KR100327952B1 - Method and Apparatus for Segmentation-based Video Compression Coding - Google Patents

Abstract

The present invention converts an analog video signal input through an image medium such as a video camera into a digital video signal, and then performs a region division for discriminating a background region and an object region, and uses the result to discriminate the background region and the object region. As a video compression encoding technique for performing compression encoding, a method and apparatus for obtaining a high compression rate while minimizing information loss in a target region are proposed.

The region segmentation based video compression encoder includes a current image frame memory, a background image frame memory, and a previous object image memory. The region segmentation based video compression encoder also has a region segmentation and encoding controller, a transform encoder, a transform decoder, an entropy encoder, and a bitstream generator. The region segmentation and encoding controller performs region segmentation of the background region or the object region for each unit region of the current image stored in the current image frame memory, and generates an encoding instruction and a motion vector differentially for each unit region by using the result. do. The transform encoder receives the encoding instruction and the motion vector for each unit region from the region segmentation and encoding controller and performs differential transform encoding on the image data of the unit region accordingly. The transform decoder performs transform decoding on the image data of the unit region transform-coded from the transform encoder to update a background image frame memory and the previous object image memory. The entropy coder performs entropy coding differentially on the transform-coded data from the transform encoder, the encoding instruction from the region segmentation and encoding controller, and the motion vector. The bitstream generator outputs the output from the entropy encoder in a bitstream in a predetermined format. According to the present invention, an image quality superior to the case of using the same size data storage space for an object area having important information as compared to the conventional compression encoding method can be realized.

Description

Region segmentation based video compression coding method and apparatus {Method and Apparatus for Segmentation-based Video Compression Coding}

The present invention relates to a compression encoding method of a video signal that can be effectively used for data compression of a video video signal in which a change of a background video does not frequently occur, such as an unmanned surveillance system, a video conference, a video call, or a video lecture. A splitting method and a compression coding method are provided.

In the field of video conferencing, video telephony, video lecture, etc. where video communication between remote areas is essential, H.26x series of techniques, which are already standardized as data compression method of video signals, are widely used in consideration of bandwidth of communication line. For the storage, broadcasting, and transmission of video signals in digital form, MPEG-1 and MPEG-2 techniques, which are standardized by the Motion Picture Expert Group (MPEG), are widely used. In most cases, unmanned surveillance systems using video signals have been used to transmit analog video signals as they are and store them on analog video tapes.However, by repeatedly storing images on tapes, image quality is reduced and the cost of tape replacement increases. In recent years, the use of surveillance systems using H.26x, MPEG or Joint Picture Expert Group (JPEG) video compression techniques has been increasingly used after converting analog video signals into digital signals.

Many video compression coding techniques developed so far for data compression of video signals divide the current video into a certain sized rectangular area (MB) and compare each area with the previous video to obtain motion compensation. After performing the method, compression encoding of the difference (difference or error) between the motion compensated image and the current image is performed.

This method has the following drawbacks.

First, the uncovered background, which is covered by the movement of the object and reappears, must be newly encoded.

Second, when the frame rate of the input video (that is, the speed at which the current video is input to the encoder) is low due to the characteristics of the video encoder, the motion prediction of the object in the adjacent video (or frame) becomes large and the motion prediction fails. The probability of doing so increases, which increases the amount of bits generated, which in turn reduces the data compression efficiency.

Third, in order to enable random access to the coded video, I-frames need to be inserted periodically, as is widely used in the MPEG coding scheme. Instead of encoding the difference from the previous image, as in encoding still images, only the current frame is independently compressed and encoded, so that the amount of bit generation increases.

Fourth, since the object region and the background region are not distinguished when encoding, an additional calculation is required to implement additional functions such as an alarm when an object appears.

Fifth, since it does not distinguish between object area and background area, it is impossible to control the quality of important areas such as lowering compression rate for relatively important object areas to improve image quality and lowering quality for less important background areas to increase compression ratio. .

An object of the present invention is to provide a method of compressing and encoding a current image using a background image and a previous image for efficient compression of a video image signal.

Another object of the present invention is to divide an input image into unit areas having a constant size, and then determine whether each unit area belongs to a background area or an object area, and segmentation and encoding to generate differential encoding instructions for each area. To provide a controller.

Another object of the present invention is to increase the compression rate by performing differential compression encoding for each region of the region-divided input image.

BRIEF DESCRIPTION OF THE DRAWINGS Fig. 1 is a diagram showing the overall configuration of an area segmentation based video compression encoder and its input / output

FIG. 2 is a diagram showing a detailed configuration of an area segmentation and coding controller (SCC) and input / output thereof shown in FIG.

FIG. 3 is a diagram showing the detailed configuration of the transform encoder TC shown in FIG. 1 and its input / output. FIG.

FIG. 4 is a diagram showing the detailed configuration of the transform decoder TD shown in FIG. 1 and its input / output.

FIG. 5 is a diagram showing a background area discrimination procedure and output thereof of the background area discriminator BAC shown in FIG.

FIG. 6 is a diagram illustrating a normality determination procedure and its output of the normality analyzer SA shown in FIG.

<Description of the symbols for the main parts of the drawings>

10: video compression encoder 12: video frame memory

14: Background image frame memory 16: Previous object image memory

18: Entropy Encoder 19: Bitstream Builder

20: segmentation and incubation controller 30: transform encoder

40: conversion decoder 50: video input source

52: network 54: storage medium

The following two terms are defined to describe the construction and operation of the present invention.

1. Macro-block (MB): In the present invention, when the image is divided into predetermined rectangular unit areas having a predetermined size as in the following example, each unit area is defined as MB (macro block). The size of the macro block MB is preferably 16 pixels in width and width, and other sizes may be used if necessary.

2. Macro-block information (MB_Info): In the present invention, information on how to encode each macro block MB of the current video is defined as macro block information MB_Info (macro block information). Each macro block MB of the current image has its own macro block information MB_Info. Specifically, the macro block information MB_Info includes a motion vector generated as a result of motion estimation and a coding directive for performing differential encoding in units of a macro block MB. The motion vector and the encoding instruction will be described later. In the present invention, the communication between the region splitter and the transform encoder is performed through this macro block information MB_Info.

The above object of the present invention is to divide a current image into a predetermined unit area (MB; Macro-Block), and to separately compress and encode each unit area according to the similarity between the stored background image and the stored object image. Segmentation-based Video Compression Encoder (SVC Encoder)

A current picture frame memory (CFM) for storing a current picture;

A background image frame memory (BFM) for storing a background image;

A previous object image memory (POM) for storing a previous object image;

Region segmentation and encoding for segmenting the current image stored in the current image frame memory into a background region or an object region for each unit region and generating differential coding instructions and motion vectors for each unit region using the result. Segmentation and Coding Control (SCC);

A transform coder (TC) for receiving a coding instruction and a motion vector for each unit region from the region segmentation and encoding controller and performing differential transform encoding on image data of a unit region according to the input signal;

A transform decoder (TD) for performing transform decoding on the image data of the unit region transform-coded from the transform encoder to update the background image frame memory and the previous object image memory;

An entropy coder (EC) for differentially performing entropy coding on the transform-coded data from the transform encoder, the encoding instruction from the region division and encoding controller, and a motion vector;

A bitstream formatter (BF) for outputting the output from the entropy encoder into a bitstream according to a predetermined format;

It is achieved by a segmentation-based video compression encoder (SVC Encoder) characterized in that it comprises a.

According to another aspect of the present invention, the current image is divided into a predetermined unit area (MB), and each unit area is stored in a background frame memory (BFM) and an object image memory (POM). : A segmentation-based video compression encoding method that differentially encodes and compresses according to similarity with an object image stored in a previous object memory.

Determining whether the unit is a background area or an object area with respect to the current image;

Generating a motion vector by differential coding instruction and motion prediction for each unit region;

Performing differential transform encoding on image data of a unit region according to the encoding instruction for each unit region;

Differentially performing entropy coding on the transform coded data, the encoding indication, and the motion vector.

An area segmentation based video compression encoding method is provided.

Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the accompanying drawings.

Referring to FIG. 1, an overall configuration and an input / output of a Segmentation-based Video Compression Encoder (SVC Encoder) 10 according to an embodiment of the present invention are shown.

The video compression encoder (SVC) 10 determines whether it is a background area or an object area in units of macro blocks (MB) with respect to the received current image, and segmentation and encoding for generating differentiated macro block information (MB_Info) for each area. A controller (SCC) 20; A current image frame memory (CFM) 12 for storing a current image frame input from an image input source VS 50; A background image frame memory (BFM) 14 for storing a background image; A previous object image memory (POM) 16 for storing a previous image; A transform encoder (TC) 30 for performing differential encoding on macro block (MB) data of a current image according to the macro block information (MB_Info); An entropy encoder (EC) 18 for entropy encoding the output data from the transform encoder 30 and the macro block information MB_Info; A bitstream generator (BF) 19 for converting the output data from the entropy encoder 18 into a bitstream according to a specific format; A transform decoder (TD) 40 for decoding the output data from the transform encoder 30 to update the background image frame memory (BFM) 14 and the previous object image memory (POM) 16. .

The input to the video compression encoder (SVC) 10 is current video data from a digital video input source (VS) such as a video frame grabber. The output from the video compression encoder (SVC) 10 is a bitstream including compressed and coded data about the input current video data and information necessary for reconstructing the current video using the data. The output bitstream may be stored in a storage medium 54 such as a hard disk or a magnetic tape, if necessary, or transmitted through a network 52.

1, a current image frame memory (CFM) 12, a background image frame memory (BFM) 14, and a previous object image memory (POM) 16 are described. Are both memories that store image data. The current image frame memory (CFM) 12 is a current image frame memory that stores a current image frame input from the image input source VS 50. The background image frame memory (BFM) 14 is a background image frame memory that stores a background image extracted from image data until immediately before current image data is input. The previous object image memory (POM) 16 is a previous object image memory for storing an object image extracted from image data immediately before the current image data.

Segmentation and Coding Control (SCC) 20 determines whether each macroblock (MB) of the current image data stored in the current image frame memory (CFM) 12 is a background image or an object image. After region division is performed, macroblock information MB_Info, which is information necessary for encoding each macroblock MB data, is generated in units of macroblocks using the result. The input of the region division and encoding controller (SCC) 20 includes a current image stored in the current image frame memory (CFM) 12, a background image stored in the background image frame memory (BFM) 14, and a previous object image memory (POM). 16) is the previous object image stored in. The area division and encoding controller (SCC) 20 compares the current image with the background image and the object image, respectively, and performs area division in units of macro blocks (MB). The output of the region division and coding controller (SCC) 20 is macro block information MB_Info. The macro block information MB_Info is composed of a motion vector (mv) generated as a result of motion estimation and a coding directive (cd: coding directive) for performing differential coding in units of macro blocks (MB). do.

Each macroblock (MB) data of the current image is transform-coded by a transform coder (TC) 30 according to the encoding instruction, and the result is entropy coding by an entropy coder (EC) 18. ) do. The data encoded by the transform encoder TC 30 may also be decoded by a transform decoder 40 according to the encoding instruction to be used for the background image frame memory BFM 14 and the previous object image memory POM. 16) used to update.

The macro block information MB_Info is also entropy coded in the entropy encoder EC 18 and included in the compressed bitstream by the bitstream generator BF 19. A detailed internal configuration and operation of the region division and coding controller (SCC) 20 that performs the above operation will be described with reference to FIGS. 2 and 5, 6, 5, 6, and 7 below.

Macro block information MB_Info will be described with reference to FIGS. 1 and 2 and Table 1 below.

As described above, the macro block information MB_Info includes a motion vector (mv: motion vector) generated as a result of motion estimation and an encoding instruction (cd :) for performing differential encoding in units of macro blocks (MB). coding directive).

The encoding instruction cd is generated as a result of the segmentation in the region division and the encoding controller SCC 20, and controls the transform encoder TC 30 and the entropy encoder EC 18 to control the macroblock. Differential coding in (MB) units, and the transform decoder (TD) 40 is controlled to perform differential decoding in units of macro blocks (MBs).

Table 1 below summarizes descriptions of six types of encoding instructions B_SKIP, B_INTER, B_UPDT, O_SKIP, O_INTER, and O_INTRA according to an embodiment of the present invention.

TABLE 1

Types of Coding Directives (cd) and Their Meanings

* 'X' in the above table ignores motion vectors and 'O' uses motion vectors

It is displayed.

** 'Location' in the above table means the current image program of the input macro block (MB).

It refers to the same position as the position on the frame.

First, B_SKIP (Background-referenced Skip) has a bit flag of '100000', and macroblock (MB) data of the current image inputted from the current image frame memory (CFM) 12 is stored in the background image frame memory (B_SKIP). BFM: There is little difference from the macroblock (MB) data in the same position on the 14), which is a directive not to perform the transform encoding.

Background-referenced inter-coding (B_INTER) has a bit flag of '010000', and macroblock (MB) data of the current image input from the current image frame memory (CFM) 12 and a background image frame memory (BFM) 14. Inter-inter encoding is performed on the difference data from the macroblock (MB) data at the same position on the image.

O_SKIP (Object-referenced Skip) has '001000' as a bit flag, and macro block (MB) data of the current image input from the current image frame memory (CFM) 12 is almost different from previous object image data by motion prediction. Since there is no, the instruction is to encode only the motion vector mv without performing transform encoding.

O_INTER (Object-referenced Inter-coding) has '000001' as a bit flag, and macroblock (MB) data of the current image inputted from the current image frame memory (CFM) 12 and previous object image data by motion prediction. Instructs to inter-convert the difference and encode the motion vector.

O-INTRA (Object-referenced Intra-coding) has a bit flag of '000010' and is an instruction to perform intra-contrast encoding of macroblock (MB) data of a current image input from a current image frame memory (CFM) 12.

Background Update (B_UPDT) has a bit flag of '000100' and has a macro on a previous object image memory (POM) 16 having the same position as the macro block MB of the current image input from the current image frame memory (CFM) 12. It is an instruction to update the background image frame memory BFM 14 by copying the block MB data to the macro block MB at the same position on the background image frame memory BFM 14. This B_UPDT means that the macroblock (MB) data of the current image inputted from the current image frame memory (CFM) 12 is the same as the macroblock (MB) data of the same position on the previous object image memory (POM) 16 for a predetermined time. Alternatively, since there is almost no difference while the predetermined number of frames have elapsed, it means that the macro frame (MB) data at the same position on the previous object image memory (POM) 16 is used as the background image from the subsequent frame.

In the above description, the bit flag is introduced in one embodiment of the present invention as a method of transmitting and receiving an encoding indication (cd) between the devices shown in FIG. 1, and only one bit of six bits is provided according to the encoding indication. It is set to '1' and the remaining bits are set to '0'. However, the method of transmitting and receiving an encoding instruction in the device implementation according to the present invention is not limited thereto, and other methods may be introduced.

Referring to Table 1, when the encoding instruction is B_SKIP, B_INTER, B_UPDT, or O_INTRA, the motion vector mv is not used. Even though it is logically assumed that a motion vector having a value of (0,0) is used. It's okay. Therefore, when the region division and encoding controller (SCC) 20 outputs B_SKIP, B_INTER, B_UPDT, or O_INTRA as an encoding instruction, the region division and encoding controller (SCC) 20 outputs (0,0) as a motion vector. do.

Referring back to FIG. 1, the transform encoder TC 30 receives macroblock information MB_Info from the region division and encoding controller SCC 20, and encodes an instruction included in the macroblock information MB_Info. After performing transform and quantization for each macroblock MB according to (cd), the result is output to the entropy encoder EC 18 and the transform decoder TD 40.

An internal configuration and operation of the transform encoder TC 30 that performs the above operation will be described with reference to FIGS. 3 and 2.

TABLE 2

Connection state of the internal switch 32 of the transform encoder TC 30 according to the encoding instruction

Table 2 shows a connection state of the internal switch 32 of the transform encoder TC 30 according to the encoding instruction cd. According to the connection state of the internal switch 32, the contents of the work performed by the transform encoder TC 30 are different.

Referring to FIG. 3 and Table 2, when the encoding instruction is B_INTER, the internal switch 32 is connected to the terminal A, in the case of O_INTRA, it is connected to the terminal B, and in the case of O_INTER, it is connected to the terminal C. Referring to Table 2, for the other coding instructions except for B_INTER, O_INTRA, and O_INTER, the terminal of the internal switch 32 may be connected to any of A, B, and C, except for B_INTER, O_INTRA, and O_INTER. For the remaining encoding instructions, macroblock (MB) data from the transform encoder TC 30 (ie, the quantizer Q in FIG. 3) by the entropy encoder EC 18 next to the transform encoder TC 30. (36) output quantization coefficients) are ignored anyway.

An operation of the transform encoder TC 30 according to the connection state of the internal switch 32 of the transform encoder TC 30 will be described with reference to FIG. 3. When the internal switch 32 is connected to the terminal A, the current image macro block (MB) data from the current image frame memory (CFM) 12 and the macro block (MB) at the same position on the background image frame memory (BFM) 14 The difference from the data is input to a transformer T 34. When connected to the terminal B, the current image macro block (MB) data from the current image frame memory (CFM) 12 is directly input to the converter T (34). When connected to terminal C, the difference between the current image macro block (MB) data from the current image frame memory (CFM) 12 and the motion compensated previous object image from the motion compensator (MC) 38 (ie, motion prediction error) Data) is input to a converter T 34. For each of the internal switches connected to terminals A, B, and C, the output data of the internal switch is transformed via a transformer (T) 34 and quantized via a quantizer (Q) 36 to transform encoder (TC; 30) It is output to the outside. The converter (T) 34 is for converting image data in a spatial domain into image data in a frequency domain, and includes a discrete cosine transform (DCT) or a discrete wavelet transform (DWT). Data conversion, which is frequently used for encoding an image, such as discrete wavelet transformation.

The number of quantization tables required by the quantizer (Q) 36 according to an embodiment of the present invention is all three, and is used when the encoding instructions are B_INTER, O_INTRA, and O_INTER, respectively. If the encoding instruction is O_INTRA or O_INTER, the Intra quantization table and the Inter quantization table, which are frequently used for video encoding, are used, respectively. If the encoding instruction is B_INTER, the same quantization table as in the case of O_INTER may be used, or a new quantization table may be devised and used.

Referring back to FIG. 1, the transform decoder (TD) 40 receives the macro block (MB) unit image data and macro block information (MB_Info) that are transform-coded from the transform encoder (TC) 30, and receives the macro block. After performing inverse quantization and inverse transform for each macroblock MB according to the encoding instruction cd included in the information MB_Info, a background image frame memory (BFM) 14 is used using the result. And the previous object image memory (POM) 16 are updated.

An internal configuration and operation of the transform decoder TD 40 that performs the above operation will be described with reference to FIGS. 4 and 3. The operation of the transform decoder (TD) 40 is basically an inverse process of the operation of the transform encoder (TC) 30, and the encoding instructions used for transform encoding in the transform encoder (TC) 30 are converted to the transform decoder (T). TD 40) is used for inverse transform decoding.

Table 3 below shows the connection state of the internal switches 44, 49, 46, and 47 of the transform decoder TD according to the encoding instruction cd.

TABLE 3

Internal switch connection state of the transform decoder (TD) 40 according to the encoding instruction

Depending on the connection state of the internal switches 44, 49, 46, 47, the contents of the work performed by the transform decoder (TD) 40 vary.

4 and 3, when the encoding instruction is B_SKIP, the terminal B2 and the terminal A4 are connected to the background image frame memory (BFM) 14, and the restored current image frame memory (RFM) 48. Data is transmitted in the order of.

When the encoding instruction is B_INTER, the terminal A1, the terminal A2, and the terminal B4 are connected to each other and the background and the data passed through the inverse quantizer (Q ^-1 , Inverse Quantizer; 42) and the inverse transformer (T ^-1 , Inverse Transformer; 43). Data from the image frame memory (BFM) 14 is added and transferred to the background image frame memory (BFM) 14 and the restored current image frame memory (RFM) 42.

If the encoding instruction is O_INTRA, the terminal B1 and the terminal C4 are connected, and the current image frame memory (RFM) in which the data passed through the inverse quantizer (Q ^-1 ; 42) and the inverse transformer (T ^-1 ; 43) is immediately restored. (42).

Subsequently, when the encoding instruction is O_INTER, the terminal C1, the terminal A3, and the terminal D4 are connected to each other, and the data passed through the inverse quantizers Q ^{-1 and} 42 and the inverse transformer T ^{-1 and} 43 and the previous object image. Data from the memory (POM) 16 and the motion compensator (MC) 45 are added and transmitted to the restored current image frame memory (RFM) 42.

When the encoding instruction is O_SKIP, the terminal B3 and the terminal E4 are connected to each other so that the data from the previous object image memory (POM) 16 and the motion compensator (MC) 45 are directly restored and the current image frame memory (RFM) 42 Is sent to.

When the encoding instruction is B_UPDT, the terminals C3 and F4 are connected so that data from the previous object image memory (POM) 16 and the motion compensator (MC) 45 is reconstructed from the background image frame memory (BFM) 14. Current image frame memory (RFM) 42 is transmitted.

The data stored in the restored current image frame memory (RFM) 42 is transferred to the previous object image memory (POM) 16 and updated.

Referring back to FIG. 1, an entropy encoder EC 18 converts macroblock (MB) unit image data (specifically, quantization coefficients) and macroblock information (MB_Info) transform-coded from the transform encoder TC 30. Receives entropy coding and outputs the result to a bitstream formatter (BF) 19 (Bitstream Formatter).

The target data to be entropy encoded depends on the encoding instruction cd included in the macro block information MB_Info. Table 4 shows entropy encoding targets of the entropy encoder (EC) 18 according to the encoding instruction.

TABLE 4

Entropy encoding target of the entropy encoder (EC) 18 according to the encoding instruction

* 'O' in the table indicates that entropy encoding is performed on the data, and 'X'

Indicates that the data is ignored and entropy encoding is not performed.

The specific entropy encoding method may vary depending on the characteristics of the entropy encoding target data. That is, the Huffman coding method is preferable as the method for entropy coding the coding instruction, and the method for entropy coding the image data (that is, the quantization coefficients) in units of macroblocks (MB) is zigzag-scanning. Modified Huffman coding or arithmetic coding of the result of run-length coding by scanning is preferable, and the modified Huffman is a method of encoding a motion vector. A modified-Huffman coding method or an arithmetic coding method is preferable. The coding instruction cd may perform fixed length coding (FLC) without entropy coding as necessary. Details of the entropy encoding method are omitted since they are well known to those skilled in the art.

1, the bitstream generator BF 19 is a bitstream formatter that outputs the output from the entropy encoder EC in a bitstream form according to a predetermined format. The bitstream syntax may vary depending on the intended use of the present invention—an embodiment of the present invention.

Hereinafter, the internal configuration and operation of the region division and coding controller (SCC) 20 will be described in detail with reference to FIGS. 2, 5, 6, and 7.

Referring to FIG. 2, the background area discriminator (BAC) 22 compares the current image frame memory (CFM) 12 and the background image frame memory (BFM) 14 in units of macro blocks (MB) to compare three bits of information. Only one bit of these three bits is set to '1' and all the remaining bits are set to '0' and output. Bit 1 of the output three bits of the background region discriminator (BAC) 22 is connected to the first bit of the encoding instruction (cd), bit 2 is connected to the second bit of the encoding instruction, and bit 3 is the movement. It is connected to a predictor (ME) 24 to indicate whether to perform motion prediction. Table 5 shows conditions of occurrence of output bits of the background area discriminator (BAC) 22 and corresponding coding instructions cd.

TABLE 5

Condition of occurrence of output bit of background region discriminator (BAC) 22 and encoding thereof

Instruction (cd)

As a specific method of determining which of the generation conditions shown in Table 5, the current image macroblock MB received from the current image frame memory (CFM) 12 may be implemented by various methods. One of the methods, that is, the similarity between the received macroblock MB and the macroblock MB at the same position on the background image frame memory BFM 14, determines which of the conditions in Table 5 corresponds to. One method is shown in FIG.

In FIG. 5, Yerror denotes a value obtained by subtracting the brightness value of the pixel at the corresponding position of the background image macro block MB from the brightness value of each pixel of the current image macro block MB received as the brightness component error. Uerror and Verror are both color difference component errors. The color difference U-component of each pixel of the current image macro block MB received and the color difference U- of the pixel at the corresponding position of the background image macro block MB in the color difference V-component value. The values obtained by subtracting the component and color difference V-component values, respectively. In general, the color difference component may be composed of two components, a U-component and a V-component, or may be composed of two components, a Cb-component and a Cr-component, and two components, an I-component and a Q-component. It may be configured as. Which one of the above three methods is selected in distinguishing the chrominance components may vary depending on the needs of the implementation of the present invention.In the description of one embodiment of the present invention, for convenience of description, the U-component and the V-component are described. The division method of was selected.

In FIG. 5, Ty and Tyy are both thresholds for the brightness component, and Tu and Tv are thresholds for the color difference U-component and the color difference V-component, respectively. Ty, Tu, and Tv all represent negligible differences (eg, 1-7), and Tyy cannot be ignored, but not so large (eg, 10-40). In addition, Tn1, Tn2, and Tn3 are thresholds for the number of pixels, and are preferably selected to be about 90 to 99% of the total number of pixels belonging to one macro block MB.

In FIG. 5, condition 1 is to check whether n1 is sufficiently large (n1 is greater than a specific threshold Tn1) when the number of pixels whose brightness component error is smaller than Tyy is n1. If condition 1 is satisfied, the difference between the brightness components of the current image macro block (MB) data from the current image frame memory (CFM) 12 and the background image macro block (MB) data from the background image frame memory (BFM) 14 is different. Since the brightness components of the two macroblock (MB) data are similar, the brightness components of the two macroblock (MB) data are similar, and the brightness components of the two macroblock (MB) data are similar.

5, if the condition 1 is not satisfied, the output bit flag is set as '001' to compare the current image macro block (MB) data with the previous object image stored in the previous object image memory (POM) 16. Will print '. If condition 1 is satisfied, condition 2 further examines the similarity between the current image macro block (MB) data and the background image macro block (MB) data.

In FIG. 5, condition 2 is to check whether n2 is large enough (that is, n2 is larger than a specific threshold Tn2) after obtaining the number n2 of pixels with small error with respect to the chrominance component. If condition 2 is satisfied, the error of the chrominance component is small enough to be ignored. If condition 2 is not satisfied, the error of the chrominance component cannot be ignored. Referring to FIG. 5, when condition 2 is not satisfied, '010' is output as an output bit flag in order to inter-code a difference between the current image macroblock (MB) data and the background image macroblock (MB) data. If condition 2 is satisfied, condition 3 examines the similarity between the current image macro block (MB) data and the background image macro block (MB) data in more detail.

In FIG. 5, condition 3 is to check whether n3 is sufficiently large (that is, n3 is greater than a specific threshold Tn3) when the number of pixels whose brightness component error is smaller than Ty (<Tyy) is n3. Referring to FIG. 5, when condition 3 is satisfied, since the error of the color difference component and the brightness component are both small enough to be ignored, '100' is output as the output bit flag so as not to perform transform encoding (B_SKIP). However, if condition 3 is not satisfied, the error of the chrominance component can be ignored but the error of the brightness component cannot be ignored. Therefore, to inter-encode the difference between the current image macroblock (MB) data and the background image macroblock (MB) data. (B_INTER) Output '010' as the output bit flag.

The determination procedure is for explanation, and the present invention is not limited to the determination procedure. This determination procedure can be added and modified as necessary.

Referring back to FIG. 2, the motion estimator (ME) 24 is enabled only when the value of bit 3 of the output bits from the background area discriminator 22 is '1'. It becomes If bit 3 of the output bits from the background area discriminator (BAC) 22 is '1', it is determined that the input current image macroblock MB does not belong to the background area. ) Performs motion estimation by comparing the input current image macro block MB and the previous object image memory 16, and according to the result, the switch 23 of the next stage of the motion predictor ME 24. Controls the connection of and outputs a motion vector. That is, when the motion prediction error is so small that it is not necessary to perform transform coding, the motion predictor (ME) 24 controls the next switch 23 to enable the normality analyzer SA 27. If the motion prediction error is not negligible, the motion predictor (ME) 24 controls the switch 23 of the next stage to enable the intra / inter mode discriminator (IID) 28. State and outputs the prediction error image macroblock (MB) data to the Intra / Inter mode discriminator (IID) 28.

As a specific method of predicting the motion in the motion predictor (ME) 24, one of the block matching algorithms (BMA) generally used in the field of the present invention may be selected and used, or a new method may be used. Can be. However, when used in an embodiment of the present invention, the search area of the motion prediction technique is different from the general case. That is, in the general case, a rectangular area having a specific size including the macro block MB is used as a search area around the position of the macro block MB received from the previous frame memory area for motion estimation. In the present invention, not all rectangular areas having a specific size including the macro block MB are located as the search area around the position of the received macro block MB in the area of the previous object image memory POM, but among the areas. The search is performed only on the object area. Therefore, the motion predictor (ME) 24 according to an embodiment of the present invention further includes an object mask memory (OMM) 25 for storing the position of the region identified as the object in the previous frame, in addition to the general motion prediction function. do. The object mask memory 25 stores one bit of mask information indicating whether or not the object area corresponds to each macro block MB. The value of the mask per macroblock MB of the object mask memory (OMM) 25 is equal to the value of the third bit among the output bits from the background area discriminator (BAC) 22. As shown in, if the third output bit of the background area discriminator (BAC) 22 is '1', the macroblock (MB) of the current image received is essentially the object area (cd = O_SKIP, O_INTRA, O_INTER) or the object area. This is because the same area (cd = B_UPDT) will be determined. In the above embodiment, it is assumed that the motion prediction is performed in units of macro blocks (MB). However, the macro block (MB) is further divided (for example, 4 equal parts) as needed to predict motions for each divided region and generate a motion vector. You may.

The description of one embodiment of the present invention has been made under the assumption that a background image and a previous object image are already secured in each of the background image frame memory (BFM) 14 and the previous object image memory (POM) 16. In practice, the stationary and time-varying regions are distinguished as time passes, and the normal region is the background image frame memory (BFM) 14, and the time-variable region is the previous object image memory. (POM; 16). The basis for updating the background image frame memory (BFM) 14 to the normal region by analyzing the normality is as follows. If the macro block MB at a specific position of the current image frame memory (CFM) 12 is not similar to the macro block MB at the same position on the background image frame memory (BFM) 14, it belongs to the background area. If not, if it remains unchanged for a sufficient time (i.e. if the same macro block (MB) image continues to appear in the same position even if new current images are continuously input), this macro block (MB) is spaced in the object constituting the background. It can be said that a normal change or a newly appearing object does not disappear and remains in the background, so that the macroblock (MB) that has not changed for a sufficient time as a result of the normality analysis is stored in the background image frame memory (BFM) 14. The next frame will be used as the background image. Using this method, the entire first input image frame (possibly a mixture of object and background regions) is stored in the background image frame memory (BFM) 14 (ie as a background image) and an arbitrary value. If enough time has elapsed even after allocating them to the previous object image memory (POM) 16, the background image is secured in the background image frame memory (BFM) 14 and the object image is correctly secured in the previous object image memory (POM) 16. do. Also, even if there is a spatial change in the object constituting the background or the scene itself changes, the background image is stored in the background image frame memory (BFM) 14, and the previous object image memory (POM) 16 is sufficient. ), The image of the object is properly secured.

In the SCR 20 according to the present invention, a timer memory 26 connected to a stationarity analyzer 27 for normality analysis is shown in FIG. 2. It is equipped together. The normality analyzer (SA) 27 receives an enable signal and a motion vector from the motion predictor (ME) 24 and references a value of a timer memory corresponding to the position of the current image macro block (MB). Outputs bit information. The timer memory TM 26 stores a time for which the normality is maintained for each macro block MB, that is, a time when the same macro block MB image appears at the same position continuously.

Table 6 shows the generation conditions of the output bits of the normality analyzer (SA) 27 and the encoding instructions accordingly, and FIG. 6 shows the normality determination procedure and its output.

TABLE 6

Condition of occurrence of output bit of normality analyzer (SA) 27 and encoding instruction (cd) accordingly

As mentioned in the foregoing description of the motion predictor (ME) 24, the motion predictor (ME) 24 may generate an enable signal to the normality analyzer (SA) 27 only when the motion prediction error is negligible. Given. 2, 6 and Table 6, when the normality analyzer (SA) 27 receives an operable signal, the normality analyzer (SA) 27 firstly receives the motion vector received from the motion predictor (ME) 24. Check if the value of is (0,0), and if it is not (0,0), output '1' and '0' to bit1 and bit2 (that is, O_SKIP as encoding instruction), and (0,0) In case of, the value of the corresponding position of the timer memory (TM) 26 corresponding to the position of the current image macro block MB is read, and if the value is greater than a specific time T _t (eg, 10 seconds), the timer memory (TM) 26 is read. ) Resets the value of the corresponding position to '0' and then outputs '0' and '1' to bit1 and bit2 (that is, B_UPDT as an encoding instruction), and if it is less than T _t , the bit1 and bit2 '1' and '0' are output (that is, O_SKIP as an encoding instruction). If the normality analyzer (SA) 27 does not receive an operable signal, it resets the value of the timer memory (TM) 26 corresponding to the position of the current image macro block (MB) to '0' and resets the bit1 and Outputs '0' to all bit2s.

Referring back to FIG. 2, when it is determined that the prediction error cannot be ignored as a result of the motion prediction in the motion predictor (ME) 24, the motion predictor (ME) 24 determines the intra / inter mode. An enable signal is given to the IID 28. Intra / Inter Mode Decision (IID) 28 receives the prediction error image macro block (MB) data from the motion predictor (ME) 24 and displays the current image frame memory, as shown in Table 7 below. (CFM) 12 to receive the current image macroblock (MB) data and compare each statistical characteristic (e.g., variance) to inter-mode encoding (i.e., to encode prediction error macroblock (MB) data) or It is determined whether to perform intra mode encoding (that is, encoding current image macroblock (MB) data), and the result is output as two bits of information.

TABLE 7

Condition of occurrence of output bit of Intra / Inter mode discriminator (IID) 28 and encoding instruction (cd) accordingly

Statistical characteristics for the determination of the Intra / Inter mode are as follows.

Use the sum of the absolute values ( A ) of the difference between the mean of the data ( mean ) and the value of each of the data ( original ), or

It is common to use variance as Detailed methods of intra / inter mode determination are well known to those skilled in the art to which the present invention pertains, and thus descriptions thereof will be omitted.

According to the present invention, since the input image is divided into macroblocks (MBs) without using contour coding, region division is performed for each macroblock (MB), so that a calculation amount is reduced and a compression ratio is increased, and a macroblock (MB) unit is used. Interworking with existing video encoders (eg, MPEG-1, MPEG-2, H.261, H.263) that perform compression coding is facilitated.

In addition, according to the present invention, it is possible to separate the input image into two regions, a background region and a target region, and the effects thereof are summarized as follows. First, if the current image macro block MB is the same as the background image macro block MB, the conversion encoding is not performed. Therefore, the uncovered background that is covered by the movement of the object and then reappeared does not need to be newly encoded. do. Second, if the input frame rate is small and the motion of the object can be processed discontinuously, the search is only required for the part of the search area that corresponds to the object area even when the search area is expanded. It is possible to lower the motion prediction failure rate without slowing down the motion prediction. Third, since the background region and the object region are distinguished from each other and encoded differently, the compression rate can be increased by intra coding only the object region without encoding the entire current image when the I-frame is inserted. Fourth, it distinguishes the background area from the object area, so it is easy to implement additional functions such as an alarm when an object appears. Fifth, it is easy to control image quality of important areas such as increasing the compression ratio in the background region and decreasing the compression ratio in the object region.

Claims (15)

A segmentation-based video compression encoder for dividing the current image into preset unit areas (MBs) and differentially compressing each unit area according to similarity between the stored background image and the stored object image. (SVC Encoder: Segmentation-based Video Compression Encoder), A current picture frame memory (CFM) for storing a current picture; A background image frame memory (BFM) for storing a background image; A previous object image memory (POM) for storing a previous object image; Region segmentation and encoding for segmenting the current image stored in the current image frame memory into a background region or an object region for each unit region and generating differential coding instructions and motion vectors for each unit region using the result. Segmentation and Coding Control (SCC); A transform coder (TC) for receiving a coding instruction and a motion vector for each unit region from the region segmentation and encoding controller and performing differential transform encoding on image data of a unit region according to the input signal; A transform decoder (TD) for performing transform decoding on the image data of the unit region transform-coded from the transform encoder to update the background image frame memory and the previous object image memory; An entropy coder (EC) for differentially performing entropy coding on the transform-coded data from the transform encoder, the encoding instruction from the region division and encoding controller, and a motion vector; A bitstream formatter (BF) for outputting the output from the entropy encoder into a bitstream according to a predetermined format; Segmentation-based video compression encoder (SVC Encoder) characterized in that it comprises a. The method of claim 1, wherein the region division and coding controller (SCC), Whether or not each unit region of the current image data stored in the current image frame memory and the background region of the same position as the unit region stored in the frame memory is judged by referring to the background image for each unit region or not. A background area classifier (BAC) for determining whether to encode in a manner; A motion estimator (ME) for performing motion prediction and outputting a motion vector for each unit region determined to be encoded in a different manner without referencing a background image by the background region discriminator; Intra / Inter mode discriminator (IID: Intra / Inter Mode Decision) for determining whether to perform intra mode encoding or inter mode encoding for the unit region; A region segmentation based video compression encoder, characterized in that it comprises a. The background image frame memory of claim 2, wherein a stationarity analysis is performed on a unit region of each unit region determined to be encoded in a different manner without referring to a background image by the background region discriminator. And a stationarity analyzer (SA) for determining whether to update or not. The mask according to claim 2 or 3, further comprising: a mask indicating whether image data stored in the previous object image memory is a background region or an object region for each unit region in order to limit a search area at the time of motion prediction to an object region; An area segmentation-based video compression encoder, further comprising an object mask memory (OMM) for storing information. The video segmentation-based video compression according to claim 3 or 4, further comprising a timer memory (TM) for storing time intervals in which normality is maintained for each unit region during normality analysis. Encoder. The method according to any one of claims 1 to 5, wherein the transform coder (TC) A motion compensator (MC) for performing motion compensation on the unit area; A transformer (T) for converting image data in a spatial domain into image data in a frequency domain and outputting transform coefficients; A quantizer (Q) for quantizing transform coefficients from the converter; An area segmentation based video compression encoder comprising a. The method of claim 6, wherein the transform decoder (TD: Transform Decoder), An inverse quantizer (Q− ¹ ) for performing inverse quantization of the change coded data in the change encoder; An inverse transformer (T- ¹ : Inverse Transformer) for performing inverse transformation on the transformation in the transformer; A motion compensator (MC) for performing motion compensation on the unit area; A segmentation-based video compression encoder comprising a. Divides the current image into pre-set Macro-Blocks (MB), and stores each unit area in the background image stored in the background frame memory (BFM) and the Previous Object memory (POM). A segmentation-based video compression encoding method for differentially compressing and encoding according to similarity with an object image. Determining whether the unit is a background area or an object area with respect to the current image; Generating a motion vector by differential coding instruction and motion prediction for each unit region; Performing differential transform encoding on image data of a unit region according to the encoding instruction for each unit region; Differentially performing entropy coding on the transform coded data, the encoding indication, and the motion vector. Region segmentation based video compression encoding method comprising a. 10. The method of claim 8, further comprising outputting the entropy-encoded data in a bitstream. The method of claim 8, further comprising performing transform decoding on the image data of the unit region that is transform-coded by the transform encoder to update the previous object image memory. Based video compression coding method. The method of claim 8, wherein the determining of whether the unit area is the background area or the object area comprises: A first brightness comparison step of comparing the brightness of the current image with the brightness of the background image with respect to pixels of each unit area; And determining the object region when the difference in brightness is greater than or equal to a first predetermined reference, and as a background region when the difference in brightness is less than or equal to a predetermined reference. The method of claim 11, wherein determining whether the unit area is a background area or an object area comprises: comparing the color difference of the current image of the unit area with the color difference of the background area when the unit area is determined as the background area. Equipped with If the difference in the color difference is more than a predetermined reference, Generating an encoding instruction for encoding the current image with reference to the background image for the unit region, If the difference in the color difference is less than the reference set trimmed, Comparing the brightness of the current image unit area with the brightness of the background area; If the difference in brightness is greater than or equal to the second criterion for brightness, Generating an encoding instruction for encoding the current image with reference to the background image for the unit region, When the difference in brightness is less than or equal to the second criterion for brightness, Generating an encoding instruction that does not encode the current video unit region. A region segmentation based video compression encoding method comprising: 12. The method of claim 8, 9, 10 or 11, wherein generating the motion vector When the current image of the unit region is determined as an object region, motion prediction is performed on each unit region and a motion vector is output, and motion estimation is performed by searching a region identified as the object region among the peripheral regions of the unit region. An area segmentation based video compression encoding method comprising a. The method of claim 8, 9, 10 or 11, further comprising: analyzing the normality of analyzing whether there is no movement for a predetermined time when the unit area is determined as the object area; The region segmentation-based video compression encoding method of claim 1, wherein the background image of the unit region is updated with the current image of the unit region corresponding to the object image without movement for a predetermined time. The method of claim 8, 9, 10, or 14, wherein the generating of the motion vector comprises: a current image of a unit area and a previous object of the unit area when it is determined that there is a motion by motion prediction. And determining a transform encoding scheme by determining similarity of the images.