searched for in the motion vector search for the forward or backward motion compensation. If the image to be encoded is a non-interlaced image, it is limited to only one image that has already been encoded.
There is a problem in that accurate motion compensation cannot be performed on a moving image in which a movement of / 2 pixels is present. If the image is an interlaced image, two consecutive screens that have already been encoded are used as described above, but since the same area is searched for on both screens in the past, the amount of calculation for searching is extremely large. Was increasing.

Further, among the moving picture coding methods as described above, a moving picture coding method targeting a transmission rate of about 1 to 2 Mbps is particularly used as a coding method for a storage medium such as a VTR or an optical disk. Is MPEG
It is being standardized under the standard name of 1. This is a system based on motion-compensated interframe prediction + DCT.

[0011] At present, for similar purposes, studies are underway on a method of encoding a moving image having a quality equal to or higher than that of TV broadcasting at about 2 to 10 Mbps. When trying to adapt the encoding method of MPEG1 to this, MPEG1 presupposes a non-interlaced image as an input image, whereas the TV signal is a signal of an interlaced image, so it is suitable for an interlaced image. New ideas are needed. In interlaced image coding, interfield / interframe adaptive prediction is known. This is a field whose scanning phase matches the field of interest (an odd field if the field of interest is an odd field, and an even field if the field of interest is an even field), but a field that does not match but is close in time (for example, if the field of interest is In the case of an odd field, an even field adjacent to the odd field, and when the field of interest is an even field, an odd field adjacent to the even field) is switched as a prediction signal. In addition, recent studies have also examined interpolative prediction that averages motion-compensated signals for multiple fields (for example, F. Wang et al., "High-quality co
ding of the even fields based on the odd fields of
interlaced video sequences ", IEEE trans.CS).

However, even in the interlaced image, the predictive coding using the past field and the future field as in the coding of MPEG1 is performed.
It is desirable to incorporate a prediction method using even / odd fields suitable for interlaced images. In this case, when the motion vector is sent to each field, the amount of information of the motion vector increases, so it is necessary to reduce the amount of information of the motion vector without reducing efficiency as much as possible. That is,
It is required to improve the prediction accuracy for interlaced images to reduce the amount of information in the coding output of prediction errors, and to minimize the increase in motion vector information. Conventionally, it is effective to meet such requirements. Technology has not been proposed yet.

[0013]

As described above, in the prior art, when the pixel at the point where the pixel on the reference image does not exist is interpolated using the images of the two fields adjacent to the reference image, the reference image moves. Since the detection is necessary, there is a problem that the motion detection circuit is required and the hardware is complicated, and the motion cannot be detected when the adjacent three fields are not already coded.

Further, in the prior art, when the image to be coded is a non-interlaced image, the reference image is limited to only one coded image in the motion vector search.
There is a problem that accurate motion compensation cannot be performed for moving images in which there is a movement of 1/2 pixel unit between adjacent images, and the encoding efficiency is reduced. Furthermore, the encoding target image is an interlaced image. In this case, the amount of calculation for the motion vector search is very large, which causes a problem that the motion vector search time becomes long and the hardware circuit scale increases.

Further, the conventional technique has a problem that the prediction accuracy for the interlaced image cannot be effectively improved and the amount of information of the motion vector sent to each field increases.

A first object of the present invention is to provide a moving picture coding apparatus capable of performing a motion-compensated prediction for an accurate interlaced picture with a small amount of calculation and performing coding without performing motion detection for a reference picture. To provide.

A second object of the present invention is to improve the coding efficiency by enabling accurate motion compensation even when there is a small motion between adjacent images, and to reduce the calculation amount for motion vector search. An object of the present invention is to provide a moving image coding apparatus that can be made smaller.

A third object of the present invention is to improve the prediction accuracy for an interlaced image to reduce the information amount of the coded output of the prediction error and to minimize the increase of the motion vector information.

[0019]

In order to achieve the first object of the present invention, a plurality of coded images for designating a reference image from among coded images and images generated using the coded images are provided. By using a motion vector candidate generating means for selectively generating motion vector candidates and a pixel value of encoded images of a plurality of screens, a pixel between those images or pixels between screens by a generation method according to the motion vector candidates Interpolation value generating means for generating an interpolated value between the motion vector candidates, and a motion vector search for searching and outputting from the motion vector candidates an optimum motion vector that maximizes the correlation between the reference image specified by the motion vector candidates and the encoding target image. It is characterized by comprising means and means for encoding an image to be encoded using an optimum motion vector.

The interpolation value generating means may be configured to continuously change the interpolation value according to the motion vector candidate, or may have a plurality of interpolation value generating circuits and at least one interpolation value generating circuit according to the motion vector candidate. It is also possible to adopt a configuration in which the interpolated value from is output selectively.

In order to achieve the second object of the present invention, the first encoded image temporally close to the image to be encoded is used as a reference image for motion compensation, and the object to be encoded is within a predetermined range of the reference image. First motion vector searching means for searching a motion vector designating the reference image or its partial image having the maximum correlation with the image or its partial image, and the image to be coded has a time longer than that of the first coded image. Second far away
Second motion vector searching means for searching a motion vector over a narrower range or area than the first motion vector searching means, and the first and second motion vectors as a reference image for motion compensation. From the motion vectors searched by the search means, the optimum motion vector designating the reference image or the partial image having the maximum correlation with the image to be coded or its partial image is selected, and the optimum motion vector and its optimal It is characterized by further comprising means for encoding a motion compensation error which is a difference between an image or partial image designated by a motion vector and an encoding target image or its partial image.

Here, in the second motion vector searching means, the center position of the search range is determined according to the motion vector searched by the first motion vector searching means, or the code of the image to be coded. It is defined in the vicinity of the digitized region.

In order to achieve the third object of the present invention, a first motion vector detecting means for detecting a motion vector between input images input in field units, and detection by the first motion vector detecting means. Prediction signal candidate generating means for generating a plurality of prediction signal candidates for the input image by performing spatiotemporal filtering on the reference image in the area specified by the selected motion vector, and a plurality of prediction signal candidates generated by this means And a second motion vector detecting means for detecting an optimum motion vector from the predicted signal candidate and the motion vector detected by the first motion vector detecting means. It is characterized by comprising information about the optimum motion vector detected by this means and an encoding means for encoding the prediction error of the prediction signal. It is.

More specifically, for example, in the prediction signal candidate generating means, a plurality of pixels in one field in the past used for prediction are subjected to spatial filter processing, and an output obtained by performing the spatial filter processing is less than one pixel. The interpolated pixel interpolated at the position of accuracy and the time-filtered output between the pixels of the other fields in the past used for prediction are generated as the plurality of prediction signal candidates, and the encoding means performs the spatial filter processing. The motion vector for the applied field is 1
The motion vector for the other fields is coded with a precision of less than a pixel, and the motion vector to be coded with a precision of one pixel or less is configured to encode the difference from the motion vector in the field.

[0025]

In general, there is a large correlation between the optimum motion vector that maximizes the correlation and the amount of motion in the reference image, and if the relative motion between the reference image and the image to be encoded is small. The probability that the motion is small in the reference image is extremely high, and conversely, if the relative motion between the reference image and the image to be encoded is large, the motion is also likely to be large in the reference image.

Therefore, in the present invention, for each motion vector candidate at the time of motion vector search, it is assumed that those motion vector candidates correspond to the actual motion amount in the reference image, and the interpolation corresponding to the motion amount is performed. A value is generated from an interpolation value generation circuit, and a reference image is configured using the interpolation value,
Perform a motion vector search. By doing this, even if the motion detection for the reference image is not performed, the motion vector search is finally performed under the reference image with the appropriate interpolation value, and the motion compensation for the interlaced image is performed with high accuracy. Is possible.

Further, in the present invention, when a motion vector is searched using a coded two-screen image as a reference image, a coded image that is temporally close to the coding target image is searched over a wide range, By performing a search for a coded image that is distant in time from the image to be coded over a more limited and narrow range, it is possible to perform accurate motion compensation while suppressing the calculation amount for motion vector search.

For example, by searching a coded image that is temporally close to the image to be coded, at least one motion vector candidate with good coding efficiency is obtained, and operations such as vector extension are performed based on the motion vector. By doing so, it is possible to limit the center of a region that is a candidate for a motion vector with good coding efficiency between the image to be coded and the coded images that are temporally separated. In this case, if the area selected as a motion vector candidate in a coded image that is close in time is coded using a motion vector between a coded image that is temporally distant, the time immediately It is possible to limit the range in which a region that is a candidate for a motion vector between an image that is distant from the target image and an image to be encoded exists.

When the motion between images is very large and there is no candidate motion vector even in a coded image that is close in time, or when the motion vector obtained as a candidate in a coded image that is close in time is very large. If it is possible to predict that there is no motion vector with high coding efficiency in a coded image that is significantly distant in time, do not search for a motion vector in a coded image that is distant in time. Further, the search range can be further limited. By doing so, it is possible to obtain a motion vector having a high computational efficiency, at least in the amount of calculation required for the search.

On the other hand, when there is a large motion between images that are close in time, there is a blur due to the motion even in one image captured by a normal camera. Therefore, when motion compensation is performed between images, 1 / It is difficult to make a difference in units of two pixels. That is, in such a case, even if the motion vector is obtained from images that are further apart in time, the improvement in coding efficiency is small despite the increase in the amount of calculation for searching. Therefore, by restricting the search range of motion vectors in temporally distant images to the vicinity of the coding region of the image to be coded, the amount of calculation for searching can be further reduced and high coding efficiency can be obtained. ..

Further, according to the present invention, the reference image in the area designated by the motion vector between the input images is subjected to the spatio-temporal filter processing to generate the prediction signal candidate, and the prediction signal is determined from these, so that the prediction signal is determined in field units. It is possible to make a prediction suitable for the input image input at, that is, the interlaced image, and the prediction accuracy is improved. In this case, the motion vector for the field to be spatially filtered is encoded with an accuracy of 1 pixel or less, but the motion vector for the other fields is encoded with the accuracy of 1 pixel or less. The information of the finally transmitted motion vector is reduced by encoding and transmitting the difference as the difference between the converted motion vector and the motion vector in the field.

[0032]

First, an embodiment according to claim 1 of the present invention will be described. FIG. 1 is a block diagram of a moving picture coding apparatus according to the first embodiment of the present invention. The moving picture coding apparatus of FIG. 1 includes a coding unit 14, two field memories 15,
16, motion vector search circuit 17, interpolation value generation circuit 19
And the local decoding unit 25.

The encoding target image signal 11 is encoded by the encoding unit 14 using the reference image signal 12 output from the interpolation value generating circuit 19 and the optimum motion vector 13 output from the motion vector searching circuit 17. It Reference image signal 12
Is the interpolation value generation circuit from the image signal at the position designated by the motion vector candidate 18 output from the motion vector search circuit 17 among the coded image signals of the past two fields stored in the field memories 15 and 16. 1
Created by 9.

The motion vector search circuit 17 comprises a correlation calculation circuit 20, a motion vector candidate generation circuit 21, an optimum vector determination circuit 22 and a switch 23, and the motion vector candidate generation circuit 21 sequentially generates during the motion vector search operation. The motion vector candidate 24 is input to the field memories 15 and 16 by the switch 23, and the correlation calculation between the reference image signal 12 and the encoding target image signal 11 generated by the interpolation value generation circuit 19 based on this is performed. It is performed by the circuit 20. The optimum vector determination circuit 22 stores the motion vector that maximizes the correlation between the encoding target image signal 11 and the reference image signal 12, and the optimum motion vector 13 is output by the switch 23 at the end of the motion vector search operation. The optimum motion vector 13 and the interpolation value generation circuit 19
The encoding target image signal 11 is encoded by the encoding unit 14 based on the optimum reference image signal 12 from, and the encoded data 26 is output.

The local decoding unit 25, based on the encoded data 26 output from the encoding unit 14, the locally decoded image signal 27.
To create. The locally decoded image signal 27 is input to either of the field memories 15 and 16 by the switch 28, and the outputs thereof are passed through the switch 29 and the interpolation value generating circuit 1 is output.
9 is input. Here, the switchers 28 and 29 are used to create a reference image which is predetermined for the image to be encoded.
It is switched so that the image signal is input to the interpolation value generation circuit 19.

The interpolated value generating circuit 19 comprises an intra-field interpolating circuit 30, a multiplier 31, a multiplier 32 and an adder 33, and a signal generated by the intra-field interpolating circuit 30 from the output signal from the field memory 15. The output signal from the field memory 16 is mixed with the mixing ratio k: 1-k.
The reference image signal 12 is obtained by mixing in.

The motion vector candidate 18 output from the motion vector search circuit 17 is also input to the interpolation value generation circuit 19 and controls the parameter k that determines the mixing ratio of the output signals from the field memories 15 and 16. That is, when the vertical component of the motion vector candidate 18 is n + 1/2 lines in the field (n is an integer), k = 1 is controlled, and the corresponding pixel value stored in the field memory 15 (here, the field memory 15 Will be output as it is as the reference image signal 12.

If the vertical component of the motion vector candidate 18 is n lines in the field, the field memory 1
Interpolation values in FIG. 2 using the output signals from 5 and 16
Is interpolated. At this time, if the absolute value of the vertical component of the motion vector candidate 18 is larger than a certain threshold value, the motion in the reference image is also regarded as large, and the interpolation value Δ is adjacent to the interpolation value Δ in the field memory 15 of the same field. The parameter k is set to a large value because it is appropriate to mainly use the signals to be interpolated. On the contrary, when the absolute value of the vertical component of the motion vector candidate 18 is smaller than the threshold value, the motion in the reference image is also regarded as small, and the interpolation value Δ is mainly used for the signal adjacent to the interpolation value Δ in the field memory. Since it is appropriate to interpolate, the parameter k is made small.

It should be noted that as shown in FIG.
4 and the reference image 41 are adjacent to each other, it is effective to bring the parameter k close to 0 and use the signal from the field memory 16 as the interpolation value of the reference image 41.
It is almost limited to the case where the absolute value of the vertical component of the motion vector candidate 18 is 0. On the other hand, when the encoding order as shown in FIG. 3 is adopted, as shown in FIG.
3 and the reference image 51 may be relatively separated from each other. In such a case, it is effective to use the signal from the field memory 16 as the interpolation value of the reference image 51 by setting the parameter k close to 0 even when the absolute value of the vertical component of the motion vector is not 0. come.

FIG. 5 shows a second embodiment of the present invention. Since the configuration is the same as that of FIG. 1 except for the configuration of the interpolation value generation circuit 19,
Detailed description thereof will be omitted. The interpolated value generation circuit 19 is composed of two interpolated value generation circuits 34 and 35, and a switch 36 for selecting the output of these. Interpolation value generation circuit 3
Reference numerals 4 and 35 denote the interpolation value generation circuit 19 shown in FIG.
Similarly to the above, the intra-field interpolation circuit 30 and the multipliers 31, 3
2 and adder 33. FIG. 6 shows the relationship between the image signals in this embodiment.

The first interpolated value generation circuit 34 uses the reference image 6
When the motion amount in 1 is large, an effective interpolation value is generated. That is, as shown in FIG. 6A, when the vertical component of the motion vector candidate 18 is n + 1/2 lines in the field (n is an integer), the corresponding pixel value ∘ stored in the field memory 15 (here Then, the description will be made assuming that the content of the field memory 15 is an image close to the encoding target image 63) is output as it is as a reference image signal (controlled to k = 1).

When the vertical component of the motion vector candidate 18 is n lines in the field (n is an integer), the first interpolation value generating circuit 34 stores in the field memory 15 adjacent to the pixel Δ to be interpolated. An interpolation value Δ is created by averaging two pixel values ◯ of the image signal 61.

The second interpolated value generation circuit 35 uses the reference image 6
When the amount of movement at 1 is small, an effective interpolation value is generated. That is, as shown in FIG. 6B, when the vertical component of the motion vector candidate 18 is n + 1/2 lines in the field (n is an integer), the corresponding pixel value ∘ stored in the field memory 15 (here: , Field memory 1
The description will be made assuming that the content of 5 is an image close to the encoding target image 63) is output as it is as a reference image signal (controlled to k = 1). For example, when n = 0, the reference image signal for the encoding target image signal 64 is the pixel value 65.
Becomes

When the vertical component of the motion vector candidate 18 is n lines in the field, the second interpolation value generation circuit 35 sets the pixel ⊚ to be interpolated as the signal ∘ in the adjacent field memory 16. For example, when n = 0, the interpolation value 68 of the reference image signal with respect to the encoding target image signal 64 is
Has a pixel value of 66.

When the vertical component of the motion vector candidate 18 is n / 2 + 1/4 lines in the field, the pixel value ◯ from the field memory 15 adjacent to the pixel Δ to be interpolated and the pixel value from the field memory 16 An interpolated value Δ is created by averaging with ○. For example, when n = 0,
The interpolation value 67 of the reference image signal with respect to the encoding target image signal 64 is an average value of the pixel value 65 and the pixel value 66.

As described above, the interpolation value generating circuit 35 can generate a reference image signal with an accuracy of a unit of 1/4 line in the field, and enables effective motion compensation for an image with few motions and high resolution. To do.

Here, when the absolute value of the vertical component of the motion vector candidate 18 output from the motion vector search circuit 17 is larger than a certain threshold value, the switch 36 selects the output of the interpolation value generation circuit 34 and sets the threshold value. Interpolation value circuit 3 if smaller
By selecting the output of 5, an appropriate reference image signal is generated for each. As another method, the motion vector candidate 18 output from the motion vector search circuit 17
When the absolute value of the vertical component of is large, the output of the interpolating value circuit 34 is selected, and when it is small, both outputs of the interpolating value generating circuits 34 and 35 are used as reference image signals, and correlation calculation is performed for both. The method of selecting either one is also effective.

Next, an embodiment according to claim 2 of the present invention will be described. FIG. 7 is a block diagram of a moving picture coding apparatus according to the third embodiment of the present invention.

In FIG. 7, the image data input to the input terminal 001 is temporarily stored in the input buffer memory 100, and then, in accordance with the order of the image to be encoded, for each partial area composed of a plurality of pixels. It is read from the partial image data input buffer memory 100. The partial image data read from the input buffer memory 100 is input to the motion vector detection circuit 200. The motion vector detection circuit 200 obtains a partial image from which the input data can be efficiently coded from the previously coded reproduced images, and the motion vector (address data) indicating the data of the partial image region and the position of the region. Is output.

The partial image data output from the input buffer memory 100 is also input to the local encoding circuit 300 together with the partial image data and the motion vector output from the motion vector detecting circuit 200. The local encoding circuit 300 encodes either the partial image data itself output from the input buffer memory 100 or the difference data from the partial image data specified by the motion vector. Here, the encoded data of the difference from the area designated by the motion vector also includes data obtained by variable-length encoding the motion vector.

The data coded by the local coding circuit 300 is input to the local decoding circuit 400 together with the partial image data output from the motion vector detection circuit 200 and decoded, whereby a reproduced image is obtained. When the encoded data is encoded using the motion vector, the decoding result is added to the partial image data output from the motion vector detection circuit 200 to obtain the reproduced image.
This reproduced image data is input to the motion vector detection circuit 200 and is temporarily stored for encoding the image data input next.

Next, a concrete operation example of the motion vector detection circuit 200, the local coding circuit 300 and the local decoding circuit 400 will be described.

First, in the motion vector detection circuit 200, the data input from the input buffer memory 100 is an image memory (any one of 211 to 214) in which image data unnecessary for the motion vector search is stored.
Are sequentially written by the write control circuit 222. The coded image data stored in the image memory (211 to 214) in this manner is sequentially read by the read control circuit 221 and the data switching circuit 231 from the image temporally close to the coded image for each area. , Difference circuit 241
The difference from the input data is calculated for each area. The evaluation circuit 242 sequentially compares the magnitudes of the total values of the differences for each area, controls the direction in which the read control circuit 221 searches the image memory, and inputs the partial image rather than the previously detected partial image. Each time a partial area of an encoded image having a smaller difference from is detected, an address indicating the area of the partial image is stored in the vector register 243, and the partial area of the encoded image closest to the input partial image is stored. Ask for. The address data indicating the partial area of the encoded image having the smallest difference from the input partial image stored in the register 243 in this way is input to the read control circuit 233 and the switching circuit 232. The reproduced image of the encoded data for the partial area of the completed image is read from the reproduced image memory (215 to 218) and input to the local encoding circuit 300 together with the address data.

On the other hand, in the local encoding circuit 300,
In this embodiment, DCT (discrete cosine transform), which is one of orthogonal transforms, and quantization and variable length coding are used as a motion compensation error coding method. Local encoding circuit 300
First, the partial image data output from the input buffer memory 100 is input to the difference circuit 311, and the partial image data and the partial image data reproduced from the motion vector detection circuit 200 are reproduced. The difference is calculated. The switching circuit 312 includes the difference circuit 31.
The differential image data input from 1 and the partial image data input from the input buffer memory 100 are sequentially switched by the control signal input to the terminal 002 and output.

The DCT circuit 320 sequentially frequency-converts and outputs the partial image data and the difference image data sequentially output from the switching circuit 312. The quantizer circuit 330
The frequency-converted data output from the CT circuit 320 is quantized and output with a preset quantization width. The entropy encoder 340 collectively encodes the data quantized by the quantization circuit 330, the quantization width information and the identification code of the partial image data and the difference data, and further encodes the difference image data by a vector. The motion vector for the partial image output from the register 243 is also variable-length coded and output using a Huffman code or the like according to the respective appearance probabilities. If one Huffman code is formed by combining the identification code and the motion vector, efficient coding is possible. In addition, in this encoding, for the image data obtained by reproducing the encoded data of the area specified by the predetermined rule, or the input image data whose difference from the fixed data is the set value or less, If the number of consecutive partial images is variable-length coded, the coding efficiency is further improved.

The code amount evaluation circuit 351 determines the code amount when the difference between the partial image to be encoded and the area specified by the motion vector is encoded and when the input data is directly DCT encoded. The comparison is performed, and the coded data having the higher coding efficiency is output to the output buffer 360 and the local decoding circuit 400.

The output buffer 360 temporarily stores the coded data in order to adjust the output data rate, and the quantization width used in the quantization circuit 330 and the coding used in the entropy encoder 340. Control the table.

In the local decoding circuit 400, the partial image data output from the motion vector detection circuit 200 is temporarily stored in the data memory 441, and the encoded data output from the code amount evaluation circuit 351 is a variable length decoder. The motion vector including the identification code and the quantized data before encoding are input to 410 and decoded. The decoded quantized data is input to the inverse quantization circuit 420, converted into a representative value having a dynamic range before quantization (inverse quantization), and then input to the addition circuit 450. The data inversely quantized by the inverse quantization circuit 420 is the inverse DCT circuit 43.
0 is input, and partial image data or difference image data is reproduced. The gate circuit 442 passes the partial image data output from the data memory 441 when the reproduction data output from the inverse DCT circuit 430 is difference image data by the identification code decoded by the variable length decoder 410, If it is not a difference image, the output data is set to zero and output to the addition circuit 450.

When the image data subjected to the inverse DCT in this way is the encoded difference image, it is added to the partial image data output from the motion vector detection circuit 200, and when it is not the difference image, the motion vector detection is performed. Circuit 2
A reproduction image can be obtained from the adding circuit 450 without using the partial image data output from 00. This reproduced image data is input to the motion vector detection circuit 200 and is temporarily stored for encoding the image data input next.

FIG. 8 shows a fourth embodiment of the present invention.
The difference is that a data memory 460 for storing quantized data is used instead of the variable length decoding circuit 410 included in the encoding circuit 400 of FIG. In this case, data obtained by performing DCT and quantization on the difference image data between the partial image data output from the input buffer memory 100 and the reproduced image specified by the motion vector for the partial image data is the data memory 4
It is temporarily stored in 60. Then, the code amount evaluation circuit 352
The image data corresponding to the encoded data selected by the output buffer 360 is output to the inverse quantization circuit 420. When this image data is difference image data, a reproduced image is obtained by adding the partial image data output from the data memory 441 in the addition circuit 450, as in the example of FIG.

This embodiment has an advantage that the processing time is shorter than that in FIG. 1 since the calculation for decoding is unnecessary.

FIG. 9 shows a fifth embodiment of the present invention, which is an example in which a part of the encoding circuit 400 is used for a motion vector detecting circuit. Also in this embodiment, the motion vector detection circuit 200 switches the partial image data from the already encoded and reproduced image according to the read control circuit 224.
The data is read via 32 and a motion vector (address data) indicating the position of the area is output.

In FIG. 9, the image data input to the input terminal 001 is temporarily stored in the input buffer memory 100 as in FIGS. 7 and 8, and then is composed of a plurality of pixels according to the order of the image to be encoded. The local encoding circuit 30 is read from the input buffer memory 100 for each region
Input to 0. The partial image data output from the input buffer memory 100 is not different from the output data from the motion vector detecting circuit 200 input via the gate circuit 313 by the control signal input to the input terminal 002 first. DCT, quantization, and variable length coding are performed as in FIGS. 7 and 8. The quantized data is stored in the data memory 460, and the encoded data and the code amount are stored in the code amount evaluator 353. Next, input terminal 002
The control signal input to is changed, the difference from the partial image data output by the motion vector detection circuit 200 is calculated, and DCT, quantization, and variable length coding are performed as in FIGS. 7 and 8. ..

The code amount evaluator 353 evaluates the encoded data from the entropy encoder 340 and controls the read control circuit 224 according to the evaluation result, so that the code amount evaluator 353 detects an area where the code amount can be further reduced. The encoded data is stored and the motion vector detection circuit 200
The partial image data output from the data memory 441 is stored in the data memory 441, and the data obtained by DCT and quantizing the difference image is stored in the data memory 460. In this way, the encoded data having the smallest code amount is finally output to the output buffer 360. The quantized data corresponding to this coded data is converted by the local decoding circuit 400 in FIGS.
Is played as well. The reproduced data is input to the motion vector detection circuit 200 and is temporarily stored for encoding the image data input next.

In the embodiments of FIGS. 7 and 8, a more accurate motion vector is obtained as the optimum motion vector, whereas in the embodiment of FIG. 9, a motion vector that maximizes the coding efficiency is obtained.

FIG. 10 shows an example of a circuit for reproducing the data coded in the embodiment shown in FIGS. The variable length decoder 510 decodes the coded data input from the input terminal 004, and includes quantization width information, motion vector (differential image data with respect to the partial image designated by the motion vector, and an identification code indicating whether or not it is different). ), And quantized data. The quantized data is the inverse quantization circuit 52.
0, reproduced through the inverse DCT circuit 530. If the reproduction data is the difference image data with the partial image specified by the motion vector, the read control circuit 521 reads the partial image data from the reproduction image memories 611 to 614 and outputs it from the motion vector detection circuit 600. Input to the decoding circuit 500. This partial image data is added to the difference image data by the addition circuit 550 via the gate circuit 540 to become reproduced image data. This reproduced image data is input to the motion vector detection circuit 600 and is temporarily stored in order to reproduce the encoded data input next. The reproduced image data is also input to the output buffer 560, returned to the original order of the images, and output.

Next, the motion vector search operation performed in the evaluation circuit 242 and the read control circuit 221 of FIGS. 7 and 8 and the code amount evaluator 352 and the read control circuit 224 of FIG. 9 will be described.

11 to 16 show examples of motion vector search according to the present invention. 11 to 16
, S1, s2, ..., s6 are 1 frame or 1
Means an image of a field, 101, 102, ..., 12
0 means one pixel or a plurality of pixels in the horizontal or vertical direction.

In the motion vector search example shown in FIG. 11, the entire encoded image s3 (10) that is temporally close to the partial image to be encoded (eg, s4-104) is first displayed.
1-120) are searched, and the optimum motion vector obtained by the search, that is, the coding efficiency is high or the more accurate motion vector (for example, s3-107) has been coded in the coding target image. The search range (area) in the coded image s2 that is temporally distant from the image s3 is limited to (105 to 115), for example.

In the search in the encoded image s2 which is distant in time, the optimum motion vector (for example, s02-109 or s03-107) including the previously calculated optimum motion vector (for example, s03-107) is included. ).

Then, the difference between the image of the area designated by the obtained optimum motion vector (for example, s2-107) and the partial image to be encoded (for example, s4-104), that is, the motion compensation error is obtained, and the difference is obtained. Encode the optimal motion vector and motion compensation error.

The motion vector search examples shown in FIGS. 12 to 14 are images (for example, s) that are temporally close to the encoding target image s4.
This is a suitable example when 3) is encoded using a motion vector between a temporally distant image (for example, s1). In this case, since it is possible to predict the motion vector between the partial image s4-104 to be encoded and the image s3 that is close in time, the image s3 that is close in time.
The search range in is limited to (103 to 112), for example. Further, if a motion vector (for example, s3-107) having a high coding efficiency with the temporally close image is obtained by this, a partial region of the temporally close image designated by the motion vector (for example, s3-107). 107) using the motion vector (for example, s1-110) between the temporally distant image (for example, s1) used for 107) and the temporal separation of the partial image s4-104 of the encoding target image s4. The area that is a candidate for the motion vector with respect to the image (for example, s2) can be limited to a narrower range, for example, (107 to 110).

In the motion vector search example shown in FIG. 15, when the motion between images is very large and there is no motion vector that is a candidate even in the image s3 that is close in time, or when the image s3 that is close in time is obtained as a candidate. This is an example of the case where the obtained motion vector (for example, s3-116) is very large, and it can be predicted that there is no motion vector with high coding efficiency in the temporally distant images s2.

In such a case, images s2 that are temporally separated
For, the motion vector search may not be performed, or the search range may be further limited, such as (116 to 120). This makes it possible to obtain a motion vector with high coding efficiency even if the amount of calculation required to search for the motion vector is small.

In the motion vector search example shown in FIG. 16, the image s3 temporally close to the partial image s4-104 to be coded is used.
This is an example of the case where the motion vector (for example, s3-107) obtained in step S1 is large to some extent, or an appropriate motion vector is not obtained. In these cases, the search for the motion vector in the temporally distant image s2 is not performed, and the search in the temporally distant image s2 is performed only when the motion vector between the temporally distant image s3 is small. The range is in the vicinity of the same position as the coding region of the coding target image s4 (for example, 103
Up to 105) and search for a motion vector.

When there is a large motion between images that are close in time, there is a "blur" due to motion even in one image captured by a normal camera. It is difficult to make a difference in pixel units. That is, in such a case, even if the motion vector is obtained from images that are further apart in time, the improvement in coding efficiency is small despite the increase in the amount of calculation for search. According to the example of FIG. 16, in such a case, since a motion vector of an image having a large motion is not encoded, the types of encoding are reduced, and the encoding efficiency is further increased.

The encoding method in the present invention may be intra-frame, intra-field encoding, or inter-frame or inter-field differential encoding, and may be combined with other encoding methods. Further, the motion vector search method according to the present invention similarly encodes when a motion vector is obtained for an image between them using a coded image previously encoded at an arbitrary number of image intervals as a reference image. The search can be performed before and after the target image.

Next, an embodiment according to claim 3 will be described. FIG. 17 is a block diagram of a moving picture coding apparatus according to the sixth embodiment of the present invention. This embodiment is based on MPEG1.
This is an example in which the present invention is applied to an encoding method such as SM3 which is a simulation model of the above. The encoding algorithm basically adopts the motion compensation + DCT (discrete cosine transform) method. However, the input image format is, for example, CCIR Rec. It is an interlaced format such as the image format defined in 601/525.
This image format is shown in FIG. In FIG. 18, Y is a luminance signal, Cr and Cb are color signals, and the number of pixels per field of each signal is as shown in the figure.

Similar to SM3, the coding unit is hierarchically structured. That is, as shown in FIG. 19, blocks, macroblocks, slices, pictures, and a group of pictures and sequences, which are not shown in FIG. 8x blocks
It consists of 8 pixels, and DCT is performed in this block unit. The macroblock is composed of two Y blocks and one Cr block and one Cb block, for a total of 4 blocks. Motion compensation and selection of each coding mode are performed in units of this macroblock.

The group of pictures (GOP) is constructed as follows. Like SM3, the picture is roughly divided into I, P, and B pictures according to the type of mode for each macroblock allowed as a prediction mode. As modes, there are four modes of intra-field prediction (Intra), forward prediction (Inter: including motion compensation), backward prediction, and bidirectional interpolation prediction (details will be described later). Of these, the types of pictures are classified into three types, I, P, and B, as shown in Table 1, depending on which prediction mode can be used.

[0081]

[Table 1] In the present embodiment, unlike SM3, an interlace format is adopted as a coded picture format, and even pictures of the same type have different prediction methods depending on their positions in the GOP, so pictures are further classified. FIG. 20 shows the structure of GOP and from which picture each picture is predicted. As shown in the figure,
A GOP is defined by a set of pictures that start with a B ₀ picture that precedes an I picture that is periodically provided as an entry point for random access and trick play, and that ends with P ₂ that precedes the next I picture. I-pictures appear only in even fields.

Table 2 and FIG. 21 show how each picture is predicted from which picture.

[0083]

[Table 2] There are two types of prediction methods: inter-field prediction in which only even fields are predicted, and inter-field / inter-frame adaptive prediction in which each even / odd field is adaptively predicted. In FIG. 21, an arrow indicates that inter-field prediction is performed, and a symbol in which two lines are integrated into one arrow indicates that inter-field / inter-frame adaptive prediction is performed. The coding order of each picture in one GOP is as shown in FIG.

Based on the above points, the moving picture coding apparatus shown in FIG. 17 will be described. In FIG. 17, the input terminal 70
An interlaced moving image signal is input to 0.
The input moving image signal is stored in the field memory 701 for eight consecutive fields. Frame memory 701
The moving image signal in the inside is supplied to the first motion vector detecting circuit 710, and the motion vector is detected with one pixel accuracy by a telescopic search (described later) using the original moving image. This is the first stage process of motion compensation.

Next, as preprocessing in the second stage of motion compensation, the motion vector detection circuit 7 is used by using the locally decoded signal stored in the field memory 708 in the local decoding loop.
The motion vector obtained using the original image in 10 is refined by searching the range of ± 1 around it.

Next, as the second stage of motion compensation, the field memory 708, the inter-field / inter-frame adaptive prediction circuit 709, and the second motion vector detection circuit 711.
At 1/2, motion vector detection with 1/2 pixel precision is performed using the locally decoded signal, and the prediction circuit 709 generates a prediction signal. This prediction signal is input to the subtractor 702 and the difference from the moving image signal from the field memory 701 is calculated, and this difference is output as the prediction error signal.

This prediction error signal is discrete cosine transformed by the DCT circuit 703 to obtain DCT coefficient data. The DCT coefficient data is quantized by the quantizing circuit 704 and adaptively scanned, and then the two-dimensional variable length coding circuit 7
It is input to the multiplexer 714 via 10. The quantized DCT coefficient data is locally decoded through the inverse quantization circuit 705 and the inverse DCT circuit 706, and only I and P pictures are written in the field memory 708. The field memory 708 is prepared for four fields required for adaptive prediction.

The multiplexer 714 multiplexes the DCT coefficient data, the motion vector from the second motion vector detecting circuit 711, which will be described later, and the additional information such as the macroblock type from the coding control unit 717 and the step size. .. The information multiplexed by the multiplexer 714 passes through the buffer 715 to have a constant transmission rate, and is output to a storage system (recording medium) such as a VTR.

In the encoding control unit 717, the buffer amount in the buffer 715 and the activity calculation circuit 71
By using the intra-macroblock activity (I picture) calculated in step 6 or the intra-macroblock activity (P, B picture) of the immediately preceding signal in the same mode before quantization, the quantization step size in the quantization circuit 704 is set. Have control.

Next, the configuration of a moving picture decoding apparatus corresponding to the moving picture coding apparatus of FIG. 17 will be described with reference to FIG. The signal read from the storage system is input to the input terminal 800 and temporarily stored in the buffer 801. Buffer 801
The signal read from is input to the demultiplexer / variable length decoding circuit 801, and the multiplexer 71 of FIG.
Additional information such as DCT coefficient data, motion vectors and step sizes multiplexed in 4 is separated and decoded.

That is, after the DCT coefficient data is two-dimensionally variable-length decoded and scan-converted, the inverse quantization circuit 8 is used as in the local decoding loop in the moving picture coding apparatus of FIG.
03 and the inverse DCT circuit 804 and locally decoded to adder 805 and field memory 8 for 4 fields.
It is input to the adaptive prediction circuit 807 via 06. On the other hand, since the motion vector is sent in the form of a difference as described later, after the variable length decoding, it is returned to the original form and supplied to the adaptive prediction circuit 807. A prediction signal is generated in the adaptive prediction circuit 807, and this prediction signal and the locally decoded signal from the inverse DCT circuit 804 are added by the adder 805,
The original moving image signal is taken out via the field memory 806. The output of the field memory 806 is supplied to an image display unit (not shown).

However, regarding the display of images, the decoding order and the display order are different, and since B pictures are not used for prediction, it is not necessary to store them in the field memory.
The I and P pictures are output from the field memory 806, and the B pictures are output as they are while being decoded.

Next, the motion compensation and motion vector detection operation in this embodiment will be described in detail. Motion compensation
As described above, it is performed in macroblock units. The motion vector detection is performed by the 1 in the first motion vector detection circuit 710.
For each point where motion vector detection by normal block matching with pixel accuracy and adaptive spatio-temporal interpolation at pixel positions with 1/2 pixel accuracy around the reference image specified by this 1-pixel accuracy motion vector , Motion vector detection in the second motion vector detection circuit 711 by searching for a motion vector. A specific method of spatiotemporal interpolation will be described later.

The motion vector detection in the first motion vector detection circuit 710 is a process of finding the optimum motion vector by a full search for each field with one pixel accuracy using each field image of the input moving image. Telescopic search (SM
3) is used. However, in the field image, a plurality of paths can be considered as the search path between the fields. In this embodiment, the search route is determined according to the following rules. 1) Search between in-phase fields This is performed using only in-phase fields. 2) Search between opposite phase fields Use fields with the same phase as much as possible.

Although it is necessary to include the search between the opposite phase fields only once in the path, in this embodiment, as shown in the order of the telescopic search in FIG. 24, the fields of different phases are searched first. I have to. For example I
→ When obtaining the motion vector of P ₂ , I → P _o → B ₁ →
Search in the order of B ₃ → P ₂ and I → B ₀ → B ₂ → P ₁ → P
Routes like ₂ do not pass. The range indicated by hatching in FIG. 24 is the search range. This telescopic search is performed separately for the forward direction and the backward direction. The search range between adjacent fields is ± 15 pixels horizontally and ± 7 pixels vertically.

Motion vector detection in the second motion vector detection circuit 711 is performed using the image stored in the field memory 708 in the local decoding loop as a reference image. First, as the pre-processing of this motion vector detection, the first
The motion vector is refined by fully searching the range of ± 1 pixel around the reference image designated by the motion vector obtained by the motion vector detection circuit 710 of FIG. Second
The main process of motion vector detection in the motion vector detection circuit 711 is a plurality of prediction signals generated by a method described later at a position of 1/2 pixel accuracy around the reference image specified by the motion vector obtained by the refinement. This is performed by selecting the most suitable prediction signal candidate by evaluating and comparing all prediction error powers of the candidates.

Motion vector detection is not performed for the color signal, and motion compensation is performed based on the motion vector obtained from the luminance signal.

Next, the processing of the inter-field / inter-frame adaptive prediction circuit 709 of FIG. 17 will be described. Figure 23
9 is a block diagram showing a partial configuration of an inter-field / inter-frame adaptive prediction circuit 709. FIG.

As described above, the second stage of motion compensation is a search for a range of 1/2 pixel accuracy around the reference image designated by the motion vector obtained by the first motion vector detection circuit 710. This is shown in FIG. 25 (b). This inter-field / inter-frame adaptive prediction is based on P ₁ , P ₂ , B
In each of the pictures, for example, a pair # 1 and # 2 of each preceding even-odd one field is used as a reference image (reference field) for the encoding target image (encoding field).

In the first stage of motion compensation shown in FIG. 25A, the motion vector detecting circuit 710 determines the optimum points (indicated by ●) in the reference fields # 1 and # 2, respectively. Suppose that it was obtained in. Figure 25
In the second stage of the motion compensation shown in (b), 1/2 of the circumference of the two optimum points designated by the motion vectors V1 and V2.
An inter-field / inter-frame adaptive prediction circuit 709 applies a spatio-temporal filter to a reference image in a pixel-precision region to obtain a plurality of prediction signal candidates, and based on the prediction signal candidates and the motion vector detected by the motion vector detection circuit 710. ,
The motion vector detection circuit 711 searches for the optimum motion vector. In this case, there are two modes of generating a prediction signal candidate, a field interpolation mode and a frame interpolation mode.

First, in the field interpolation mode, the prediction signal candidate is generated only by the spatial filter among the spatiotemporal filters. That is, the pixel values of the pixels A to C in FIG. 25B are created by A = (O + D) / 2 B = (O + D + E + F) / 4 C = (O + F) 2. In this field interpolation mode, there are a total of 18 prediction signal candidate search points, 9 points in each of the even and odd fields.

On the other hand, in the frame interpolation mode, the prediction signal candidate is created by subjecting the signal subjected to motion compensation with one pixel precision in each field to the processing of space-time filter. For example, the pixel values of the pixels A to C in FIG. 25B are created by A = G / 2 + (O + D) / 4 B = G / 2 + (O + D + E + F) / 8 C = G / 2 + (O + F) 1/4 It In this case, the field on the side which provides the data of the pixel position with the 1/2 pixel precision is referred to as a reference field. Also in this frame interpolation mode, there are nine prediction signal candidates in each of the even and odd fields, but the prediction signals at the pixel O positions match, so there are a total of 17 search points for prediction signal candidates.

In the second stage of motion compensation, the sum of the two is 3
The prediction signal candidates of 5 points are searched and the prediction signal candidate having the smallest prediction error is determined as the prediction signal. However, if the direction of the motion vector pointing to both fields is largely deviated, the frame interpolation mode is not selected (details will be described later).

Information indicating which of the field interpolation mode and the frame interpolation mode is selected and information indicating which field is selected as the reference field used in the field interpolation mode or the reference field in the frame interpolation mode are 1-bit information. It is transmitted by a flag. Regarding the prediction by the field immediately before P ₀ and P _2, the prediction is performed only by the mode in which the field interpolation mode is applied to a single field. In this case, the mode selection flag is not transmitted.

The above description is for the P picture, but the same is true for the B picture. However, in the case of a B picture, as the search field in the field interpolation mode and the reference field in the frame interpolation mode, only fields having the same phase as the encoded field are selected. Of course, in this case, the flag indicating which field is selected is not transmitted.

Next, with reference to FIG. 23, how the above-mentioned principle is actually realized on hardware will be described. Cache memories 901a, 901b and 902
In a and 902b, of the image signals from the field memory 708 for four fields of the local decoding loop of FIG. 17, the portion designated by the motion vector obtained by the preprocessing of the second motion vector detection is stored. It Switching circuit 900
Distributes the outputs of the cache memories 901a and 901b to the temporal filter 903 and the spatial filter 905 according to the control signal from the second motion vector detection circuit 711,
The outputs of the cache memories 902a and 902b are distributed to the temporal filter 904 and the spatial filter 906.

In accordance with a control signal from the second motion vector detection circuit 711 indicating which of the field interpolation mode and the frame interpolation mode is to be selected, the selector 907 receives the signal passing through only the spatial filter 905, the spatial filter 905 and the time filter. Either of the signals that have passed through both filters 903 is selected. The same applies to the selector 908. For example, from the switching circuit 900 to the temporal filter 903.
If the signal of the pixel G of FIG. 25 (b) is input to the pixel and the signal of the pixel B of FIG. 25 (b) is output from the spatial filter 905, the signals of the pixel G and the pixel B are output to the output of the temporal filter 903. An averaged signal, that is, a signal subjected to spatiotemporal filtering is obtained. Therefore, the output of the spatial filter 905 may be selected in the field interpolation mode, and the output of the temporal filter 903 may be selected in the frame interpolation mode.

The outputs of the selectors 907 and 908 are directly input to the selector 911, added by the adder 909 and halved by the divider 910, and then input to the selector 911. The selector 911 uses these three inputs, that is, the signal predicted using the reference fields # 1 and # 2 from the selector 907, the signal predicted using the reference fields # 3 and # 4 from the selector 908, and the divider 91.
Signals predicted using reference fields # 1 to # 4 starting from 0 are selected and output to the second motion vector detection circuit 711 as predicted signal candidates. Motion vector detection circuit 71
In No. 1, the prediction signal candidate having the smallest prediction error is determined as the prediction signal from the 35 prediction signal candidates as described above, and the information indicating that is returned to. As a result, the inter-field / inter-frame adaptive prediction circuit 709
Outputs the prediction signal instructed by the motion vector detection circuit 711 to the subtractor and adder 707.

Next, the second motion vector detection circuit 711.
How to encode and send the motion vector from the above will be described. First, the motion vector in the field interpolation mode or the motion vector on the reference field side in the frame interpolation mode is coded by the variable length coding circuit 713 as a reference vector and sent to the multiplexer 714. This motion vector (reference vector) has 1/2 pixel precision. When the frame interpolation mode is selected, the motion vector of the field other than the reference field and the motion vector of the reference field (reference vector)
Is similarly encoded and transmitted. That is, the difference between the extension line of the motion vector in the reference field, the closest point to the intersection of the fields other than the reference field, and the motion vector is 1
It shall be sent with pixel accuracy. It is considered that the frame interpolation mode is not effective except when the directions of these two motion vectors are close to each other, and the frame interpolation mode is not selected when the value of the difference exceeds ± 1.

FIG. 26 shows a specific example of the method of sending the motion vector. The motion vector of 1/2 pixel precision of the reference field # 1 (indicated by the arrow on the upper side in the figure) and the reference field # 2. The difference (the arrow extending in the vertical direction in the figure) between the point ● closest to the intersecting point Δ and the 1-pixel precision motion vector of the reference field # 2 (indicated by the lower arrow in the figure) is determined with the 1-pixel accuracy. send. In the example of FIG. 26, the difference is -1. By doing so, it is possible to save the information amount of the motion vector without degrading the performance of the re-prediction.

Next, motion compensation of color signals will be described.

As shown in FIG. 19, the luminance signal Y and the color signals Cr and Cb have the same number of pixels in the vertical direction in one macroblock, but the number of pixels of the color signal becomes half in the horizontal direction. ing. Therefore, when the motion vector obtained from the luminance signal is applied to the color signal, the horizontal component of the motion vector is halved. The division by ½ is performed by rounding an integer or less in the direction of 0. This is the same in the field interpolation mode and the frame interpolation mode.

27 (a) and 27 (b) show specific examples in the case of obtaining the motion vector of the color signal from the motion vector of the luminance signal in the field interpolation mode and the frame interpolation mode. Circles shown by dotted lines in FIG. 27 are pixel positions where no color signal exists. The pixel position obtained in the first step is indicated by a circle in the center, and the point obtained in the second step is indicated by x. In this example, the horizontal coordinate of the origin is larger than the horizontal coordinate of the circle point in the center in any case. In the case of the field interpolation mode of FIG. 27 (a), ×
When the motion vector of the luminance signal is obtained at the position, the 1/4 pixel precision of the color signal is rounded in the A direction to form the interpolated pixel Δ. This interpolated pixel Δ is created by Δ = 1/4 (A + B + C + D). Also for the reference field # 1 in the frame interpolation mode of FIG. 27B, the interpolation pixel Δ in the reference field # 1 is similarly created by Δ = 1/4 (E + F + G + H).

As for the interpolated pixel in the first reference field # 2, the pixel at the I position is used by rounding the 1/2 pixel precision of the color signal in the A direction. After all, in the case of the example of FIG. 27, the value of the interpolation pixel Δ is obtained by I × 1/2 + (E + F + G + H) × 1/8.

[0115]

As described above, according to the present invention, it is possible to perform highly accurate motion compensation predictive coding on an interlaced image without performing motion detection on a reference image. Therefore, not requiring a motion detection circuit not only reduces the circuit scale, but also reduces, for example, VT.
It is possible to apply motion compensation predictive coding even when using a coded sequence in which motion detection cannot be performed in order to support R special reproduction.

Further, according to the present invention, it is possible to perform accurate motion compensation and obtain high coding efficiency while reducing the amount of calculation required for motion vector search.

Further, according to the present invention, it is possible to effectively use a large number of reference fields to perform efficient and highly accurate prediction suitable for interlaced images, and increase motion vector information accompanying this. Can be minimized.

[Brief description of drawings]

FIG. 1 is a block diagram showing a first embodiment.

FIG. 2 is a diagram showing a relationship between image signals of respective screens in the related art.

FIG. 3 is a diagram showing an example in which three adjacent fields are not already coded.

FIG. 4 is a diagram showing a relationship between image signals of respective screens in the first embodiment.

FIG. 5 is a block diagram showing a second embodiment.

FIG. 6 is a diagram showing a relationship between image signals of respective screens in the second embodiment.

FIG. 7 is a block diagram showing a third embodiment.

FIG. 8 is a block diagram showing a fourth embodiment.

FIG. 9 is a block diagram showing a fifth embodiment.

FIG. 10 is a block diagram of a video decoding device.

FIG. 11 is a diagram showing an example of motion vector search according to the present invention.

FIG. 12 is a diagram showing an example of motion vector search according to the present invention.

FIG. 13 is a diagram showing an example of motion vector search according to the present invention.

FIG. 14 is a diagram showing an example of motion vector search according to the present invention.

FIG. 15 is a diagram showing an example of motion vector search according to the present invention.

FIG. 16 is a diagram showing an example of motion vector search according to the present invention.

FIG. 17 is a block diagram of a moving picture coding apparatus according to a sixth embodiment of the present invention.

FIG. 18 is a diagram showing an input image format in the embodiment.

FIG. 19 is a diagram showing a hierarchical structure of coding units in the embodiment.

FIG. 20 is a diagram showing the structure of a group of pictures and the coding order in the embodiment.

FIG. 21 is a diagram for explaining a prediction method for each picture in the embodiment.

22 is a block diagram of a moving picture decoding apparatus corresponding to the moving picture coding apparatus of FIG.

23 is a block diagram of the inter-field / inter-frame adaptive prediction circuit in FIG.

FIG. 24 is a diagram showing the order of telescopic search in the embodiment.

FIG. 25 is a diagram for explaining an inter-field / inter-frame adaptive prediction process in the embodiment.

FIG. 26 is a diagram showing an example of how to send a motion vector in the embodiment.

FIG. 27 is a diagram showing a specific example of obtaining a motion vector of a color signal from a motion vector in the embodiment.

[Description of Codes] 14 ... Encoding unit 17 ... Motion vector search circuit 19 ... Interpolation value generation circuit 20 ... Correlation calculation circuit 21 ... Motion vector candidate generation circuit 22 ... Optimal vector determination circuit 200, 600 ...
Motion vector detection circuit 221, 223, 224, 621 ... Image data read control circuit 242, 351, 352, 353 ... Evaluation circuit 300 ... Local encoding circuit 400 ... Local decoding circuit 101 ... Field memory 103 ... DCT circuit 104 ... Quantization Circuit 105 ... Inverse quantization circuit 106 ... Inverse DCT circuit 108 ... Field memory 109 ... Adaptive prediction circuit 110 ... Motion vector detection circuit 111 ... Motion vector detection circuit 112 ... Variable length coding circuit 113 ... Variable length coding circuit 114 ... Multiplexer 115 ... Buffer 116 ... Activity calculation circuit 117 ... Code control unit

 ─────────────────────────────────────────────────── ─── Continuation of front page (72) Hideyuki Ueno 1 Komukai Toshiba-cho, Sachi-ku, Kawasaki-shi, Kanagawa Inside the Toshiba Research Institute Co., Ltd. (72) Inventor Noboru Yamaguchi Komukai-shi Toshiba, Kawasaki-shi, Kanagawa No. 1 in Toshiba Research Institute, Inc. (72) Inventor Tomio Yamakage No. 1 Komukai Toshiba-cho, Saiwai-ku, Kawasaki City, Kanagawa Incorporated Toshiba Research Institute

Claims (4)

[Claims] 1. A motion vector candidate generating means for selectively generating a plurality of motion vector candidates for designating a reference image from an encoded image and an image generated using the encoded image. Interpolation value generating means for generating an interpolation value between pixels between the images or pixels within the screen by a generation method according to the motion vector candidate using pixel values of encoded images of a plurality of screens; Motion vector search means for searching and outputting an optimum motion vector having the maximum correlation between a reference image specified by a vector candidate and an encoding target image from the motion vector candidate, and the encoding target image being the optimum motion vector And a means for encoding by using the moving picture encoding apparatus. 2. A first coded image temporally close to the image to be coded is used as a reference image for motion compensation, and a predetermined range of the reference image has a maximum correlation with the image to be coded or a partial image thereof. First motion vector searching means for searching for a motion vector designating a reference image or a partial image thereof, and second encoding which is farther in time from the first encoded image to the encoding target image. Second motion vector searching means for searching a motion vector over a narrower range or area than the first motion vector searching means, using the completed image as a reference image for motion compensation, and these first and second motion vector searching means The optimum motion vector that specifies the reference image or its partial image that has the maximum correlation with the target image or its partial image from the motion vectors searched by And a means for encoding the difference between the optimum motion vector and the image or partial image designated by the optimum motion vector and the encoding target image or the partial image thereof. Device. 3. A first motion vector detecting means for detecting a motion vector between input images input in field units, and a reference to an area designated by the motion vector detected by the first motion vector detecting means. Prediction signal candidate generating means for generating a plurality of prediction signal candidates for an input image by performing spatio-temporal filtering on an image, and a prediction signal for determining a prediction signal by comparing a plurality of prediction signal candidates generated by this means Determining means, second motion vector detecting means for detecting an optimum motion vector from the predicted signal candidate and the motion vector detected by the first motion vector detecting means, and information on the optimum motion vector detected by this means And a coding means for coding a prediction error of the prediction signal. 4. The prediction signal candidate generating means outputs one of a plurality of pixels in one field in the past used for prediction, which has been subjected to spatial filter processing, and one output which has been subjected to this spatial filter processing.
An output obtained by performing temporal filtering between an interpolated pixel interpolated at a position of accuracy equal to or less than a pixel and a pixel of another past field used for prediction is generated as the plurality of prediction signal candidates, and the encoding means is a spatial filter. The motion vector for the field to be processed is encoded with an accuracy of 1 pixel or less,
The motion vector for other fields is coded as a difference between a motion vector in the field and a motion vector encoded with the precision of 1 pixel or less and a motion vector in the field. 3. The moving picture coding device according to 3.