JPH1198508A - Method for detecting motion vector - Google Patents

Abstract

PROBLEM TO BE SOLVED: To improve encoding efficiency by means of dual prime prediction and field prediction by permitting referring picture elements required for searching the reference motion vector of dual prime prediction and a frame movement vector to be common. SOLUTION: A motion compensating means 140 interpolates the picture elements at every field screen so as to generate then reference movement vector detecting referring picture elements of dual prime prediction through the use of the referring picture elements which are read, based on reference vectors for searching the reference motion vector of dual prime prediction and the frame motion vector with respect to an encoding object block supplied from a motion vector candidate detecting means 130. Then, the reference motion vector of a 1/2 picture element unit in dual prime prediction is detected. Besides, the referring picture elements for detecting the frame motion vector are generated after picture element interpolation as the frame screen and the frame motion vector of the 1/2 picture element unit is detected. The access speed of a referring screen memory is reduced and the reference motion vector is detected with high precision.

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 used for recording, communication, transmission and broadcasting of a moving picture signal, and more particularly to a partial area of a screen to be coded from which position on a reference screen. The present invention relates to a motion vector detection method for detecting a motion vector indicating whether a motion vector has come.

[0002]

2. Description of the Related Art Conventionally, in motion vector detection in dual prime prediction, a frame motion vector and a top motion vector are searched in a primary motion vector search performed at intervals of one pixel or more.
In a secondary motion vector search performed by detecting a field motion vector and a bottom field motion vector at an interval of 1/2 pixel, re-search is performed for each motion vector, and three motions obtained by the re-search are further performed. Based on the motion vector that minimizes the error among the vectors, the minute motion vector dmv of the dual prime prediction
And a prediction method that minimizes a prediction error is selected from three types of prediction, that is, prediction using a frame motion vector, prediction using two field motion vectors, and dual prime prediction (see JP-A-09-65340).

However, in the detection of the bottom field motion vector in the primary motion vector search, the motion vector for the bottom field of the coded picture is detected from the top field of the reference picture whose temporal inter-picture distance is long. The motion vector for the bottom field is not optimized because it is hard to detect or it is not detected. Even if the coding efficiency is improved by the optimal field prediction without this restriction, the suboptimal prediction method is used. Therefore, there is a problem that the coding efficiency is reduced.

[0004] Further, since the reference motion vector of the dual prime prediction is substituted by a frame motion vector or a field motion vector of 1/2 pixel precision, there is a problem that the utilization efficiency of the dual prime prediction is not so much improved.

Further, in this secondary motion vector search,
If the reference motion vector used for dual prime prediction is a field motion vector, the reference field for the other field for the reference motion vector is insufficient, so that
The reference region for frame prediction (18 pixels × 18 pixels), the reference region for field prediction (18 pixels × 10 pixels × 2), and the reference region for small motion vector dmv detection of dual prime prediction (18 pixels × 10 pixels × 2). In addition, there is a problem that a reference area (18 pixels × 10 pixels) that is insufficient for detecting a reference vector in dual prime prediction is required, and the access speed of the reference screen memory is increased.

[0006]

As described above, in order to improve the utilization efficiency of the dual prime prediction, the reference screen for the field motion vector detection has been limited in the past. There was a problem to say.

Further, when the reference motion vector of the dual prime prediction is a field motion vector, there is a problem that the access to the reference screen memory must be performed at high speed in order to detect the minute motion vector dmv of the dual prime prediction. Was.

Furthermore, since the reference motion vector of the dual prime prediction is substituted by a frame motion vector or a field motion vector with half-pixel accuracy, there is a problem that the utilization efficiency of the dual prime prediction is not so much improved. .

[0009] It is an object of the present invention to eliminate the restriction on the reference screen in the field motion vector detection in the primary motion vector search and to improve the coding efficiency by dual prime prediction and field prediction.

The access speed of the reference screen memory is reduced by overlapping the reference region of the reference motion vector for dual prime prediction with the reference region for frame prediction or the reference region for detecting a minute motion vector dmv for dual prime prediction. It is in. Another object of the present invention is to detect a dual prime reference motion vector with high accuracy.

[0011]

SUMMARY OF THE INVENTION According to the present invention, there is provided a motion vector detecting system for detecting a motion vector for generating a reference image for an input image from a partial image of a previously encoded image. In the detection method, a first motion vector detection step of searching for a motion vector at intervals of a predetermined pixel unit or more, and a portion of the input screen from an odd field forming a reference screen detected in the first motion vector detection step On the basis of at least one of a motion vector for an odd field forming an area and a motion vector for an even field forming a partial area of the input screen from an even field forming a reference screen, an inter-frame prediction code in units of 1/2 pixel. Motion vector used in the encoding method and the reference when performing the dual prime prediction encoding method The second of searching the motion vector of
Characterized by a motion vector detecting step.

[0012] In order to achieve the above object, the present invention provides
A first motion vector detection step of searching for a motion vector at intervals of one pixel or more, and a motion vector for frame prediction, field prediction and dual prime prediction based on the motion vector obtained in the first motion vector detection step. In the motion vector detecting method for detecting an optimal motion vector for a partial area of a current coded screen from a previous coded screen as a reference screen by a second motion vector detecting step of re-searching in units of １ ／ pixel, The reference motion vector and the frame motion vector for the dual prime prediction are searched for in units of 1/2 pixel with reference to the field motion vector from the in-phase reference field obtained in the motion vector detection step. As a result, the reference pixels required for searching for the reference motion vector and the frame motion vector for dual prime prediction are shared, the access speed of the reference screen memory is reduced, and the reference motion vector for dual prime prediction is detected with high accuracy. .

[0013] In order to achieve the above object, the present invention provides
If at least one of the field motion vectors from the in-phase reference field obtained in the first motion vector detection step is equal to or less than a predetermined value, the first motion vector detection step uses a dual Search for a reference motion vector for prime prediction with half-pixel accuracy, and search for a minute motion vector dmv for dual prime prediction in half-pixel units based on the resulting reference motion vector with half-pixel accuracy. I do. As a result, the common range of the reference pixel necessary for searching for the reference motion vector of the dual prime prediction and its minute motion vector dmv is secured, the access speed of the reference screen memory is reduced, and each motion vector of the dual prime prediction is increased. Detect with accuracy.

The present invention also achieves the above-mentioned object.
When both of the field motion vectors from the in-phase reference field obtained in the first motion vector detection step are equal to or larger than a predetermined value, the search for each motion vector of the dual prime prediction in the second motion vector detection step is performed. By prohibiting, the access speed of the reference screen memory is reduced.

[0015] In order to achieve the above object, the present invention provides
When both of the field motion vectors from the in-phase reference field obtained in the first motion vector detection step are equal to or larger than a predetermined value, the search for each motion vector of the dual prime prediction in the second motion vector detection step is performed. By prohibiting and increasing the number of search points of the motion vector of the frame prediction in the second motion vector detection step, a highly accurate frame motion vector is detected within the allowable range of the access speed of the reference screen memory. To improve the coding efficiency.

According to the present invention, in the first motion vector detection step, when a field motion vector candidate for each field of the motion vector detection target block is detected, a search for a field motion vector candidate from in-phase reference fields is performed. This is performed before searching for a field motion vector candidate from a reference field having an opposite phase. Thereby, in the second motion vector detection step, detection of in-phase field motion vector candidates used for searching for a reference motion vector for dual prime prediction is obtained as an intermediate result of detection of each field motion vector candidate,
Eliminates the need for a dedicated detection step for reference motion vector candidates in dual prime prediction.

[0017]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 shows a flowchart of the motion vector detection method according to the first embodiment of the present invention.

First, processing step 1 is a first motion vector detection step of searching for a motion vector at intervals of one pixel or more, and comprises a field motion vector from the top field of the reference screen to the top field of the coding block. Processing step 11 for detecting a candidate, processing step 12 for detecting a motion vector candidate for the top field of the coded block from the top field and the bottom field of the reference screen, and processing of the coded block from the bottom field of the reference screen. A processing step 13 for detecting a field motion vector candidate for the bottom field and a processing step 14 for detecting a motion vector candidate for the bottom field of the coding block from the bottom field and the top field of the reference screen. It is configured.

The processing step 11 is performed at the top of the reference screen.
The sum of the absolute value of the difference between the field and the top field of the coding block (matching error) is sequentially calculated, and the position where the matching error is minimized is determined by the field from the in-phase reference field with respect to the top field of the coding block. It is detected as a motion vector candidate.

The processing step 12 is a step of
The matching error between the field and the top field of the coding block is sequentially calculated, and the position where the matching error is minimized using the matching error detected in the processing step 11 as an initial value is determined by the field motion vector for the top field of the coding block. Detect as a candidate.
That is, in the processing step 12, when the position smaller than the matching error detected in the processing step 11 does not exist in the bottom field of the reference screen, the field motion from the top field of the reference screen detected in the processing step 11 If a vector candidate is detected and a position smaller than the matching error detected in processing step 11 is present in the bottom field of the reference screen, a field motion vector candidate from the bottom field of the reference screen is detected.

The processing step 13 is performed at the bottom of the reference screen.
The matching error between the field and the bottom field of the coding block is sequentially calculated, and the position where the matching error is minimized is detected as a field motion vector candidate from the in-phase reference field with respect to the bottom field of the coding block.

The processing step 14 is performed at the top of the reference screen.
The matching error between the field and the bottom field of the coding block is sequentially calculated, and the position where the matching error is minimized using the matching error detected in the processing step 13 as an initial value is determined by the field motion vector for the bottom field of the coding block. Detect as a candidate.
That is, in the processing step 14, when the position smaller than the matching error detected in the processing step 13 does not exist in the top field of the reference screen, the field motion from the bottom field of the reference screen detected in the processing step 13 If a vector candidate is detected and a position smaller than the matching error detected in the processing step 13 exists in the top field of the reference screen, a field motion vector candidate from the top field of the reference screen is detected.

In the processing step 2a, the smaller one of the candidate field motion vectors from the in-phase reference fields detected in the processing steps 11 and 13 is used to determine the smaller one of the matching errors between the reference motion vector and the frame motion vector of the dual prime prediction. It is determined as a motion vector candidate to be used for a search reference point in units of 1/2 pixel.

The processing step 3a is a second motion vector detection step of searching for a motion vector at intervals of 1/2 pixel based on the motion vector candidates detected in the processing step 1. It comprises a processing step 31 for detecting a motion vector, a processing step 32 for detecting a frame motion vector, a processing step 33 for detecting a field motion vector, and a processing step 34 for detecting a minute motion vector dmv for dual prime prediction. I have.

In processing step 31, processing step 2a
Based on the motion vector candidates determined in the above, to create a reference region pixel-interpolated reference screen for each field, sequentially calculate the matching error between each field of the encoding block and the reference region of the in-phase field, The position at which the matching error is minimized is detected as a reference motion vector with half-pixel accuracy of dual prime prediction.

In processing step 32, processing step 2a
Based on the motion vector candidates determined in step (1), a reference area is created by interpolating pixels with the reference screen as one screen, and a matching error (frame prediction error) between the coding block and the reference area is sequentially calculated. The position where the error is minimized is detected as a frame motion vector with half-pixel accuracy.

In processing step 33, processing step 12
Based on each of the field motion vector candidates detected in steps 14 and 14, a pixel-interpolated reference area is created for each reference field, and the matching error between each field of the coding block and each reference area is calculated separately and sequentially. The position where the matching error (field prediction error) for each field of the encoded block is minimized is detected as a 1/2 pixel-accurate field motion vector for each field of the encoded block.

In processing step 34, processing step 31
Based on the motion vector (scaling motion vector) obtained by scaling the reference motion vector of 1/2 pixel precision of the dual prime prediction detected in the above, the reference region of the reference field of the opposite phase is determined for each field of the coding block. , Dmv with pixel interpolation for each reference field
Create a reference area for detection. Further, a difference block is created by subtracting each pixel value of the reference block determined in the processing step 31 from the encoded block in which each pixel value is doubled,
The absolute value sum (dual prime prediction error) of the value obtained by subtracting each pixel value of the reference area for dmv detection from this difference block and halving the value is sequentially calculated, and the position of the reference area where the dual prime prediction error is minimized Is detected as a small motion vector dmv for dual prime prediction.

The processing step 4a is a processing step for determining an optimal prediction method for the coded block, and calculates a frame prediction error when using the frame motion vector detected in the processing step 3a and two field motion vectors. The sum of the field prediction error when used, the dual prime prediction error when using the reference motion vector and the small motion vector dmv for dual prime prediction, and the prediction error when the prediction signal is set to 0 are compared, and the minimum prediction is performed. An error prediction method is detected, and a prediction method and a motion vector used for encoding are determined.

In the first embodiment, when the coding block size is 16 pixels × 16 pixels, the frame prediction reference pixels (18 pixels × 18 pixels) are necessary for detecting the reference vector in the dual prime prediction. Since it is included in the reference pixels (18 pixels × 10 pixels × 2), in the second motion vector detection step, the reference pixels for field prediction (18 pixels × 10 pixels × 2) and the minute motion vectors dmv for dual prime prediction are detected. Reference pixel (18 pixels x 10 pixels x
Even if 2) is added, the access of the reference screen memory required for each coding block is 18 pixels × 10 pixels × 6, which is about 6/7 of the conventional. However, the reference motion vector of the dual prime prediction is also more than the conventional. Can be detected accurately.

In this embodiment, in-phase field motion vector candidates are used as frame motion vector candidates, but the prediction error is smaller than in a frame motion vector search based on opposite-phase field motion vector candidates. In many cases, the effect on coding efficiency is small.

In the first motion vector detecting step, when detecting a field motion vector candidate for each field of the motion vector detection target block, the search for the field motion vector candidate from the in-phase reference field is performed by the opposite phase reference. In the second motion vector detection step, in-phase field motion vector candidates used for searching for a reference motion vector for dual prime prediction are detected before searching for a field motion vector candidate from a field. This is obtained as an intermediate result of detection of vector candidates, and eliminates the need for a detection step dedicated to reference motion vector candidates for dual prime prediction.

FIG. 2 shows a flowchart of the motion vector detecting method of the first embodiment according to the second embodiment of the present invention. First, processing step 1 is a first motion vector detection step of searching for a motion vector at intervals of one pixel or more, and performs the same processing as in the first embodiment.

In the processing step 2b, the smaller of the matching errors among the field motion vector candidates from the in-phase reference fields detected in the processing steps 11 and 13 is determined as one of the reference motion vectors of the dual prime prediction.
Determined as a motion vector candidate to be used as a search reference point in units of / 2 pixels. Further, the processing step 2b determines the smaller one of the field motion vector candidates detected in the processing steps 12 and 14 as a motion vector candidate to be used as a search reference point in units of 1/2 pixel of the frame motion vector. I do.

The processing step 3b is a second motion vector detection step of searching for a motion vector at intervals of a half pixel based on the motion vector candidates detected in the processing step 1. A processing step 30 for determining whether or not to search for a motion vector with a / 2 pixel accuracy; a processing step 31 for detecting a reference motion vector for dual prime prediction; processing steps 32 and 36 for detecting a frame motion vector;
It comprises processing steps 33 and 37 for detecting a field motion vector, and processing step 34 for detecting a minute motion vector dmv for dual prime prediction.

In processing step 30, it is checked whether or not the magnitude of the sum of the absolute values of the horizontal and vertical components of the reference motion vector candidate for dual prime prediction is equal to or smaller than a predetermined value (for example, 20 pixels per frame pixel). In the case of, the process proceeds to a processing step 31 for searching for a motion vector with a half-pixel accuracy of dual prime prediction. Proceed.

Note that the processing contents of the processing steps 31 to 37 are the same as those of the first embodiment, so that the description will be omitted. Processing step 4b is a processing step for determining an optimal prediction method for the coded block, and includes a frame prediction error when using the frame motion vector detected in processing step 3b and a frame prediction error when using two field motion vectors. Of the field prediction error of the reference, the reference motion vector and the small motion vector dmv of the dual prime prediction
Is compared with the dual-prime prediction error when using (except when dual-prime search is not performed) and the prediction error when the prediction signal is set to 0, and the prediction method that minimizes the prediction error is detected. To determine the prediction method and the motion vector to be used.

In the first example according to the second embodiment, the access to the reference picture memory required for each coding block in the second motion vector detection step is performed for a motion having a small effect of dual prime prediction. This is a large coding block, which is a reference pixel for frame prediction (18 pixels × 18 pixels) and a reference pixel for field prediction (18 pixels × 10 pixels × 2), which is about 4/7 of the prior art. In a certain coding block, about の of the reference pixels (18 pixels × 10 pixels × 2) necessary for detecting the reference vector of the dual prime prediction are used as the reference pixels (18 pixels × 18) of the minute motion vector dmv detection of the dual prime prediction.
10 pixels × 2), and a frame prediction reference pixel (1
Since one half of (8 pixels × 18 pixels) is included in the field prediction reference pixels (18 pixels × 10 pixels × 2), it is about 6/7 of the conventional one, and the reference motion vector of the dual prime prediction is more accurate than the conventional one. Can be detected.

FIG. 3 shows a flowchart of a motion vector detecting method according to a second embodiment of the present invention. First, processing step 1 is a first motion vector detection step of searching for a motion vector at intervals of one pixel or more, and performs the same processing as in the first example according to the second embodiment.

In processing step 2c, the smaller of the matching errors among the field motion vector candidates from the in-phase reference fields detected in processing steps 11 and 13, is determined as one of the reference motion vectors of the dual prime prediction.
Determined as a motion vector candidate to be used as a search reference point in units of / 2 pixels.

Processing step 2c is the same as processing step 1
Of the field motion vector candidates detected in 2 and 14, the one with the smaller matching error is used as a first motion vector candidate to be used as a search reference point in units of 1/2 pixel of the frame motion vector, and the other is used as the first frame motion vector candidate. 1
It is determined as a second motion vector candidate used for a search reference point in units of / 2 pixels.

The processing step 3c is a second motion vector detection step of searching for a motion vector at an interval of a half pixel based on the motion vector candidate detected in the processing step 1. Processing step 30 for determining whether or not to search for a motion vector with a / 2 pixel precision, processing step 31 for detecting a reference motion vector for dual prime prediction, and processing steps 32, 35, 36 for detecting a frame motion vector. And 38, processing steps 33 and 37 for detecting a field motion vector, and processing step 34 for detecting a minute motion vector dmv for dual prime prediction.

In processing step 30, it is checked whether or not the sum of the absolute values of the horizontal and vertical components of the reference motion vector candidate for dual prime prediction is equal to or smaller than a predetermined value (for example, 20 pixels per frame pixel). In the case of, the process proceeds to a processing step 31 for searching for a motion vector with a half-pixel accuracy of dual prime prediction. Proceed.

Note that the processing contents of the processing steps 31, 33, 34 and 37 are the same as those of the first example according to the second embodiment, so that the description will be omitted. Processing step 3
In step 5, based on the second frame motion vector candidate determined in processing step 2c, a frame motion vector having half-pixel accuracy as in processing step 32 of the first example according to the second embodiment. Perform a search.

In processing steps 32 and 36, based on the first frame motion vector candidate determined in processing step 2c, as in processing step 32 of the first example according to the second embodiment, 1 / A frame motion vector search with 2-pixel accuracy is performed.

In the processing step 38, based on the motion vector 0, a frame motion vector search with 1/2 pixel precision is performed as in the processing step 32 of the first example according to the second embodiment.

Processing step 4c is a processing step for determining an optimal prediction method for the coded block. If a dual prime search is performed, processing step 3c
Frame prediction error when using the frame motion vector detected in step 2, the sum of the field prediction errors when using two field motion vectors, the reference motion vector for dual prime prediction and the dual prime when using the small motion vector dmv The prediction error and the prediction error when the prediction signal is set to 0 are compared, the prediction method with the minimum prediction error is detected, the prediction method and the motion vector used for encoding are determined, and the dual prime search is not performed. The frame prediction error when the first candidate frame motion vector detected in the processing step 3c is used,
A frame prediction error when using a candidate frame motion vector, a frame prediction error when using a frame motion vector near the motion vector 0, a sum of field prediction errors when using two field motion vectors, and a prediction signal Are compared with each other, a prediction method that minimizes the prediction error is detected, and a prediction method and a motion vector used for encoding are determined.

In the second example according to the second embodiment, the access to the reference picture memory required for each encoded block in the second motion vector detecting step is performed for a motion having a small effect of the dual prime prediction. In a large coding block, the number of half of each of the first and second candidate frame prediction reference pixels (18 pixels × 18 pixels) is changed to the field prediction reference pixels (18 pixels × 10 pixels ×
2), and adding a frame prediction reference pixel (18 pixels × 18 pixels) near the motion vector 0, the conventional
/ 7, which can be expected to detect a frame motion vector with good coding efficiency instead of the dual prime prediction.

FIG. 4 is a diagram showing an example in which the motion vector detecting method according to the embodiment of the present invention is applied to an encoding device. As shown in FIG. 5, an image to be encoded is sequentially input from the input terminal 101 in the encoding target block in the horizontal direction and in the encoding target block in the vertical direction.

[0050]

(Equation 1)

As shown in FIG. 6A, the interpolation thinning-out means 110 interpolates the "luminance signal of the input image" in the horizontal direction and sub-samples the two fields at different phases at two-pixel intervals. A “primary search reference screen” is generated and stored in the internal image memory 120.

The motion vector candidate detecting means 130 performs a process corresponding to the above-described first motion vector detecting step. The motion vector candidate detecting means 130 subdivides the encoding target block input from the input terminal 101 in the horizontal direction at an interval of two pixels. Two types of search target blocks sampled (Fig. 6 (b) and (c))
Is generated, and two types of “signals of the first field of the search target block” are read out from the internal image memory 120 using the peripheral pixels at the same screen position input one screen before as reference regions, and Of the field motion vector candidate for the first field of.

In the detection of the first field motion vector candidate, first, two types of "first"
Is evaluated with respect to the "signal of the in-phase field of the primary search reference screen" with respect to the "signal of the field of the first search", and a field motion vector candidate with the minimum matching error is detected. In-phase field motion vector candidates for the field are assumed.

Using the matching error and the field motion vector candidate as initial values, matching with the "signal of the opposite phase field of the reference screen for the primary search" for the "signal of the first field" of the two types of search target blocks. The error is evaluated, a field motion vector candidate with the minimum matching error is detected, and the first
Field motion vector candidates for the field of.

Next, the motion vector candidate detecting means 130
Reads a reference area for the “signal of the other field of the search target block” from the internal image memory 120 based on the detection result of the field motion vector candidate for the “signal of one field of the search target block”,
A field motion vector candidate for the second field of the current block is detected.

In the detection of the second field motion vector candidate, first, two types of “first motion vector
Of the field signal of the primary search reference screen with respect to the signal of the in-phase field of the primary search reference screen, a field motion vector candidate with the minimum matching error is detected, and the second In-phase field motion vector candidates for the field are assumed.

Using the matching error and the field motion vector candidate as initial values, matching with two types of "signals of the second field of the block to be searched" and "signals of the opposite phase field of the reference screen for primary search" The error is evaluated, a field motion vector candidate with the minimum matching error is detected, and the second
Field motion vector candidates for the field of.

The motion compensating means 140 determines which of the two in-phase field motion vector candidates detected by the motion vector candidate detecting means 130 has the smaller matching error the reference motion vector and the frame motion vector of the dual prime prediction. It is selected as a reference vector for the search and supplied to the address generating means 150 together with the field motion vector candidate.

The address generation means 150 is stored in the external memory 160 based on the field motion vector candidates for each field of the block to be encoded, the reference motion vector for dual prime prediction, and the reference vector for frame motion vector search. The reference pixel of the local reproduction screen is read out and supplied to the motion compensation means 140 via the buffer memory 170.

The motion compensating means 140 performs a process corresponding to the above-described second motion vector detecting step. The motion compensating unit 140 applies, to the encoding target block supplied from the motion vector candidate detecting unit 130, a reference pixel read based on a reference motion vector for dual prime prediction and a reference vector for searching for a frame motion vector. To generate a reference pixel for detecting a reference motion vector in dual prime prediction by performing pixel interpolation for each field screen, detect a reference motion vector in units of 1/2 pixel in dual prime prediction, and perform pixel interpolation as a frame screen. To generate a frame motion vector detection reference pixel, and detect a frame motion vector in units of 1/2 pixel.

Further, the motion compensating means 140 generates a reference pixel interpolated for each field screen from the reference pixels read based on the field motion vector candidates, and detects a 1/2 pixel unit field motion vector. Supplies the reference motion vector of the half-pixel unit of the previously detected dual prime prediction to the address generation unit 150, and generates a reference pixel interpolated for each field screen from the reference pixel read based on the motion vector. Then, a minute motion vector dmv of the dual prime prediction is detected.

Further, the motion compensating means 140 detects an optimal motion vector from the motion vectors detected as described above based on the matching error, determines an optimal motion compensation mode, and determines the optimal motion vector and the motion compensation mode. The prediction signal corresponding to the motion compensation mode is generated and supplied to the local playback unit 220, and the difference (prediction error data) between the prediction signal and each pixel of the encoding target block is supplied to the variable length coding unit 230. Is generated and supplied to the discrete cosine transform means 180.

The discrete cosine transform means 180 calculates the prediction error data supplied from the motion compensation
The DCT coefficient data is generated by performing a dimensional discrete cosine transform process and supplied to the quantization unit 190.

The quantization means 190 performs a quantization process for dividing the DCT coefficient data supplied from the discrete cosine transform means 180 by a quantization width corresponding to each frequency component. Variable length coding means 2
30.

The variable length coding means 230 performs variable length coding on the motion compensation mode and the motion vector supplied from the motion compensation means 140 and the DCT coefficient data supplied from the quantization means 190, multiplexes them, and outputs Output from terminal 102.

The inverse quantization means 200 includes a quantization means 190
Is multiplied by the quantization width according to each frequency component to obtain the DCT of the original dynamic range.
DC restored to T coefficient and restored to original dynamic range
The inverse discrete cosine transform unit 210 that supplies the T coefficient data to the inverse discrete cosine transform unit 210
DCT coefficient data supplied from 0 is subjected to inverse discrete cosine transform processing to generate local reproduction prediction error data.

The local reproduction means 220 adds the local reproduction prediction error data supplied from the inverse discrete cosine transform means 210 and the prediction pixel data supplied from the motion compensation means 140 to generate a local reproduction image signal. And temporarily store it in the external memory (at this time, the past reference screen that is no longer referenced is overwritten on the stored address).

[0068]

As described above, according to the present invention, the frame motion vector candidate and the reference motion vector candidate for dual prime prediction are shared, or the frame motion vector candidate and the field motion vector candidate are shared and dual prime prediction is performed. Due to the limitation of the size of the reference motion vector candidate, the overlap of the reference pixel for frame prediction and the reference pixel necessary for the detection of the reference vector for dual prime prediction, or the overlap of the reference pixel for frame prediction and the reference pixel for field prediction and dual prime The reference pixel for reference motion vector search for prediction and the reference pixel for micro motion vector search for dual prime prediction are overlapped, and the access of the reference screen memory required for each coding block in the second motion vector detection step is performed in a conventional manner. Approximately 6/7, the reference motion vector for dual prime prediction It can also be detected accurately than conventional.

In the first motion vector detection step, when detecting a field motion vector candidate for each field of the motion vector detection target block, the search for the field motion vector candidate from the in-phase reference field is performed by the opposite phase reference. In the second motion vector detection step, in-phase field motion vector candidates used for searching for a reference motion vector for dual prime prediction are detected before searching for a field motion vector candidate from a field. This is obtained as an intermediate result of detection of vector candidates, and a detection step dedicated to reference motion vector candidates for dual prime prediction is not required.

[Brief description of the drawings]

FIG. 1 is a flowchart of a motion vector detection method according to a first embodiment of the present invention.

FIG. 2 is a flowchart of a motion vector detection method according to a second embodiment of the present invention.

FIG. 3 is a flowchart of a motion vector detection method according to a third embodiment of the present invention.

FIG. 4 is an explanatory diagram relating to a search for 1/2 pixel accuracy of a motion vector.

FIG. 5 is an explanatory diagram relating to an image input method.

FIG. 6 is an explanatory diagram of a search method for a motion vector.

[Explanation of symbols]

1... First motion vector detection step, 2a, 2
b, 2c: a step of selecting a reference motion vector candidate and a frame motion vector candidate for dual prime prediction; 3a, 3b, 3c: a second motion vector detection step; 30: a determination step of whether to search for a motion vector for dual prime prediction; 4b, 4c ...
Prediction method and motion vector determination step.

Claims (7)

[Claims] 1. A motion vector detecting method for detecting a motion vector for generating a reference image for an input image from a partial image of an image coded in the past, in order to encode the input image. A first motion vector detecting step of searching for a motion vector at the above interval, and a first field forming a partial area of the input screen from a first field forming a reference screen detected in the first motion vector detecting step. Based on at least one motion vector of a motion vector and a motion vector for a second field forming a partial area of the input screen from a second field forming a reference screen, at least an inter-frame prediction code in pixel units smaller than a predetermined pixel unit. Motion vector or dual prime predictive coding method Motion vector detecting method characterized by comprising a second motion vector detection step of searching a motion vector of a quasi. 2. A motion vector detecting method for detecting a motion vector for generating a reference image for an input image from a partial image of an image encoded in the past in order to encode the input image. A first motion vector detecting step of searching for a motion vector at the above intervals, and a motion vector for an odd field forming a partial area of the input screen from an odd field forming a reference screen detected in the first motion vector detecting step And a motion vector used in the inter-frame predictive encoding method in units of 1/2 pixel based on at least one of motion vectors for an even field forming a partial area of the input screen from an even field forming a reference screen. Search for the motion vector that is the reference when performing the prime prediction coding method. Cormorant motion vector detecting method characterized by comprising a second motion vector detection step. 3. A first motion vector detecting step of searching for a motion vector at intervals of one pixel or more, and a motion vector in half pixel units based on the motion vector detected in the first motion vector detecting step. In the motion vector detecting method for detecting a motion vector for a partial region of the current coded screen from the previous coded screen as the reference screen by the second motion vector detecting step of re-searching, the second motion vector detecting step Is a motion vector corresponding to the top field of the reference screen detected from the top field of the reference screen detected in the first motion vector detection step and the bottom field of the partial screen of the coded screen obtained from the bottom field of the reference screen. Frame with respect to at least one of the motion vectors for the field Motion vector detecting method characterized by comprising the steps of searching the reference motion vector of measurements of the motion vector and the dual prime prediction. 4. A first motion vector detecting step of searching for a motion vector at intervals of one pixel or more, and a motion vector in a 1/2 pixel unit based on the motion vector detected in the first motion vector detecting step. In the motion vector detecting method for detecting a motion vector for a partial region of the current coded screen from the previous coded screen as the reference screen by the second motion vector detecting step of re-searching, the second motion vector detecting step Is a motion vector corresponding to the top field of the reference screen detected from the top field of the reference screen detected in the first motion vector detection step and the bottom field of the partial screen of the coded screen obtained from the bottom field of the reference screen. The horizontal and vertical components of at least one of the motion vectors for the field Is less than or equal to a predetermined value, a search for a reference motion vector for dual prime prediction based on the motion vector and a search for a minute motion vector corresponding to the reference motion vector are performed; otherwise, a reference for dual prime prediction is performed. A method for detecting a motion vector, comprising: determining a motion vector as a reference motion vector for dual prime prediction without searching for a motion vector, and searching for a minute motion vector corresponding to the reference motion vector. 5. A first motion vector detecting step of searching for a motion vector at intervals of one pixel or more, and a motion vector in a 1/2 pixel unit based on the motion vector detected in the first motion vector detecting step. In the motion vector detecting method for detecting a motion vector for a partial region of the current coded screen from the previous coded screen as the reference screen by the second motion vector detecting step of re-searching, the second motion vector detecting step Are the motion vectors for the top field of the reference screen from the top field of the reference screen detected in the first motion vector detection step and the bottom field of the partial area of the coding screen from the bottom field of the reference screen. The horizontal and vertical components of at least one of the motion vectors for the field Is less than or equal to a predetermined value, a search for a reference motion vector for dual prime prediction based on the motion vector and a search for a minute motion vector corresponding to the reference motion vector are performed; otherwise, a reference for dual prime prediction is performed. A method for detecting a motion vector, comprising a step of prohibiting a search for a motion vector and a minute motion vector. 6. The method according to claim 5, further comprising the step of, when prohibiting the search for the reference motion vector and the minute motion vector for dual prime prediction, increasing the number of search points for the motion vector for frame prediction. Motion vector detection method. 7. A first motion vector detecting step for searching for a motion vector at intervals of one pixel or more, and a motion vector in 1/2 pixel units based on the motion vector detected in the first motion vector detecting step. In the motion vector detecting method for detecting a motion vector for a partial area of the current coded screen from the previous coded screen as the reference screen by the second motion vector detecting step of re-searching the first motion vector, Performs the field motion vector detection of the partial area of the coded screen, the motion vector search from the in-phase reference field, and then the motion vector search from the opposite-phase reference field. A method for detecting a motion vector, comprising the step of detecting a field motion vector.