JPH11308617A - Digital image coder and motion vector detector used therefor - Google Patents

Abstract

PROBLEM TO BE SOLVED: To reduce the power consumption of an MPEG2 coder by reducing the arithmetic operation amount for motion estimation, while minimizing deterioration in image quality and to attain motion estimate with high precision by panning information. SOLUTION: A motion vector detection section 7c retrieves a motion vector of an MB to be coded in an input image buffer 7a from reference image data in a reference image buffer 7b and selects a width of a retrieval range set to a motion vector detection section 7c, depending on whether or not a variance of the motion vector obtained by a variance calculation section 7h is smaller than a threshold level and applies an offset depending on the mean value of the motion vector which is obtained by a mean value calculation section 7g for a narrow retrieval range. Furthermore, when panning information is fed to a value latch section 7e via a CPU IF 8 externally, the retrieval range of the motion vector detection section 7c is made wide or narrow, depending on the panning information and an offset depending on the panning information is applied to the retrieval range, when the retrieval range is narrow.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a digital image signal encoding apparatus for encoding a digital image signal with high efficiency, and a motion compensation (inter-frame prediction) used in the digital image signal encoding apparatus for reducing redundancy in a time axis direction. The present invention relates to a motion vector detecting device that detects a motion vector required for a motion vector.

[0002]

2. Description of the Related Art As a digital image signal encoding device,
Conventionally, for example, “ITU-T White Book, Audiovisual / Multimedia-Related (H Series) Recommendations,” Japan ITU Association, published February 18, 1995, pp. 375-595
2. Description of the Related Art There is known an encoding apparatus based on a moving image compression technique H.262 (commonly referred to as an MPEG2 system) specified in (hereinafter, referred to as Document 1). The MPEG2 system is a general-purpose standard that is premised on being applied to various applications such as broadcasting, communication, and storage media.

[0003] In the MPEG2 system, in order to reduce the redundancy in the time axis direction, motion compensation is performed to obtain high compression efficiency. Motion compensation is the image to be compressed (picture)
Is divided into macroblocks MB consisting of 16 × 16 pixels, a difference between the macroblock MB and image data consisting of 16 × 16 pixels indicated by a motion vector in the reference image is obtained, and the difference image data is encoded. Things.

In the MPEG2 system, a macroblock MB obtained by encoding differential image data by motion compensation using a motion vector is called a non-intra (inter-frame predictive encoding) MB, and no difference is obtained without using a motion vector. The macroblock MB coded in (1) is called an intra (intra-frame predictive coding) MB. Also, a picture composed of only intra MBs is called an I picture, a picture that uses only image data of a past picture in display order as reference image data is called a P picture, and past and future pictures are displayed as reference image data in display order. A picture in which image data of both pictures can be used is called a B picture. Note that the image data of the I picture and the P picture are used as the reference image data of the B picture.
The image data of the B picture is not used as reference image data.

[0005] Here, I and P pictures that can be used as reference image data are referred to as reference images. When a reference image appears for every x images (pictures) arranged continuously, coding is performed. It is said that the m value of the MPEG2 bit stream thus obtained is x.

In order to detect a motion vector used for motion compensation, a method called block matching in which input image data to be coded and reference image data are matched in blocks of 8 × 8 pixels is mainly used. Have been.

In the coding by the MPEG2 system, the amount of computation required for the processing of motion estimation (hereinafter referred to as ME) for searching for the motion vector generally involves discrete cosine transform or quantization which is other computation. , Variable length coding and the like. And
The large amount of calculation leads to an increase in power consumption, and MPEG2
Power consumption in a coding apparatus based on the scheme (hereinafter, referred to as an MPEG2 coding apparatus) is mainly spent on ME processing.

In order to reduce the amount of calculation of the ME processing while suppressing the deterioration of the image quality of the encoded image, conventionally, for example, "1997 IEICE General Conference C-12-34"
(Hereinafter referred to as Reference 2), the shape of the search range is adaptively changed using the history of past motion vectors, thereby narrowing the search range of the motion vector while suppressing image quality deterioration. There is a known method.

[0009]

With the recent development of digital technology, devices for handling moving images have become smaller and lighter. Due to the development of the technology for reducing the size and weight, the demand for portable devices will continue to increase.

Incidentally, since these portable devices are driven by a battery, it is desirable to use a circuit with lower power consumption in order to extend the life of the battery.
In the MPEG2 encoding apparatus, most of the calculation amount is devoted to the ME. To reduce the power consumption, the search range of the motion vector is reduced or the search is simplified to reduce the ME. Although it is conceivable to reduce the amount of calculation, there is a first problem that this adversely affects the encoded image quality.

Further, as described in the above reference 2, the method of adaptively changing the shape of the search range of the motion vector is as follows:
By controlling the shape of the search range intricately, an image quality almost equal to that having a large search range can be obtained.
Does not show a control method for determining the shape of the search range, and it is conceivable that the function may not be fully used.

On the other hand, when panning is performed when MPEG2 encoding is considered as a whole system, pan information of the entire screen may be detected by means other than the MPEG2 encoding device. Also, when panning is performed, the motion vector often indicates the destination moved by panning, and therefore the range in which the apparatus performing the ME process searches for the motion vector is limited to the area around the destination moved by panning. However, there is a second problem that searching for other places is wasteful.

A first object of the present invention is to solve the first problem and to reduce the amount of operation of the ME while suppressing the deterioration of the quality of an encoded image by using a simple circuit configuration.
An object of the present invention is to provide a digital image encoding device capable of reducing power consumption of an encoding device and a motion vector detecting device used for the digital image encoding device.

A second object of the present invention is to solve the second problem and effectively utilize the information of the pan detected externally in the MPEG2 encoding apparatus so as to eliminate unnecessary search operations for motion vectors. It is another object of the present invention to provide a digital image encoding device and a motion vector detecting device used therefor.

[0015]

In order to achieve the first object, the present invention provides a circuit block for performing ME, a mode for limiting a search range of a motion vector and a function for offsetting the search range. A function of autonomously returning these to the initial state is provided.

Normally, the correlation of a moving image in the time axis direction is high, but in some cases, the correlation is low. Generally, when a moving image having a high correlation in the time axis direction is encoded by MPEG2,
In many cases, the motion vectors point in substantially the same direction. On the other hand, when a moving image having low correlation in the time axis direction is encoded by MPEG2, the motion vector points in a random direction. Therefore, as a result of a search for a motion vector using past reference image data, when the direction of the motion vector is concentrated in a similar place, a mode for limiting the search range is set, and (Which determines the position within the original wide search range) is set as appropriate. If the limited search range is out of the range in which a correct motion vector can be detected, the search range is returned to the initial wide state autonomously.

By doing so, the search range can be narrowed without greatly affecting the search result of the motion vector, and as a result, the amount of calculation of the ME is reduced. MPE while suppressing image quality degradation
Low power consumption of the G2 encoding device can be realized.

In order to achieve the second object, the present invention provides a mode for limiting the search range and a function for offsetting the search range from the outside. The information on the motion vector detected by the conversion device can be externally referred to.

A motion vector to be detected is predicted from pan information detected outside the MPEG2 coding apparatus, and the predicted value is set as the above-mentioned offset value. In addition, it is determined whether or not the search range may be limited based on information on the actually detected motion vector, and a mode for limiting the search range is set. By doing so, the information of the bread detected externally can be stored in the MPE.
It can be used effectively in the G2 encoding device.

[0020]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram showing a first embodiment of a digital image encoding device according to the present invention.
1 is an image data input terminal, 2 is an MPEG2 encoding device, 3 is an SDRAM (semiconductor storage element that can be read and written at any time that requires a synchronous storage operation), and 4 is a CPU (central processing unit) block. , 5 is a bit stream output terminal, 6 is VIDEO (video)
Block, 7 is a ME (motion estimation processing) block, 8 is C
PU IF (CPU interface) block, 9 is MC (motion compensation) block, 10 is JUDGE (judgment)
Block 11 is DCT / IDCT (discrete cosine transform /
Inverse discrete cosine transform) block, 12 is a Q / IQ (quantization / inverse quantization) block, 13 is a RATE (rate control) block, 14 is a VLC (variable length coding) block, 15 is a CODE (code) block, 16 is SDR
An AM address bus 17 is an SDRAM data bus.

Referring to FIG. 1, the SDRAM block 3 stores image data of I and P pictures as reference image data. When a read address is supplied from an SDRAM address bus 16, the SDRAM block 3
The reference image data is read from the area corresponding to the read address at step (1) and output to the SDRAM data bus 17.
This is called reading from the SDRAM block 3. Also, the write address from the SDRAM address bus 16 is
When the image data is supplied from the SDRAM data bus 17, respectively, the image data can be written in the area of the SDRAM block 3 corresponding to the write address. This is SD
This is called writing to the RAM block 3.

Next, the operation of each unit of the MPEG2 encoder 2 will be described.

In the MPEG2 system, both past and future images can be used in the display order as reference images for performing motion compensation of an image (picture). For this reason, the input order of the images (that is, the display order) does not always match the encoding order. For example, as shown in FIG. 7A, assuming that pictures P ₁ , P ₂ , P ₃ , and P ₄ are input in that order, when these are coded, as shown in FIG. Next, pictures P ₁ , P ₄ , P ₂ , and P ₃ are arranged in this order.

The image data of each picture input from the input terminal 1 is supplied to the VIDEO block 6. In the VIDEO block 6, the order of the input pictures is rearranged in the encoding order as described above. And a luminance signal Y and two color difference signals (RY) for each picture.
_, (BY), and each signal is divided into macroblocks MB. Here, the luminance signal in the macro block MB is composed of 16 × 16 pixels, and the color difference signals (R−Y), (B
-Y) (these are collectively referred to as a color difference signal C)
Is composed of 8 × 8 pixels thinned out in both the horizontal and vertical directions. The luminance signal Y is sequentially supplied to the ME block 7, the MC block 9, and the JUDGE for each macro block MB.
The two color difference signals C are also supplied to the MC block 9 for each macroblock MB.

The ME block 7 includes a luminance signal Y from the VIDEO block 6 and an SDRA for each macro block MB.
The motion vector MV is searched using the reference image data D ₄ read from the M block 3 and supplied to the MC block 9. Thus, the MC block 9 uses the searched motion vector MV to generate a VI
It generates difference image data D ₂ between the luminance signal Y and the macro block MB of a luminance component of the reference image data D ₄ read from SDRAM block 3 corresponding to this macro block MB from DEO block 6. The JUDGE block 10 calculates the variance of each of the difference image data D ₂ and the luminance signal Y from the VIDEO block 6 for each macro block MB, and the smaller one of these variances has a smaller code amount. it is determined that can be coded, ME block 7 the determination result as an intra-MB determination signal S _4,
It is supplied to the MC block 9 and the CPU IF block 8.

When the variance of the luminance signal Y is determined to be smaller based on the intra MB determination signal S ₄ , the MC block 9 determines the luminance of the macro block MB supplied from the VIDEO block 6. 1 of signal Y
6 × 16 pixels and 4 × 8 × 8 pixels each
Of the macro block MB supplied from the VIDEO block 6 and each of the 8 color difference signals C of the macro block MB.
× 8 pixels are directly used as blocks, and these six blocks are time-division multiplexed to obtain DCT / I as block data D _3.
Supply to DCT block 11. When it is determined that the variance of the difference image data D ₂ is smaller,
This difference image data D ₂ is divided into four blocks each consisting of 8 × 8 pixels, and two color difference signals C supplied from the VIDEO block 6 and the SDRAM block 3
From the reference image data D ₄ read out from the image data and the respective color difference signals of 8 × 8 pixels determined by the motion vector MV, and these are used as blocks as they are, and these four blocks of luminance signals Y time division multiplexing the blocks of two six 8 × 8 pixels comprising a block of the color difference signal C, DC as block data D ₃
Supply to T / IDCT block 11.

The block data D ₃ as described above is a DC data
The discrete cosine transform is performed by the T / IDCT block 11, and Q /
After being quantized in the IQ block 12, the VLC block 1
4 and is subjected to variable length encoding to form an MPEG2 bit stream. This MPEG2 bit stream is temporarily stored in a transmission buffer by the CODE block 15 and read-out is controlled, so that it is smoothed to a predetermined constant bit rate and output from the output terminal 5.

Further, information on the code amount of the image data generated by the VLC block 14 is supplied to the RATE block 13, and from this and the preset bit rate information, the bit rate of the MPEG2 bit stream output from the output terminal 5 is determined. In order to keep the rate constant, the Q / IQ block 12 calculates how coarsely the quantization should be applied.
The / IQ block 12 performs quantization according to the calculated value. Further, the RATE block 13 calculates a generated code amount per one macroblock MB that can be allowed so that the fixed bit rate does not fail, and calculates the VLC block 14.
The encoding process is controlled according to the calculated value so that the generated code amount per MB does not exceed a value that does not break down the bit rate.

By the way, in the MPEG2 system, in order to perform irreversible encoding, the original image (input image) input from the input terminal 1 and the MPEG2 image output from the output terminal 5 are output.
It will be different from the image obtained by decoding the bit stream. The decoder on the receiving side that decodes this MPEG2 bit stream uses the decoded image data as reference image data in order to reproduce the original image data from the differential image data. 2, when the difference image data is generated by the MC block 9, if the input image data is used as it is as the reference image data, the reference image data in the MPEG2 encoding device and the reference image data in the decoder on the receiving side are used. The difference between is important. In order to eliminate this difference, the MPEG2 encoder 2 must use the image data obtained by decoding the MPEG bit stream as reference image data. This is called local decoding.

This local decoding is performed as follows. That is, as described above, the VLC block 14
In this document, the generated code amount per MB is limited for bit rate control, and image data exceeding that amount is not subjected to variable length coding. For this reason, a part of the image data output from the Q / IQ block 12 may not be subjected to the variable length coding. Therefore, in order to match the value obtained by decoding the MPEG2 bit stream with the value obtained by local decoding, the image data obtained by the variable length coding from the VLC block 14 is supplied to the Q / IQ block 12 and dequantized. The DCT / IDCT block 11 performs an inverse discrete cosine transform and supplies the result to the MC block 9. In the MC block 9, the reference image data is added to the difference image data of the non-intra MB, and the difference image data is returned to the image data.
The decoded image data is generated with the data of B
As decoded image data is reference image data D ₁ of the picture and P-pictures, via the SDRAM data bus 17, SD
It is stored in the RAM block 3.

FIG. 2 is a block diagram showing a first embodiment of the ME block 7 as the motion vector detecting device in FIG. 1, wherein 7a is an input image buffer, 7b is a reference image buffer, and 7c is a motion vector detecting unit. , 7d is a reference image read address generation unit, 7e is a value holding unit, 7f is a judgment and counting unit, 7g is an average value calculation unit, and 7h is a variance value calculation unit. Parts corresponding to those in FIG. And a duplicate description will be omitted.

A macroblock MB composed of 16 × 16 pixels of the luminance signal Y supplied from the VIDEO block 6 (FIG. 1) is stored in the input image buffer 7a, and the macroblock MB is read from the SDRAM block 3 The output reference image data D ₄ is stored in the reference image buffer 7.
b. Here, the reference image buffer 7b stores the horizontal 128 pixels × vertical 56 pixels of the reference image data D _4.

From this reference image buffer 7b, reference image data in the search range of the motion vector MV for the macro block MB stored in the input image buffer 7a is read, and the image of the macro block MB read from the input image buffer 7a is read. The search range of the motion vector is supplied to the motion vector detection unit 7c together with the data. The search range of the motion vector can be set to a normal wide search range and a limited narrow search range, as described later. Here, the normal wide search range is -56 to +56 pixels in the horizontal direction and -20 to +20 in the vertical direction, centering on the position corresponding to the macroblock MB read from the input image buffer 7a.
This is a range composed of pixels (including the region (16 × 16 pixels) corresponding to the macroblock MB,
The limited narrow search range is a range of −24 to +24 pixels in the horizontal direction and −8 to +8 pixels in the vertical direction (corresponding to the macroblock MB). Including the region, the size is 64 pixels horizontally × 32 pixels vertically).

The motion vector detecting section 7c performs block matching between the image data of the macroblock MB supplied from the input image buffer 7a and the image data in the search range from the reference image buffer 7b to obtain the macroblock MV. A motion vector MV is detected.
This block matching is 16 ×
By moving the search range up, down, left, and right with respect to a macroblock MB consisting of 16 pixels, a correlation (prediction error) with a 16 × 16 pixel area at each moving position is obtained.

When the motion vector MV is obtained for one macroblock MB of the input image in this way, the search for the motion vector MV of the next macroblock MB is performed. The search range for this macroblock MB is as follows. The search range is different from the search range for the immediately preceding macroblock MB. For this reason, when searching for the motion vector MV of the next macroblock MB, reference image data is read from the SDRAM block 3 in order to set a search range for the motion vector MV. Most of the search ranges for the two macroblocks MB to be performed overlap, and a part of the reference image data in the search range used for searching for the motion vector MV of the immediately preceding macroblock MB can be reused. For this reason, the range of the reference image data read from the SDRAM block 3 and taken into the reference image buffer 7b is a range of 16 horizontal pixels × 56 vertical pixels in the above-described ordinary wide search range, and is in the limited narrow search range. , 16 horizontal pixels × 32 vertical pixels.

The reference image read address generating section 7d is an SD
A mode for limiting the read range of the reference image data from the RAM block 3 and a function for offsetting the read position are provided. These modes and functions are set by the CPU.
This is performed by the IF block 8.

Next, the modes and functions of the reference image read address generator 7d by the CPU IF block 8 will be described. Before that, the correlation between the input image to be encoded and the reference image will be described with reference to FIGS. The relationship between the motion vector and the motion vector and its search range will be described. 3 is a distribution diagram of motion vectors when the correlation between the input image and the reference image is low, FIG. 4 is a distribution diagram of motion vectors when the correlation between the input image and the reference image is high, and FIG. FIG. 6 is an explanatory diagram of the limitation of the search range and the offset when the correlation with the reference image is high, and FIG. 6 is an explanatory diagram of the case where the correlation between the input image and the reference image to be encoded with the search range limited is low. .

In FIGS. 3 to 6, the range α
-Β represents an ordinary wide search range, and the center of the vertical axis of the search range α-β is set in the search range α.
Macroblock M of input image data to be encoded by -β
B position. When searching for a motion vector, a macroblock MB is relatively moved within a search range having a pixel range as described above, for example, by a half image, thereby searching for a highly correlated portion. The “distribution density of motion vectors” on the vertical axis in FIGS. 3 to 6 indicates the degree of distribution of motion vectors detected for a plurality of MBs. For example, in the case of FIG. 3, the correlation between the input image and the reference image is low, indicating that a motion vector is randomly detected. Also,
FIG. 4 shows that the rate of detecting the motion vector is high.

When the correlation between the input image to be encoded and the reference image is low, the motion vectors are uniformly detected in the search range α-β in the motion vector detecting section 7c as shown in FIG. If the correlation between the input image and the reference image is low, the variance of these difference images tends to be large. Therefore, the determination result in the JUDGE block 10 is
This means that there are many B's.

Conversely, when the correlation between the input image and the reference image is high, for example, as in an image in which simple panning is performed, the detected motion vector is as shown in FIG. It gathers in the area centered on the vector representing the movement of the image by panning, and the judgment result of the JUDGE block 10 indicates that there are many non-intra MBs.

As described above, when the motion vectors are concentrated in a narrow range within the search range α-β, as shown in FIG. 5, the motion vector in the search range with respect to the average value Va of the motion vector representing the position is obtained. Even if the offset value N _OF is specified and the search range is limited to a narrow search range including most of the motion vectors (that is, the search range α′−β ′), the image quality of the encoded image is greatly affected. Absent. Note that, with respect to this limited narrow search range α′-β ′, the above-described search range α-β becomes the normal wide search range.

Further, during encoding, a narrow search range α in which the location where the motion vector is concentrated is limited because the correlation between the input image and the reference image is reduced due to scene switching or the like and the pan direction is changed. If it deviates from '-β', the search range must be returned to the original wide range α-β.

FIG. 6 shows the limited narrow search range α ′.
This indicates a case where the correlation between the input image and the reference image is low when −β ′ is set. In this case, it should be detected in a hatched portion outside the search range α′−β ′. Is not correctly detected, and the number of intra MB determinations becomes very large. Therefore, when it is determined that the number of intra MBs is large when the limited narrow search range is set, the normal wide search range α− shown in FIG.
It is necessary to return to β.

As described above, the setting of the wide search range and the narrow search range and the setting of the offset of the narrow search range are performed. First, in FIG. The value of ₃ is 0 and the CPU
Reference image read address generator 7 from IF block 8
The case where d is not set will be described.

Whether the search range of a motion vector is set to the above-described wide search range or limited to a narrow search range is determined at the time of encoding a P-picture as follows.

That is, the average value Va (FIG. 5) of the motion vector MV detected by the motion vector detecting unit 7c is calculated by the average value calculating unit 7g during encoding into the P picture, and the variance of the motion vector MV is calculated. The values are calculated by the variance value calculation unit 7h and supplied to the value holding units 7e. Also, JUDGE
The decision counting unit 7f counts the number of intra MBs in the P picture from the intra / non-intra decision result of the block 10, and supplies the counted value to the value holding unit 7e.

[0047] When the coding of the P-picture is completed, the value held unit 7e calculates an offset value N _OF readout position from the average value Va taken, the motion vector detecting section 7c this result as a read position offset signal S ₅ And the reference image read address generation unit 7d. Calculation of the read position offset S ₅ will be described later. Also, the variance calculation unit 7h
When the variance value calculated in step (b) is equal to or less than a preset threshold, the readout range of the reference image data (that is, the search range)
May be limited, and the read-out range limiting signal S ₆ is sent to the motion vector detecting unit 7 c and the reference image read address generating unit 7.
d and the reference buffer 7b. In this case, the reference image read address generator 7d is by the read range limit signal S _6, the read area of the reference image data from the SDRAM 3, the position corresponding to the macro block MB is stored in the input image buffer 7a Restrict narrowly, as for a narrow search center.

[0048] Further, when the read area limit signal S ₆ from the value holding section 7e is output, when the even read position offset S ₅ is output, in response to the read position offset S _5, the reference image from SDRAM3 Set the position of the data read range. If the absence of read position offset S ₅ are read area of the reference image data in SDRAM3 is the position corresponding to the macro block MB is stored in the input image buffer 7a is in the range of the center position, the read position If there is an offset S _5, it becomes this center position is shifted by an amount corresponding to the offset value N _oF represented by the read position offset S _5. The reference image buffer 7
In b, will be the reference image data of the narrow search range read area of the reference image data is written in SDRAM3 set by the reference image read address generation section 7d is stored, according to the read range limit signal S ₆ The reference image data is read and supplied to the motion vector detection unit 7c. Thus, the motion vector detecting unit 7c
Is, for example, as shown in FIG.
A limited narrow search range α′−β ′ having an offset value N _OF corresponding to ₅ is set.

[0049] Incidentally, when the determination count portion 7f is outputting a constant or count values (i.e., when the number of intra MB is large), the read position offset signal S ₅ is 0 also, the reading range limit signal S ₆ is not output. Further, when the dispersion value calculated by the variance calculation section 7h is greater than the threshold value are likewise read position offset signal S ₅ and the read range limit signal S ₆ is not output. In this case, the reference image read address generation unit 7d reads out the reference image data so as to read a wide search range centered on the position corresponding to the macroblock MB stored in the input image buffer 7a. Reference image buffer 7
b. In the motion vector detecting section 7c,
The reference image data is read from the reference image buffer 7b, and a wide search range centered on a position corresponding to the macroblock MB stored in the input image buffer 7a, like a normal wide search range α-β shown in FIG. Is set.

A narrow search range α'- with a limited search range
In the mode of β ′, the readout range of the reference image data from the reference image buffer 7b is 縦 each in the vertical and horizontal directions, and the operation time of the motion vector detection unit 7c is shorter than in the case of the normal wide search range α−β. It becomes 1/4. In addition, SDR is used to search for a motion vector in the next macroblock MB.
S for reading reference image data from AM block 3
The amount of access to the DRAM block 3 is also reduced, and the power consumption of the ME block 7 can be reduced. Also, when a wide search range α-β is required in such a mode, as described above, the wide search range α-β is returned autonomously, so that the influence on the encoded image is minimized. It is held down to the limit.

[0051] During the encoding of the next picture, the reference image read address generator 7d is unless the read range limit signal S ₆ is 0, limits the read range, further, at the read position offset signal S ₅ 0 without reads reference image data from the SDRAM block 3 the read position offset accordingly, also the read range limit signal S ₆ 0
, The SDRAM block 3 in the normal wide read range
From the reference image data.

Here, if the m value of the moving picture to be encoded is constant, the read position offset in the next P picture may be the average value Va of the motion vector. When the m value changes, scaling is performed according to the m value.

Next, the determination of the read position offset in the B picture will be described. FIGS. 7A and 7B are explanatory diagrams of the scaling of the motion vector in the first embodiment. FIG. 7A shows the arrangement of pictures in display order (input order), and FIG. Are shown, respectively.

[0054] In FIG. 7 (a), now paying attention to the four pictures P _1, P2, P _3, P ₄ inputted sequentially, a specific image is a picture P ₁ in these pictures, P _2, P ₃ ,
Shall will change the position in the order of P _4, in these respective pictures, the macro block MB included in the specific image and the macro block MB ₁ in picture P _1, and the macro block MB ₂ in pictures P _2, picture a macroblock MB ₃ at P _3, and the macro block MB ₄ in picture P _4. Here, it is assumed that panning is being performed, and therefore, the entire image moves in the same direction and at the same speed. In Figure 7, the macro blocks MB of the display position of the specific image, picture _{P 1, P 2, P 3} , in the order of P ₄ macroblock MB ₁ from the lower left corner to the upper right corner, MB _2, M
It shows a state in which it moves to B ₃ and MB ₄ . These macro blocks MB ₁ to MB ₄ image is very highly correlated with each other. For convenience of explanation, these macro blocks MB
It is shown in an exaggerated spacing between ₁ to MB _4.

The pictures P _{1 to} P ₄ correspond to the picture P ₁
Picture but as an I picture, a picture P ₄ is P picture, and if the picture P _2, P ₃ are respectively coded as a B-picture, as shown in FIG. 7 (b), which is coded to an I picture P ₁ , a picture P ₄ encoded into a P picture, a picture P ₂ encoded into a B picture,
It will be arranged in the order of the picture P _3, which is encoded into B pictures. Accordingly, the m value of the MPEG2 bit stream in this case is 3.

When the pictures P _{1 to} P ₄ are encoded as described above, since the picture P ₁ is encoded as an I picture, no motion vector is involved in this encoding.

Next, when the picture P ₄ is encoded into a P picture, the macro block MB _{4 of the} picture P ₄ is encoded.
Respect, the image data of the picture P ₁ of the range corresponding to the macroblock MB ₄ is read from the SDRAM 3, the reference image data D _4, reference image buffer 7b
Is stored in In this case, also read area read position offset signal S ₅ from the value holding section 7e limit signal S ₆
Is not output, the reference image buffer 7
From d, the reference image data of the read range including the search range of normal broad range around the position of the macro block MB ₄ is supplied to the read and motion vector detection unit 7c, a macroblock in this search range a search for a motion vector MV is performed for MB _4.

Here, assuming that the motion vector MV ₁ is within this search range, the pictures P ₁ and P ₄ have very high correlation, and the distribution density of the motion vector in this case is shown in FIG. as indicated, so that the concentrated around MV _1. Average value Va of the motion vector calculated by the average value calculating section 7g in this case is close to MV _1, also
It is assumed that the variance of the motion vector calculated by the variance calculator 7h is lower than a preset threshold.

Picture P ₂ to be encoded into the next B picture
When encoding a macroblock MB _2, the search (forward motion vector search of a motion vector MV ₂ as an input a preceding image data reference image data of the picture P ₁ (referred forward motion vector this) ) and the search for the motion vector MV ₃ to a reference image data to the image data of the picture P ₄ after the input order and (backward motion vector search) is performed.

First, in the search for the forward motion vector MV ₂ , the search range offset value N _OF (FIG. 5) and the SDRAM
Read position offset value of the reference image data from the 3, the average value Va of the motion vector MV ₁ picture P ₄ obtained above
Size of 1/3 in this motion vector MV ₁ is set in the same direction.

Next, the search for the backward motion vector MV ₃ is performed. In this case, the offset value N _OF of the search range (FIG. 5) and the read position offset value of the reference image data from the SDRAM 3 are: Picture P ₄ obtained earlier
2/3 the size of the average value Va of the motion vector MV ₁ and the motion vector MV ₁ is set in the reverse direction.

[0062] Further, when encoding a macroblock MB ₃ at the next picture P _3, as in the case of the picture P _2, motion vector MV ₄ before the reference image data to the image data of the picture P ₁ a forward motion vector search of performing the search, the motion vector search after performing a search of motion vector MV ₅ after a reference image data to the image data of the picture P ₄ is performed.

In this case, the forward motion vector MV ₄
In the search for, the offset value N _OF of the search range (FIG. 5) and S
Read position offset value of the reference image data from DRAM3 is set to the motion vector in the same direction at 2/3 the size of the average value Va of the motion vector MV ₁ picture P ₄ obtained previously.

Similarly, the backward motion vector M
While the search for V ₅ is than is carried out, in this case, the read position offset value of the reference image data from the offset value N _OF (Fig. 5) and SDRAM3 search range, the motion of the picture P ₄ obtained above the average value Va of the vector MV ₁ 1 /
The motion vector Va is set in a direction opposite to the motion vector Va at a size of 3.

When the m value of the MPEG2 bit stream is 3, the pictures P ₅ and P ₆ following the picture P ₄ (FIG. 7)
(Not shown in FIG. 7), and encodes the picture P ₇ (not shown in FIG. 7) into a P picture using the image data of the picture P ₄ as reference image data. In this case, the average value V of the motion vector MV ₁ picture P ₄ to the value holding section 7e
a is held, also, since the variance value of the motion vector MV ₁ calculated at the picture P ₄ is lower than the threshold value, from the value holding unit 7e, according to the dispersion value and these average values Va, read position offset signal S ₅ and the read range limit signal S ₆ is output, a narrow search range is limited to the motion vector detecting portion 7e is set, the offset value N of the search range in this case
_OF (Fig. 5) and read position offset value of the reference image data from SDRAM3 is the same size as the average value Va of the motion vector MV ₁ picture P ₄ obtained above, in the same direction as the motion vector MV ₁ Is set. Where MPEG2
When the m value of the bit stream changes to a value other than 3, the search range offset value N _OF (FIG. 5) or SDRA
The read position offset value of the reference image data from M3 is
Average value Va × m value of motion vector MV ₁ of picture P ₄ /
3 in size, are set in the same direction as the motion vector MV _1.

When the average value Va of the motion vector MV newly changes in the encoding of the picture P ₇ , the value holding unit 7 e
In this case, the average value Va is updated to the new average value Va, and the average value Va is held until the new average value Va is updated in the encoding of the next P picture.

In the above search for the motion vector MV,
When the dispersion value calculated by the variance calculation section 7h is higher than the threshold value, the read position offset signal S ₅ and the read range limit signal S ₆ is not output from the value holding section 7e (Even in this case, the value holding section 7e In this case, the average value Va is held as it is.) Therefore, the range of reading the image from the SDRAM 3 is also expanded, and a normal wide search range is set in the motion vector detection unit 7c.

The above description is for the case where the m value of the MPEG2 bit stream is 3, but when the m value is 2 (one reference picture (I or P picture) for every two pictures)
However, the reference image data read position offset value N _OF of the SDRAM 3 in the forward motion vector search of the picture coded into the B picture is equal to the previously input P
The motion vector MV is set in the same direction as the motion vector MV at half the average value Va of the motion vector MV of the picture, and the read position offset value N _OF of the reference image data of the SDRAM 3 in the backward motion vector search is P The motion vector MV is set in a direction opposite to the motion vector MV at a half of the average value Va of the motion vector MV of the picture. Otherwise,
Same as above.

Next, the operation when the value of the external setting valid mode signal S ₃ from the CPU IF block 8 is 1 will be described.

In FIG. 1, the CPU block 4 includes a CP
From the U IF block 8, the detection result of the motion vector MV and the intra MB determination signal S ₄
Receipt and values, using a user-defined algorithm, the calculated and reference values of the image data read position offset value and a vector search range limited signals S ₁ of SDRAM 3 (wherein, e.g., instead of the average value Va of the An algorithm that uses a mode value and a sum of absolute value differences instead of a variance value can be considered.)

Then, the value of the external setting valid mode signal S ₃ is set to 1, the value of the offset value setting signal S ₂ is set to the calculated reference image data reading position offset, and the value of the vector search range limiting signal S ₁ is When the vector search range is limited, it is set to 1 in each of the CPU IF blocks 8.

[0072] In FIG. 2, the value holding section 7e, the value of the read range limit signal S ₆ to the value of the vector search range limited signals S ₁ from the CPU IF block 8, the offset value the value of the read position offset signal S ₅ the value of the setting signal S ₂ respectively set to be output to. Thus, the reference image data read position offset value and a vector search range limited signals S ₁ and the value C
By enabling the setting from the PU block 4, information on the pan detected from the outside can be stored in the MPEG format.
2 can be used in the encoding device 2.

In the first embodiment, as described above, the average value and the variance value are used. However, this calculation is performed by another calculation method, for example, the mode value is replaced with the variance value instead of the average value. May be used instead of the absolute value difference sum.

In the first embodiment, the restriction between the reference image data read position offset and the search range is updated when a P picture is coded. It is also possible to calculate and update the position offset and the limit of the search range. In addition, based on the information of the average value and the variance value of the motion vectors MV in the past determined number of macroblocks MB, instead of the picture unit, the reference image is set for each macroblock MB or for each of a plurality of macroblocks MB. It is also possible to update the data reading position offset and the limit of the search range.

Further, in the first embodiment, the search range is set to two types, that is, a wide mode and a restricted narrow mode. However, the search range is selectively controlled to three or more modes having different sizes. You can also.

FIG. 8 is a block diagram showing a second embodiment of the motion vector detecting device used in the digital image encoding device according to the present invention, wherein 7i is a search point counter, and 7j is a search range judgment result holding unit. 2 are denoted by the same reference numerals, and redundant description will be omitted. This embodiment can also be used as the ME unit 7 in FIG.

In the figure, the ME block 7 has two modes, a full search mode and a narrow search mode, and the narrow search mode has five search points (search ranges).

FIGS. 9A and 9B show the search range of the ME block 7. FIG. 9A shows the search range when encoding a P picture, and FIG. 9B shows the search range when encoding a B picture. Are shown respectively.

In FIG. 9A, the search range in the full search mode at the time of P-picture coding is as follows: (a), horizontal -56 to +56 pixels, vertical -20 to +20 pixels (macro block MB very a wide search range SP _1, also the search range in the narrow search mode of 16 × 16 pixel region be included horizontal 128 pixels × vertical 56 pixels in size) of at a position corresponding to the (b ), The search points of horizontal −24 to +24 pixels and vertical −6 to +6 pixels (horizontal 64 pixels × vertical 28 pixels including the 16 × 16 pixel area at the position corresponding to the macroblock MB)
A narrow search range SP ₂₁ of (1,1). In this narrow search mode, in addition to the search range SP _21, as shown as (c), with respect to a wide search range SP _1, the search range of a search point in the central portion of the left half portion and (0,1) and SP _22, the search point of the top half of the central part and a narrow search range SP ₂₃ to (1,0), a narrow search range SP ₂₄ that the search point in the central portion of the right half and (2,1) , also four narrow search range SP ₂₅ that the search point lower half of the central portion (1,2), these five search range SP ₂₁ to SP ₂₅ are horizontal, the number of the same pixel in the vertical Range.

In FIG. 9 (b), the search range in the full search mode at the time of B picture encoding is as shown in FIG.
(Macro area of 16 × 16 pixels at a position corresponding to the block MB be included horizontal 64 pixels × vertical 56 pixels) is a broad search range SB ₁ of the search range in the narrow search mode horizontal -8 + 8 pixels , Vertical -6 to +6 pixels (32 horizontal pixels × 28 vertical pixels including a 16 × 16 pixel area at a position corresponding to the macroblock MB). Search range in the narrow search mode, the positional relationship and the search points for the search range SB ₁ in all search mode is a same manner as in the narrow search mode at the time of P-picture coding, (ii), as (c) As shown, five search ranges SB
_21, there is _{SB 22, SB 23, SB 24} , SB 25.

First, the case where the ME block 7 operates in the full search mode will be described with reference to FIG.

In the search point counter 7i, a counter is provided for each search point in the narrow search mode, and the number of occurrences of motion vectors entering each search point is counted by each counter. The counting result S ₉ of each search point is reported to the CPU block 4 through the CPU IF block 8 on a picture basis.

In the MPEG2 system, there are two types of pictures, field pictures and frame pictures.
In the case of a frame picture, there are two types of motion vectors, a frame motion vector and a field motion vector. Therefore, when the motion vector detecting unit 7c detects a field motion vector, the search point counter 7
i converts this into a frame motion vector and counts it.

In the case of a B picture, there are a forward motion vector and a backward motion vector, and the backward motion vector is added as being included in a search point at a position symmetrical with respect to the center of the search range. For example, when a backward motion vector is detected at the search point (0, 1),
The search point (0, 1) is added to the counter of the search point (2, 1) symmetrical with respect to the center of the search range.

Further, when a motion vector is detected in an area where search points overlap, the counters of all search points to which the motion vector belongs are counted in duplicate.

Next, the determination process of the search mode in the CPU block 4 will be described.

If the frequency of occurrence of the motion vector in the diagonal direction is high (there are many motion vectors that do not belong to any search point), the next picture is set as the picture in the full search mode. If the motion vector occurrence frequency at any of the search points is equal to or greater than a preset threshold, the next picture may be set to the narrow search mode, but in order to prevent transition to the narrow search mode unnecessarily, The determination results for several consecutive pictures are stored, and if the rate at which the corresponding narrow search point is determined is equal to or higher than a certain value, the mode transits to the narrow search mode. At this time, if the frequency of occurrence is the same among a plurality of search points, the search point (1, 1) has the highest priority, and the following (0, 1), (2, 1), (1, 0),
The priority becomes lower in the order of (1, 2).

When the transition to the narrow search mode is determined by the above determination, the next picture is set to the narrow search mode,
The search point is the search point with the highest motion vector occurrence frequency. This result, through the CPU IF block 8, ME block 7 as a search range selection signal S ₈
Is supplied to the judgment result holding unit 7j. Determination result holding unit 7
j supplies the search range selection signal S ₈ held at the right time, the reference buffer 7b as the search range signal S _7, the motion vector detecting section 7c and the reference image read address generation unit 7d. In the next picture, the reference image read address generator 7d reads the search point specified by the search range signal S _7, the motion vector detecting section 7c searches only search point specified by the search range signal S7.

When the ME block 7 is operating in the narrow search mode, as in the case of the ME 7 shown in FIG. 2, the correlation between the input image and the reference image becomes lower due to scene switching or the like, and the pan direction changes. If, for example, the place where the motion vector MV concentrates deviates from the search range, it is necessary to return to the full search mode. Here, macroblocks MB from the leftmost macroblock MB to the rightmost macroblock MB in the same horizontal arrangement in one picture
Is referred to as a macroblock MB line, in order to determine the return to the full search mode, in some macroblock MB lines, a motion vector is detected in advance in the full search mode (hereinafter, such a motion vector is referred to as a macroblock MB line). The number of motion vectors generated in each search point in the narrow search mode is counted. The CPU block 4 determines that it is the full search macroblock MB line, and specifies the full search mode for the determination result holding unit 7j in the macroblock MB line even if the narrow search mode is determined. Thus, in this macro block MB line, the ME block 7 operates in the full search mode.

[0090] search point counter 7i is a full search macroblock MB lines, the number of occurrences of motion vectors entering each search point is counted by the macro block MB line units, C the result as the search point counting result S ₉
Report to PU block 4. The CPU block 4
Search point counting result receives S _9, to specify the motion vector search when motion is vector number generated is equal to or greater than the threshold of the search point specified in the current picture, subsequently in the narrow search mode in the same search points. Otherwise,
From the next macroblock MB line in the current picture,
Specify search of motion vector in full search mode. The change of the search point in the narrow search mode in the picture or the change from the full search mode to the narrow search mode is not performed because it greatly affects the image quality.

Further, the CPU block 4 receives the intra / non-intra MB from the JUDGE block 10 (FIG. 2).
Has been received, and all the search macro blocks M
Even if it is other than the B line, when the ratio of the intra MB increases in macroblock MB line units, the mode is returned to the full search mode from the next macroblock MB line.

In this embodiment, the number of search points in the narrow search mode (narrow search point: narrow search range) is five.
However, the number of search points is not limited to this and may be 5
It may be more or less than the number.

In this embodiment, the narrow search points are set so as to partially overlap each other. However, a wide search range may be divided into a plurality of areas, and each of the divided areas may be set as a narrow search point.

Further, in this embodiment, one narrow search point can be used in one search, but a search may be performed using a plurality of narrow search points.

Furthermore, in this embodiment, the number of occurrences of motion vectors entering each search point is counted, and a narrow search point having a high count value is selected. An algorithm that does not search for a motion vector at a point is also conceivable.

By combining some of the above examples, for example, the search range in the full search mode is divided into 4 × 4 = 16 areas, each of which is a narrow search point. An embodiment in which the number of occurrences of motion vectors at each narrow search point is counted, a plurality of narrow search points are selected so as to include 95% or more of the motion vector, and the narrow search point is used for subsequent motion estimation is also considered. Can be

FIG. 10 is a block diagram showing one application example of the digital image encoding apparatus according to the present invention to a portable video camera. Reference numeral 18 denotes an MPEG2 encoder (a digital image encoding device according to the present invention), 19 denotes an imaging unit, 20 denotes a camera shake correction unit, 21 denotes a pan detection sensor, 22 denotes a CPU block, 23a denotes a recording unit using a magnetic tape, and 23b denotes a fixed disk device. Is a recording unit, and 24 is a CPU bus.

[0098] In the figure, in a portable electric device, since the power supply of the device depends on a battery, it is necessary to increase the battery life by reducing power consumption. In a portable video camera, as is often used in a conventional portable video camera, a pan detection sensor 803 detects panning of an image and corrects camera shake according to the detection result. Like that.

The pan detection sensor 21 detects pan information from the output image signal of the imaging section 19 and supplies the information to the camera shake correction block 20 and the CPU block 22. In addition, a gyro, an acceleration sensor, and the like may be built in as the pan detection sensor 21 so that the pan information is detected by these. The camera shake correction block 20 corrects the influence of the camera shake of the image signal based on the pan information, and the image signal with the camera shake corrected is supplied to the MPEG2 encoder 18. The image information encoded by the MPEG2 encoder 18 is recorded in the recording unit 23a or 23b.

Note that this MPEG2 encoder 18
It comprises an MPEG2 encoding device 2 and an SDRAM block 3 in FIG.

On the other hand, the pan information from the pan detection sensor 21 is analyzed by the CPU block 22, and an offset value of an appropriate search range in the ME block 7 (FIG. 1) in the MPEG2 encoder 18 is calculated. CPU
The data is supplied to the MPEG2 encoder 18 via the bus 24. Thereby, the MPEG2 encoder 18 can detect an appropriate motion vector corresponding to pan.

Further, the CPU block 22 stores the MPEG2
The data from the encoder 18 is fetched via the CPU bus 24 for calculation, and the limit of the search range of the motion vector in the MPEG2 encoder 18 is specified according to the calculation result. Thereby, low power consumption of the entire system can be achieved. Further, when it is necessary to urgently reduce the power consumption due to a shortage of the remaining battery capacity or the like, the CPU block 2
2 keeps the search range limited, thereby further reducing power consumption.

FIG. 11 is a block diagram showing an example of application of a digital image encoding apparatus according to the present invention to a remote control automatic camera, wherein 25 is a carriage, 26 is a camera, and 27 and 28.
Is a transmission unit, 29 is a control unit, 30 is a transmission line, and FIG.
The same reference numerals are given to the portions corresponding to and the duplicate description will be omitted.

In the figure, when the camera 26 is remotely operated, a video signal obtained by photographing with the camera 26 must be supplied to the control unit 29. However, there is a case where the image data PD output from the camera 26 cannot be transmitted as it is because the band of the transmission path 30 is narrow. In this application example, the output image data PD from the camera 26 is once supplied to the MPEG2 encoder 18 to encode and compress the amount of information, and then the transmission unit 27 is suitable for transmission on the transmission path 30. The signal is converted to the original signal form and transmitted through the transmission path 30, and the transmission section 28 converts the signal form to the original signal form and supplies it to the control section 29, where it is decoded and reproduced. Thus, the MPE
The G2 encoder 18 is effective for transmitting the image data PD through the transmission path 30 having a narrower band.

In this application example, the camera operation signal CO from the control unit 29 is processed by the transmission unit 28 and transmitted through the transmission path 30, and is processed by the transmission unit 27 to be processed by the camera operation signal CO '.
Is supplied to the pan head 25. Thereby, the control unit 29
, The direction of the camera 26 can be controlled.

Here, the camera operation signal CO 'is analyzed by the CPU block 22, and how the camera 26 moves is calculated in advance. In this case, the number of pixels panned from the previous picture to the current picture can be calculated based on the movement of the camera 26 and the focal length of the lens of the camera 26. Based on the calculation result, the MPEG2 encoder 1
8 and calculate the appropriate offset of the search range at M
It can be set in the PEG2 encoder 18. Thereby, the MPEG2 encoder 18 can detect an appropriate motion vector corresponding to pan.

In the above description, the MPEG2 encoding apparatus has been mainly described. However, for example, a camera shake of a digital image encoding apparatus or a portable video camera conforming to another standard such as MPEG1 or H.263. The present invention can be similarly applied to a device using a motion estimation device, such as a pan detection device of a correction device.

[0108]

As described above, according to the present invention,
A limited mode for narrowing the search range of the motion vector and a function for offsetting the search range are provided, and the offset value and the search range are operated based on the result of the past vector search, so that the deterioration of the image quality is minimized. It is possible to reduce the amount of ME operation that occupies most of the power consumption while keeping
As a result, the power consumption of the entire MPEG2 encoding device can be reduced.

According to the present invention, with respect to the optimization of the motion estimation by the external pan detecting device, the offset value and the search range are set by the external CPU, so that the motion estimation according to the pan detected externally is performed. Can be optimized.

[Brief description of the drawings]

FIG. 1 is a block diagram showing a first embodiment of a motion vector detection device and a digital image encoding device using the same according to the present invention.

FIG. 2 is a block diagram showing a first embodiment of a motion vector detecting device according to the present invention used as an ME block in the MPEG2 coding device shown in FIG. 1;

FIG. 3 is a distribution diagram of a motion vector when a correlation between an encoded image and a reference image is low.

FIG. 4 is a distribution diagram of a motion vector when a correlation between an encoded image and a reference image is high.

FIG. 5 is an explanatory diagram of a search range limitation and an offset when a correlation between an encoded image and a reference image is high.

FIG. 6 is an explanatory diagram in a case where a correlation between an encoded image and a reference image is low in a state where a search range is limited.

FIG. 7 is an explanatory diagram of scaling of a motion vector in the first embodiment shown in FIG. 1;

FIG. 8 shows a digital image encoding device shown in FIG.
Second of motion vector detecting devices usable as blocks
It is a block diagram showing an embodiment.

9 is an explanatory diagram of a search range for detecting a motion vector in the embodiment shown in FIG.

FIG. 10 is a block diagram illustrating an application example of a digital image encoding device according to the present invention.

FIG. 11 is a block diagram showing another application example of the digital image encoding device according to the present invention.

[Explanation of symbols]

 Reference Signs List 1 image data input terminal 2 MPEG2 encoding device 3 SDRAM 4 CPU block 5 bit stream output terminal 6 VIDEO block 7 ME block 7a input image buffer 7b reference image buffer 7c motion vector detector 7d reference image read address generator 7e value Holding unit, 7f Judgment counting unit 7g Average value calculation unit 7h Variance value calculation unit 7i Search point counter 7j Judgment result holding unit 8 CPU IF block 9 MC block 10 JUDGE block 11 DCT / IDCT block 12 Q / IQ block 13 RATE block 14 VLC block 15 CODE block 16 SDRAM address bus 17 SDRAM data bus 18 MPEG2 encoder 19 Imaging unit 20 Camera shake correction unit 21 Pan detection sensor 22 CPU 23 , 23b recording unit 25 pan head 26 camera 29 control unit 30 the transmission line

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Satoshi Motosu 292 Yoshida-cho, Totsuka-ku, Yokohama-shi, Kanagawa Prefecture Inside the multimedia system development headquarters of Hitachi, Ltd. (72) Tomoo Kobori Yoshida-cho, Totsuka-ku, Yokohama-shi, Kanagawa 292 Hitachi Multimedia Systems Development Division

Claims (6)

[Claims] 1. A digital image coding apparatus for compressing and coding a moving image, comprising: input means for inputting image data of the moving image; storage means for storing image data of a reference image; A search range of a motion vector is set for each image data of the moving image input from the means, and the image data of the reference image in the set search range is read from the storage means.
A motion vector detecting unit that detects a motion vector for the image data of the moving image input using the image data in the search range; and using the motion vector detected by the motion vector detecting unit, Encoding means for encoding the corresponding input image data of the moving image, and control means for varying the size of the search range set in the motion vector detecting means. Digital image coding device. 2. The reference image reading address generating means according to claim 1, wherein said motion vector detecting means generates an address signal for reading image data of said reference image in said storage means. Reference image buffer means for accumulating image data of the reference image read from the accumulating means in response to the image data of the search range; and an input for accumulating image data of the moving image from the input means. Image buffer means, image data of the reference image stored in the reference image buffer means and image data of the moving image stored in the input image buffer means are supplied, and the reference image is stored within the search range. While changing the relative position between the image data of the reference image and the input image data of the moving image, the image data of the reference image A prediction error is calculated between the image data of the moving image and the positional relationship between the image data of the reference image and the image data of the moving image in which the prediction error is minimized. Image position detecting means for generating and outputting a motion vector representing the positional relationship that minimizes the relationship, wherein the control means includes: the image data of the reference image detected by the image position detecting means; The relative position between the image data of the moving image and the size of the search range is determined using information corresponding to the relative position, and information indicating whether or not encoding has been performed from the encoding unit. And a reference range control means for outputting the control signal designating the determined size in accordance with the range. 3. Input image data of a moving image is input from an input unit in a desired block unit, and a motion vector is detected for each input image data in the desired block unit using image data of a reference image in a search range. A storage means for storing image data of a reference image, an address generation means for generating an address signal for designating an area from which the image data of the reference image is read by the storage means, Reference image buffer means for storing the read image data of the reference image and storing image data of the reference image in the search range; input image buffer means for storing the input image data of the desired block unit; The image data of the reference image in the search range is sent from the reference image buffer means to the desired block unit from the input image buffer means. Force image data, and sequentially changing the relative positional relationship between the image data of the reference image and the input image data of the desired block unit within the search range, while changing the relative image relationship for each of the relative positional relationships. Calculating a prediction error between the image data of the desired block unit and the input image data of the desired block unit, detecting the relative positional relationship in which the predicted error is minimized, representing the detected relative positional relationship, and representing the detected A motion vector detection unit that generates and outputs a motion vector for encoding the input image data of the above; a distribution of the prediction error for each of the relative positional relationships detected by the motion vector detection unit; Search range control means for generating a control signal for designating the size of the search range in accordance with information indicating whether or not the input image data has been used for encoding the desired block unit of input image data. Motion vector detecting apparatus characterized by a variable according the size of the search range set by the motion vector detection unit to the control signal. 4. The motion vector detection unit according to claim 3, wherein the motion vector detection unit has two types of operation modes, a wide range and a narrow range, as a range for moving the reference image up, down, left, and right, and searches in the narrow range. A motion vector detecting device having a plurality of positions. 5. The method according to claim 4, wherein a transition from detection of the reference image position in the narrow range to detection of the reference image position in the wide range is performed.
If it is set in advance, the detection of the reference image position in a wide range is unconditionally performed, and it is determined whether or not the detection of the reference image position at the set position is appropriate. Characteristic motion vector detection device. 6. The method according to claim 4, wherein, when the search range is a narrow range, the search range control unit adjusts the bias position of the distribution of the prediction error for each of the relative positional relationships detected by the motion vector detection unit. In response to,
A motion vector detecting device, wherein an offset is set at a position in the search range.