KR20010026899A - Apparatus for motion estimation so as to moving picture coding - Google Patents

Abstract

PURPOSE: A motion estimator for coding a motion picture is provided to expand a range of motion estimation by using a multitude of chip for motion estimation. CONSTITUTION: The first scan type converter(110) converts an original image to a vertical direction scan type. The second scan type converter(120) converts a locally decoded image to the vertical direction scan type. A reference data processor(140) stores the original data as the reference data and outputs the delayed reference data. A search data processor(160) stores the decoded image data and outputs search data. A motion estimator(180) performs a motion estimation process and outputs motion vectors. A motion vector processor(190) extracts only one motion vector of the motion vectors.

Description

Apparatus for motion estimation so as to moving picture coding}

The present invention relates to a motion estimation apparatus for video encoding. More particularly, the present invention provides a motion estimation apparatus using a plurality of well-known motion estimation dedicated chips to provide an easy extension of the motion estimation range, The present invention relates to a motion estimation apparatus for video encoding that is performed quickly and accurately.

At present, interest in video encoding has increased, and many video compression standards such as H.261, MPEG-1, and MPEG-2 have been enacted. These standards are applied to various applications such as video telephony, digital broadcasting, and high-definition television (HDTV), and most of them adopt complex motion compensation / discrete cosine coding as video encoding. In this way, motion compensation is very important because it effectively compresses large amounts of digital data by effectively reducing redundancy between successive images. In order to efficiently perform such motion compensation, a process called motion estimation is necessary as a prerequisite.

As shown in FIG. 1, a configuration of a general MPEG image encoder having a motion estimator necessary to efficiently perform motion compensation and increase a compression rate of image data in image encoding is shown.

Here, reference numeral 1 denotes a subtractor for calculating a difference between an input image and a predicted image for motion compensation, 2 denotes a discrete cosine transformer for discrete cosine transforming the output image of the subtractor 1, and 3 denotes a discrete cosine converter 2. The quantizer quantizes the data output from.

Further, reference numeral 4 is a variable length encoder for variable length coding the quantized data by the quantizer 3 and outputs it as a bit string, and 5 is a quantizer 3 according to the bit rate control signal output from the variable length encoder 4. Is a bit rate controller that controls the output bit rate of the.

Reference numeral 6 is an inverse quantizer 6 for inverse quantization of data output from the quantizer 3, and 7 is an inverse discrete cosine transformer for inverse discrete cosine transforming the data output from the inverse quantizer 6. , 8 is an adder that adds the output data of the inverse discrete cosine converter 7 and the predicted image for motion compensation, and 9 is a frame memory that stores and retrieves the output data of the adder 8.

Reference numeral 10 denotes a motion estimation unit for motion estimation of the input image using a previous image obtained from the frame memory 9, and reference numeral 11 denotes the frame memory 9 using a motion vector obtained from the motion estimation unit 10. The motion compensation unit obtains the prediction value of the image block to be currently processed from the previous image stored in).

The general MPEG image encoder configured as described above performs motion estimation using the input image and the previous image stored in the frame memory 9 for prediction.

The motion vectors obtained at this time are input to the variable length encoder 4 and the motion compensator 11, respectively, and the motion compensator 11 is currently processing from the previous image stored in the frame memory 9 using the motion vectors. The motion compensation prediction value of the image block is extracted.

The motion compensation prediction value thus obtained and the current input image are subtracted by the subtractor 1 to obtain the difference, and the difference is subjected to the discrete cosine transform (DCT) in the discrete cosine converter 2.

The DCT-encoded data is quantized in the quantizer 3 and then variably coded in the variable length encoder 4 to be output as a bit string. Inverse quantize image blocks.

In addition, the inverse discrete cosine transformer 7 performs inverse discrete cosine transform (IDCT) on the inverse quantized data, and the adder 8 adds a motion compensation prediction value to the image of the current image block that has been inverse discrete cosine transformed. The result is stored in frame memory 9 for motion estimation and compensation of the next input image block.

That is, compression of image data is performed through motion estimation and compensation, discrete cosine coding, variable length coding, and the like. Of these, motion estimation is a relatively simple task, but requires massive hardware (H / W).

Therefore, the general image coding system performs motion estimation using the ASIC chip manufactured for this purpose.

However, currently available motion estimation ASICs have a limited maximum search range of vectors. Therefore, in applications (MPEG-1, MPEG-2, etc.) that require a large search range, they must be implemented using a plurality of them. In particular, when the image resolution is very large, such as a high-definition TV, the motion estimation range is limited.

Accordingly, the present invention has been proposed to solve various problems occurring in the conventional motion estimation as described above.

The object of the present invention is to implement a motion estimation apparatus using a plurality of well-known motion estimation dedicated chips, thereby providing an easy extension of the motion estimation range, and to estimate the motion for moving picture encoding in which motion estimation is performed quickly and accurately. To provide a device.

Technical means for achieving the object of the present invention,

A first scan format converter for converting an original image to be encoded into a vertical scan format;

A second scan format converter for converting the locally decoded image into a vertical scan format;

The reference image data is stored as reference data in the original image data obtained by the first scan format converter, and the reference data extracted from the reference data storage is output with a time delay for synchronization with the search data. A processing unit;

A search data processor for storing the decoded image data obtained by the second scan format converter as search data in the search data storage, and extracting and outputting search data for motion estimation of the input original image from the search data storage; ;

A motion weight estimator comprising a plurality of motion estimators for estimating motion by comparing the reference data to the search data according to a control signal obtained from a motion estimation controller for generating a control signal for motion estimation and outputting a motion vector obtained as a result. With the government;

And a motion vector processor for extracting only one motion vector from among the motion vectors obtained by the plurality of motion estimators in the motion estimator and outputting the final motion vector.

The first scanning format converting unit may include: first and second memory modules configured to alternately write and read original image data input through two paths; A memory regulator for controlling the writing and reading of the first and second memory modules; And a data selector for selecting one of the original image data output from the first and second memory modules and outputting the converted scan format as data.

The second scanning format converter may include: first and second memory modules configured to alternately write and read locally decoded image data input through two paths; A memory regulator for controlling the writing and reading of the first and second memory modules; And a data selector configured to select one of the decoded image data respectively output from the first and second memory modules and output the converted scan format as the converted data.

In addition, one motion estimator in the motion estimator includes first and second registers for temporarily storing search data of two input paths and a plurality of registers for sequentially shifting reference data input through another path. A plurality of motion estimation dedicated to performing motion estimation by comparing a register section, search data of two paths output from the first and second registers and one reference data in the register section, and outputting motion data according to the result value. Characterized in that it consists of a chip module dedicated motion estimation consisting of a chip.

The motion vector processor may include: a motion vector processor configured to compare a prediction error of each motion vector obtained by the motion estimator, select a vector having a minimum error, and output the motion vector as a motion vector; And a time delay adjusting the delay time of the motion vector finally selected and output by the motion vector processor.

1 is a block diagram of a typical MPEG video encoder,

2 is a block diagram showing an embodiment of a motion estimation apparatus for video encoding according to the present invention;

3 is a block diagram illustrating an example of a scan type converter of FIG. 2;

4 is a diagram illustrating a signal configuration of a search data processor of FIG. 2;

5 is a memory structure diagram illustrating a memory allocation method for storing discovery data in the present invention;

6 is a diagram illustrating a signal configuration of a reference data processor of FIG. 2;

FIG. 7 is a block diagram illustrating an embodiment of the motion vector processor of FIG. 2;

FIG. 8 is a block diagram illustrating an embodiment of one motion estimator in one motion estimator of FIG. 2.

<Description of the code | symbol about the principal part of drawing>

110,120: first and second scanning format conversion unit

130: reference data storage unit

140: reference data processing unit

150: search data storage unit

160: search data processing unit

170: motion estimation control unit

180: motion estimation unit

190: motion vector processing unit

Hereinafter, a preferred embodiment of the present invention according to the technical spirit as described above will be described in detail with reference to the accompanying drawings.

2 is a block diagram of an embodiment of a motion estimation apparatus for video encoding according to the present invention.

Here, reference numeral 110 is a first scan format converter for converting an original image to be encoded into a vertical scan format, and 120 is a second scan format converter for converting a locally decoded image into a vertical scan format. In step 140, the original image data obtained by the first scan format converter 110 is stored as reference data in the reference data storage unit 130, and the reference data extracted from the reference data storage unit 130 is search data. It is a reference data processor that outputs with time delay for synchronization with.

In addition, reference numeral 160 stores the decoded image data obtained by the second scan format converter 120 as search data in the search data storage unit 150 and searches the search data for motion estimation of the input original image. The search data processor 150 extracts and outputs the data from the data storage unit 150.

In addition, reference numeral 170 is a motion estimation controller for generating a control signal for motion estimation, and reference numeral 180 is a motion estimation control unit for estimating motion with respect to the search data according to the control signal obtained from the motion estimation controller 170. And a motion estimator comprising a plurality of motion estimators 181 to 180 + N for outputting a motion vector obtained as a result, and reference numeral 190 denotes a plurality of motion estimators 181 to 180 + N in the motion estimator 180. ) Is a motion vector processing unit which extracts only one motion vector from each of the motion vectors obtained in the &quot;

In the motion estimation apparatus according to the present invention configured as described above, since the original input image and the locally decoded image are scanned in the horizontal direction, the first and second scan format converters 110 and 120 convert these scan formats. ) Into the vertical scan format.

Here, as illustrated in FIG. 3, the first scan format converter 110 for converting the scan format of the original image includes first and second alternately recording and reading the original image data input through the two paths. Memory modules 111 and 112; A memory regulator (113) for controlling the writing and reading of the first and second memory modules (111, 112); The data selector 114 selects one of the original image data output from each of the first and second memory modules 111 and 112 and outputs the converted data as a scan format.

Assuming that the size of the input image (original image) for motion estimation is 16 * 16 blocks which are basic units, a memory having a capacity of 256 bytes for processing one image block is required. In addition, in order to input the reference data used for the motion estimation to the motion estimation unit 180 at a later stage, the data must be converted into data in a vertical scanning format in units of blocks for reference in order to parallelize the data. In addition, in order to convert the horizontally scanned original image data in block units in the vertical direction, eight memories and a memory controller 113 for controlling the memories are required. In this case, two pairs of paths are used to process the motion estimation at high speed.

In more detail, while reading data from one memory module (eg, the first memory module) using the first and second memory modules 111 and 112, each of which consists of four memories, Original image data input to another module (for example, a second memory module) is recorded. Four memories (eg, 111a, 111b, 111c, and 111d) existing in each of the memory modules 111 and 112 are paired to process original image data of two paths inputted and outputted. That is, when writing data input to the first memory module 111, data stored in the second memory module 112 is read, and when writing data to the second memory module 112, the first memory module 111 is read. To read the data. At this time, in order to write data to one memory module 111, the upper two memories 111a and 111b and the lower two memories 111c and 111d of the first memory module 111 are alternately alternated every row scanned horizontally. Alternately use to write input data, and likewise, to read data from one memory module 111, the left two memories 111a and 111c and the right two memories 111b and 112d are alternated every column. Read by reading the stored data. The memory read / write control in each memory module as described above is controlled by generating an adjustment signal from the memory controller 113. The memory usage method is changed after processing of one image block. That is, when the first memory module 111 was previously used for storage, it is used for reading, and conversely, the second memory module 112 previously used for reading is used for storage. The data selector 114 selects and outputs one of the data output from the first and second memory modules 111 and 112, respectively. That is, when used for data storage in the first memory module 111 selects and outputs the data output from the second memory module 112, otherwise, if the second memory module 112 is used for data storage, 1 Data to be output from the memory module 111 is selected and output.

In this process, the scanning format of the original image data inputted is converted, and similarly to the first scanning format converting unit 110, the second scanning format converting unit 120 also includes the first and second memory modules; The memory controller and the data selector convert the scan format of the input decoded image.

Next, the reference data processor 140 stores the original image data converted from the scan format in the vertical direction by the first scan format converter 110 as the reference data in the reference data storage 130 and stores the reference data. When the reference data stored in the unit 130 is extracted and output, the delay time with the search data is adjusted and output. That is, as illustrated in FIG. 6, when the scan-converted original image data is input, the reference data processor 140 may store two memories 131 in the reference data storage 130 to adjust delay with the search area data. The data is alternately stored at 132 and input to the motion estimator 180 at an appropriate time. At this time, the data output from the reference data processor 140 is input one pixel in the vertical direction in the convex. When the reference data is input to the reference data processor 140, the necessary search data is data of a previous image, and thus are already stored in the search data storage 150. Therefore, since the memory for storing the reference data only serves to match the time with the search data, the size of the memory may be small.

Next, the search data processor 160 stores the locally decoded image data whose scan format is converted in the vertical direction by the second scan format converter 120 in the search data storage 150, and then stores the decoded image data in an appropriate timing. And extracts it into the motion estimator 180. Since the search area data and the current reference data differ by more than one frame in time, the search data must be stored in the frame memory. As illustrated in FIG. 4, the search data processor 160 stores input search data (decoded image data) in the first and second memory modules 151 and 152 in the search data storage 150. Then, it reads for motion estimation of the next input original image and transmits it to the motion estimation unit 180. In this case, the input search data is image data decoded locally by the image encoder, which is stored in one memory module of two memory modules 151 and 152 for motion estimation of the next image, and simultaneously the other memory module. Reads and outputs the stored search data to estimate the motion of the original image currently input. Each of the memory modules 151 and 152 in the search data storage 150 is one more than the number of motion estimators 181 to 180 + N in the motion estimation unit 180 used to extend the vertical motion estimation range. It consists of many (N + 1) memories. Therefore, each memory (eg, memory 1, memory 2, ..., memory N + 1) in each memory module stores the search data in units of a set (slice) of blocks in the horizontal direction, and simultaneously stores the search data. Will read. That is, in the case where the motion estimator 180 includes two motion estimators, if the vertical search range of the motion estimation dedicated chip is K pixels, the total search range is 2 * K. The search data will be read.

5 illustrates a method of using each memory in the search data storage unit 150 that stores and retrieves search data. Each memory module 151, 152 is composed of a total of (N + 1) memory, Figure 5 shows only the memory configuration in one memory module. The search data is stored in units of slices, which are sets of blocks in the horizontal direction in the image. For example, data of the first slice is stored in the memory 1, and data of the (N + 1) th slice is stored in the memory N + 1. Next, the (N + 2) th slice data is stored in the memory 1 again. This is because it considers reading data from memory. That is, (n + 1) slice data is read at the same time to expand the search range. At the same time, data of [1, 2, ..., N + 1] is read for motion estimation of blocks existing in N / 2th slice, and for motion estimation of next (N / 2 + 1) th slice. The search data of [2, 3, ..., N + 2] is read.

Next, the motion estimator 180 estimates a motion by comparing the reference data processed by the reference data processor 140 and the search data processed by the search data processor 160 and performs a motion vector according to the result value. Outputs N motion estimators (181 to 180 + N) were used to extend the vertical motion estimation range. In each motion estimator, the same reference data is input at the same time, search data of two paths is input, and motion estimation is performed. As in the description of FIG. 4, the search data for motion estimation for the block of the (N / 2) th slice is the data of the [1, 2, ..., N + 1] th slice that are simultaneously input, and the motion estimator 1st and 2nd slice data are input to 1 (181), 2nd and 3rd slice data are input to motion estimator 2 (182), and Nth and (N) to Nth motion estimator (180 + N). The +1) th slice data is input and motion estimation is performed.

Meanwhile, the motion estimation controller 170 provides an operation control signal of each motion estimation dedicated chip in each motion estimator 181 to 180 + N in the motion estimation unit 180.

Here, each motion estimator in the motion estimation unit 180 is as shown in FIG. 8.

That is, the first and second registers 210 and 220 temporarily store search data of two input paths and transfer the temporary search data to the motion estimation dedicated chip 241 of the motion estimation dedicated chip module 240. The register unit 230 sequentially shifts the reference data input through the two input paths and other paths into a plurality of registers 231 to 230 + N, and the output of one of the registers is the chip module dedicated to motion estimation. The reference data is transmitted to the motion estimation dedicated chip in response to one motion estimation dedicated chip in 240. Then, each motion estimation dedicated chip 241 to 240 + N of the motion estimation dedicated chip module 240 performs motion estimation by comparing search data of two input paths with reference data of another path, and according to the result value. Outputs a motion vector.

That is, search data corresponding to two slices are simultaneously input to the motion estimation dedicated chip, and one reference data is input thereto. At this time, a register for each data is used to input stabilized data. In addition, the motion estimation dedicated chip that performs the motion estimation for the motion estimation of the neighboring motion estimation dedicated chip first transmits the search data input to the next motion estimation dedicated chip. By performing motion estimation by connecting a plurality of motion estimation chips in series, the search range in the horizontal direction is expanded. In other words, if the search range of one motion estimation chip is K pixels, since N chips are used, the motion vector up to (K * N) pixels can be estimated.

Next, motion vectors estimated by each motion estimator in the motion estimator 180 are transmitted to the motion vector processor 190 as described above.

Then, the motion vector processor 190 compares the prediction error for each motion vector obtained by the motion estimators 181 to 180 + N in the motion vector processor 191 to determine a vector having the minimum error as the final motion vector. Decide on In addition, since the search areas of each motion estimation chip are different, the function of converting the size of the vector into the area of the corresponding chip is also performed. The time delayer 192 adjusts the delay time of the motion vector finally selected and output from the motion vector processor 191.

According to the present invention described above, by implementing a motion estimator using a plurality of motion estimation dedicated chips generally known, there is an advantage that provides an easy to expand the motion estimation range.

In addition, by implementing a motion estimator using a general motion estimation dedicated chip and a few programmable logic devices, there is an advantage in that the motion estimator can be implemented at low cost.

In addition, since the motion estimation range can be extended, it can be applied to applications such as high-definition television (HDTV) that require larger motion estimation.

Claims (5)

In the motion estimation apparatus for motion compensation of an image, A first scan format converter for converting an original image to be encoded into a vertical scan format; A second scan format converter for converting the locally decoded image into a vertical scan format; The reference image data is stored as reference data in the original image data obtained by the first scan format converter, and the reference data extracted from the reference data storage is output with a time delay for synchronization with the search data. A processing unit; A search data processor for storing the decoded image data obtained by the second scan format converter as search data in the search data storage, and extracting and outputting search data for motion estimation of the input original image from the search data storage; ; A motion weight estimator comprising a plurality of motion estimators for estimating motion by comparing the reference data to the search data according to a control signal obtained from a motion estimation controller for generating a control signal for motion estimation and outputting a motion vector obtained as a result. With the government; And a motion vector processor for extracting only one motion vector from among the motion vectors obtained by each of the plurality of motion estimators in the motion estimator and outputting the final motion vector. The method of claim 1, wherein the first scanning format conversion unit, First and second memory modules configured to alternately write and read original image data input through two paths; A memory regulator for controlling the writing and reading of the first and second memory modules; And a data selector configured to select one of the original image data output from each of the first and second memory modules, and output the converted data as a scan format. The method of claim 1, wherein the second scan format converter, First and second memory modules configured to alternately write and read locally decoded image data input through two paths; A memory regulator for controlling the writing and reading of the first and second memory modules; And a data selector configured to select one of the decoded image data respectively output from the first and second memory modules, and output the converted data as a scan format. The apparatus of claim 1, wherein the one motion estimator in the motion estimation unit sequentially shifts the first and second registers temporarily storing search data of two input paths and the reference data input through another path. A plurality of registers configured to perform motion estimation by comparing two paths of search data output from the first and second registers and one reference data in the register unit, and outputting motion data according to the result value Motion estimation apparatus for video encoding, characterized in that consisting of a motion estimation dedicated chip module consisting of three motion estimation dedicated chips. The method of claim 1, wherein the motion vector processing unit, A motion vector processor for comparing a prediction error with respect to each motion vector obtained from each motion estimator, selecting a vector having a minimum error and outputting the vector as a motion vector; And a time delay device configured to adjust a delay time of a motion vector finally selected and output by the motion vector processor.