KR20020015231A - System and Method for Compressing Image Based on Moving Object - Google Patents

Abstract

PURPOSE: A system and a method for object-oriented video encoding are provided to flexibly change the size of an image by orienting to a moving object and using wavelet transformation, to enhance speed gain and compressing efficiency by supporting multiple resolution and increasing processing speed, and to reduce the size and capacity of a memory by enhancing the compressing efficiency. CONSTITUTION: An object-oriented video encoding system comprises a subtracter(10), an object sampling unit(12) sampling the moving objects included in a following picture and calculating the motion vector of the moving object, a discrete wavelet transformation unit(16) wavelet-transforming a reference picture, wavelet-transforming the difference image between a prior picture and the following picture, and outputting a wavelet transformation value, a quantization unit(18) outputting a quantized value by quantizing the wavelet transformation value, an entropy coder(20) outputting an encoded value by arithmetically coding the quantized value, a multiplexer/buffer(22) temporarily storing the encoded value and motion vector and forming a bit stream including the encoded value and motion vector, an inverse quantization unit(24), an invert discrete wavelet transformation unit(26), an adder(28), and a frame memory(30). The video encoding system encodes the plural pictured including a reference picture and at least one of following pictures arranged after the reference picture. Even the minute motion of the object in a picture is identified.

Description

System and Method for Compressing Image Based on Moving Object

The present invention relates to an image processing system and method, and more particularly, to a system and method for compressing and storing or transmitting a moving image.

The image information can be compressed by removing spatial and temporal redundancy of the image. Transform encoding is mainly used as a method of reducing spatial correlation. On the other hand, many motion compensation coding is performed in the time domain, and this motion compensation coding enables to reduce the amount of data more than the coding techniques in the spatial domain. In a conventional video compression algorithm, it is common to employ a combination of the transform coding and the motion compensation coding.

An example of the existing video compression algorithm is a block-based discrete cosine transform (DCT) method. For example, in MPEG-1 / 2 and H.261, an image is processed by DCT transforming a block having a size of 8x8 in basic units, and a macroblock of 16x16 size is used as a unit for motion estimation and compensation. This method is the most widely used video compression and decompression system to date because of its simple hardware implementation and excellent motion estimation and compensation performance in moving images.

However, such a block-based DCT method exhibits a stable performance only in motion estimation and compensation in a limited area, and has a problem in that it is difficult to properly perform a function and processing speed is slow in a wide range of motion estimation and compensation. In addition, this method has a disadvantage in that it is difficult to use in satellite cameras or surveillance systems because the image quality is severely deteriorated at a low bit rate and it is difficult to identify when the object in the image is almost stationary.

Another example of an existing video compression algorithm is a method using an embedded zero-tree wavelet. This method sends the important bits first by partial alignment of the data, and generates an embedded bit string and compresses it by zero-tree coding using similarity between subbands in the wavelet transformed image. . The encoder can encode at a desired bit rate with an embedded bit string, and the decoder can also decode at a desired bit rate.

However, according to the embedded wavelet method, although encoding and decoding can be performed to have multiple resolutions, there is a problem in that one layer has most of the bit strings, and there is a problem in that there is no efficiency because there is no gain of time in the decoding process. . In addition, this method has the advantage of lower image quality deterioration and higher resolution in still image than block-based DCT method, but it is not suitable for processing video because of low compression ratio and time-consuming motion estimation and compensation. I have a problem.

The present invention has been made to solve such a problem, and provides a video encoding system capable of identifying a high compression rate, reducing image quality deterioration at a low bit rate, and even identifying small movements of an object in an image. do.

Another object of the present invention is to provide an image encoding method and a video decoding method suitable for this, in which a high compression ratio and a low bit rate reduce image quality deterioration, and even when a motion of an object in a image is small.

1 is a block diagram of a preferred embodiment of a video encoding system according to the present invention.

FIG. 2 is a detailed block diagram illustrating an example of the object extractor shown in FIG. 1. FIG.

3 is a flowchart showing a camera motion detection and compensation process.

4 is a diagram illustrating a search area for camera motion detection.

5 is a diagram illustrating a frame system for explaining a process of forming a motion object frame;

6A to 6I illustrate examples of forming a moving object frame.

7 is a flowchart illustrating an example of a bounding box configuration process.

FIG. 8 shows an example of a bounding box configured for the mask of FIG. 6H.

9A-9C show examples of bounding boxes in the case where two or more shapes are present.

10 is a flow chart showing a minimum bounding box extraction process for each of the individual moving objects in the bounding box.

FIG. 11 is a diagram illustrating an example of grouping shapes in a bounding box according to a motion vector in the images illustrated in FIGS. 6A to 6I and 8.

12A and 12B show examples of minimum bounding boxes for individual moving objects.

FIG. 13 shows a background mask for the image illustrated in FIGS. 6A-6I.

14A and 14B show an example of a moving object appearing from an edge like this.

FIG. 15 is a block diagram of a discrete wavelet transform unit shown in FIG. 2; FIG.

FIG. 16 is a detailed block diagram of the filtering unit illustrated in FIG. 15. FIG.

17 and 18 illustrate examples of decomposing an input image into sub-images.

19A to 19G are diagrams showing the configuration of a bit stream output by the video encoding system of FIG.

20 is a block diagram of a preferred embodiment of an image decoding system according to the present invention.

21 is a flowchart showing a video encoding method according to the present invention.

22 is a flowchart illustrating an image decoding method according to the present invention.

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

10: subtractor 12: object extraction unit

16: Discrete Wavelet Transform 18: Quantization

20: Entropy Coder 22: Multiplexer / Buffer

24: inverse quantization unit 26: inverse transform unit

28: adder 30: frame memory

290: bounding box 340, 342: minimum bounding box

F _i : Picture frame ER _{i, j} : Car picture frame

MASKER _{i, j} : Mask Frame MASKER _i : Mask

MOBFR _i : Motion Object Frame

In order to achieve the above technical problem, the present invention extracts and encodes only a motion object and a main still object for a screen other than a reference screen, thereby flexibly changing an image size to remove redundancy over time. In encoding each object, wavelet transform is used to support multiple resolutions while increasing processing speed to increase speed gain and compression efficiency. In particular, for the moving object, the difference image between the original image and the previous image is obtained, and the difference between the original image and the next image is obtained again. The logical product of two difference images is calculated to obtain a motion region, and a motion vector is obtained for the motion region. Next, the difference image and the motion vector of the object are compressed by wavelet transform.

In detail, an image encoding system for achieving the above technical problem compresses and encodes a plurality of screens including a reference screen and at least one subsequent screen arranged after the reference screen. The object extracting means extracts a motion object included in the subsequent screen with respect to the subsequent screen, and calculates a motion vector of the motion object. The wavelet converting means wavelet transforms the reference screen and wavelet transforms the difference image with the previous screen of the subsequent screen with respect to the motion object in the subsequent screen. The quantization means quantizes the wavelet transform value and outputs a quantized value. The entropy encoding means encodes the quantized value by arithmetic coding. The multiplexing and buffering means form a bit stream of a predetermined format comprising the encoded value and the motion vector of the motion object.

Meanwhile, in the image encoding method for achieving the above another technical problem, first, (a) wavelet transforms, quantizes, and arithmically encodes a reference picture to generate coded reference picture data. (b) A motion object is extracted for the subsequent screen, and a motion vector of the motion object is calculated. (c) Wavelet transforms, quantizes, and arithmetic-codes the difference image with respect to the previous screen before the motion object to generate coded motion object data. (d) A bit stream having a predetermined format including encoded reference picture data, encoded motion object data, and a motion vector is formed.

In the image decoding method, first (a) the bit stream is decomposed to restore the encoded reference picture data, the encoded motion object data, and the motion vector. (b) Variable length decoding of the coded reference picture data, inverse quantization, and inverse wavelet transform to restore the reference picture. (c) For the subsequent screen, the motion object data is variably decoded, dequantized, and inverse wavelet transformed to restore the motion object. (d) After moving the moving object part in the previous screen by the motion vector, the moving screen object is overwritten on the reference screen to restore the subsequent screen.

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

1 shows an embodiment of an image encoding system according to the present invention. The illustrated system includes a subtractor 10, an object extractor 12, a discrete wavelet transformer (DWT) 16, a quantization unit 18, an entropy coder 20, a multiplexer / buffer 22, an inverse quantizer ( 24), an inverse transform unit (IDWT) 26, an adder 28, and a frame memory 30.

The video encoding system sets a predetermined number of frames, for example, about 5 to 30 consecutive frames, into a picture group (GOP), and encodes pictures in GOP units. For each GOP, the video encoding system encodes the entire picture by wavelet transforming the first frame (I-picture), and encodes the remaining frames in the GOP with reference to the previous frame. That is, for a frame other than an I-picture, the encoding system obtains a difference image between a frame to be encoded and a previous frame, extracts motion objects, and then obtains a motion vector of each of the motion objects. The difference between the previous frame and the minimum bounding box including each motion object is obtained, and the difference image is wavelet-transformed and stored or transmitted together with the motion vector.

First, a case in which an I-picture is input to the video encoding system of FIG. 1 will be described. The input video signal is directly input to the wavelet converter 16. The wavelet transform unit 16 performs wavelet transform on the image signal and outputs transform coefficients. The quantization unit 18 quantizes the transform coefficients from the wavelet transform unit 16. In this case, the quantization rate may vary according to the first control signal CON1 applied from the outside. The entropy coder 20 compresses the signal by arithmetically encoding the quantized data. On the other hand, the inverse quantizer 24 inverse quantizes the quantized data again, and the inverse transformer IDWT 26 performs inverse wavelet transform on the output signal of the inverse quantizer 24. Therefore, the signal output from the inverse transform unit IDWT 26 is almost the same as the signal before the wavelet transform.

The adder 28 adds the output signal of the frame memory 30 to the output signal of the inverse transform unit IDWT 26 and supplies the addition result to the frame memory 30. Here, since the frame memory 30 is reset before encoding of the I-picture, the signal supplied by the adder 28 to the frame memory 30 becomes a video signal for the entire frame of the I-picture. The frame memory 30 stores a signal supplied from the adder 28.

Next, a case in which a frame other than an I-picture, that is, a B-picture, is input to the video encoding system will be described. The subtractor 10 obtains a difference image between the input image and the previous frame image from the frame memory 30 and outputs a frame difference image signal. The object extractor 12 extracts the objects included in the frame from the frame difference image signal, obtains the object difference image signal for each object, outputs the object difference image signal to the wavelet converter 16, and outputs a motion vector for each object. The calculation is performed to the multiplexer / buffer 22.

In this case, the wavelet converter 16 wavelet transforms the object difference image signal for each object included in the frame. The operations of the quantization unit 18, entropy coder 20, inverse quantization unit 24, and inverse transform unit (IDWT) 26 are the same as in the case where an I-picture is input. The adder 28 adds the difference image signal of each object output from the inverse transform unit IDWT 26 to the output signal of the frame memory 30 and supplies the addition result to the frame memory 30. Accordingly, the frame memory 30 stores an image signal for the current frame F _N , which may be used as a reference signal for encoding the next frame image signal. In addition, the frame memory 30 may include the previous frame F _N-1 and the next frame F _{N + 1} of the frame to be encoded, and the inter-frame difference images ER _{N, N-1} and ER _{N + 1, N.} It is desirable to have sufficient capacity to store the) together.

The multiplexer / buffer 22 temporarily stores the encoded data for the entire entropy-coded I-picture frame, the encoded data of each object for the B-picture, and the motion vector, and arranges them to form a bit stream of a predetermined format. The structure of the bit stream will be described later. Data output from the multiplexer / buffer 22 may be transmitted remotely through a communication network or stored in a storage medium.

The bit stream configuration operation of the multiplexer / buffer 22 is adjusted in response to the second control signal CON2. In the present embodiment, the first control signal CON1 for the quantization unit 18 and the second control signal CON2 for the multiplexer / buffer 22 are supplied by a controller (not shown). The controller may be implemented as a high performance RISC processor. Meanwhile, in another embodiment of the present invention, all or part of the image encoding system of FIG. 1 may be implemented by software running on such a processor.

FIG. 2 shows an example of the object extractor 12 shown in FIG. 1. The object extractor 12 includes a camera motion detector and compensator 100, a motion object frame constructer 102, a bounding box constructer 104, and a motion object extractor 106.

The camera motion detection and compensator 100 compares the current frame with the previous frame to determine whether the camera has moved regardless of the movement of the object within the frame, and compensates the movement of the camera when it is determined that the camera has moved. That is, the camera motion detection and compensation unit 100 obtains a full frame motion vector for each frame, and when the encoding system compresses the difference image between the previous frame (Previous frame or Reference frame) and the current frame, It allows the frame motion vector to be reflected in the error frame calculation.

The camera motion detection and compensation process is shown in FIG. 3. Referring to FIG. 3, first, as shown in FIG. 4, a frame motion search block (180, 182, 184, 186) is set to a 32 × 32 pixel block near four corners of the current frame. In addition, a 160 x 160 pixel block is set as the search areas 190, 192, 194, and 196 for each search block (operation 150). A frame motion vector is obtained for each search block by block matching (step 152). In other words, while shifting each search block in the search area, the pixel value difference is calculated for each pixel in the search area, and the square values of the pixel value difference are accumulated. The amount of motion is determined by the motion vector for the corresponding search block.

The entire frame motion vector is obtained based on the obtained four frame motion vectors (steps 154 to 162). At this time, if the four frame motion vectors are similar in direction and magnitude, and the magnitude of the minimum error is small, the average of the four motion vectors is taken as the entire frame motion vector (steps 154 and 156). If three of the four frame motion vectors are similar and one is different, the average of the three similar motion vectors is taken as the full frame motion vector (steps 158 and 160). On the other hand, if two or more of the frame motion vectors are different from each other, it is assumed that a screen transition has occurred and the current frame is set to a new reference frame, that is, an I-screen, so that the entire frame is encoded (162). step).

Referring back to FIG. 2, the moving object frame constructing unit 102 forms a moving object frame indicating a portion in which only a moving object exists in the image frame. FIG. 5 illustrates a frame system for explaining a process of forming a motion object frame by the motion object frame constructing unit 102. For example, when the _Nth frame (F _N : 202) is encoded, the difference image (ER _{N, N-} ) between the _N- 1th frame (F _N-1 : 200) and the Nth frame (F _N : 202) is first described. ₁ : 206). As described above, in obtaining a difference image, a frame motion vector is considered. That is, in obtaining the difference images ER _{N and N-1} , the pixel value with the N th frame F _{N in} pixel units after the N-1 th frame F _N-1 is moved by the frame motion vector. You get a car. Similarly, the difference image (ER _{N + 1, N} : 208) between the _Nth frame (F _N : 202) and the N + 1th frame (F _{N + 1} : 204) is obtained.

Next, in the difference image ER _{N, N-1} , the mask frame MASKER _{N, N-1} : 210 is obtained by replacing the pixel values of all pixels whose pixel value is not "0" with "1" and binarizing them. . Similarly, a mask frame (MASKER _{N + 1, N} : 212) is obtained by binarizing the difference image ER _{N + 1, N.} The mask MASAKER _N 214 is obtained by performing an AND operation on two mask frames MASKER _{N, N-1} and MASKER _{N + 1, N.} The N th frame (F _N ) and the mask (MASKER _N : 214) are ORed again to obtain a motion object frame (MOBFR _N ) having only a moving object.

An example of obtaining the motion object frame MOBFR ₂ for the second frame F ₂ is shown in FIGS. 6A-6I. 6A to 6C show images (F ₁ , F ₂ , F ₃ ) of _three consecutive frames, in which two moving objects exist. Figure 6d and 6e are respective first frame (F ₁₎ and two difference between the first frame (F ₂₎ between the difference image (ER _1,2) and the second frame (F ₂₎ and the third frame (F ₃₎ video (ER _2,3 ). Figure 6f and 6g is a difference image of the mask frame (MASKER _1,2) and a substituted primary image the pixel values of all pixels in the pixel value (ER _1,2) other than "0" to "1" (ER _{2, 3} shows a mask frame (MASKER _2,3 ) in which the pixel value of all pixels whose pixel value is not "0" is replaced with "1". Figure 6h shows the mask (MASKER ₂₎ is formed by calculating a logical product of two mask frame (MASKER _1,2, MASKER _2,3) pixel by pixel, and Fig. 6i is a second frame (F ₂₎ and the mask ( MASKER ₂ ) shows the motion object frame (MOBFR ₂ ), which is obtained by logical ANDing again. As shown, the moving object frame MOBFR ₂ essentially includes portions in which the moving object exists in the second frame F ₂ , and almost all other portions of the screen are filled with pixel values of zero.

In FIG. 2, the bounding box constructing unit 104 configures a bounding box including a moving object in the image. Here, the bounding box refers to a rectangular box including a motion object. Such a bounding box may be configured in various ways. In a preferred embodiment, the bounding box constructer 104 may have a pixel value of "1" in the mask frame MASKER _{i, i-1} or the mask MAKER _i . The bounding box of the minimum size is derived by scanning the shape connected with the pixels by raster scanning. The bounding box configuration process is shown in FIG.

First, scanning starts from the left top of the frame (step 250). When a pixel having a pixel value of "1" appears for the first time in the scanning process (step 252), the X and Y coordinates of the corresponding pixel are set to the X and Y coordinates of the upper left and lower right corners of the bounding box (step 254). . The shape scanning is continued until the pixel having the pixel value of "1" appears again (step 256). If a pixel having a pixel value of "1" appears again (step 258), the coordinates of the pixel are compared with the preset X and Y coordinates of the upper left and lower right corners to update the X and Y coordinates of the upper left and lower right corners. (Steps 260 to 274).

That is, if the X coordinate of the pixel whose newly found pixel value is "1" is smaller than the upper left corner X coordinate (X_top_left), the upper left corner X coordinate (X_top_left) is updated to the X coordinate of the corresponding pixel (260 and 262). step). If the X coordinate of the pixel is larger than the right lower corner X coordinate (X_bottom_right), the lower right corner X coordinate (X_bottom_right) is updated to the X coordinate of the pixel (steps 264 and 266). On the other hand, if the Y coordinate of the pixel is larger than the lower right corner Y coordinate (Y_bottom_right), the lower right corner Y coordinate (Y_bottom_right) is updated to the Y coordinate of the pixel (steps 268 and 270). The scanning and coordinate updating procedure of steps 256 to 270 is continued until the scanning is completed (step 272).

On the other hand, the bounding box is configured by scanning pixels having a pixel value of "1" in the mask frame MASKER _{i, i-1} or mask MASK _i , and having a pixel value of "0" in the motion object frame MOBFR _i . It may be configured by scanning pixels other than ". 8 shows an example of a bounding box 290 configured for the motion object frame MOBFR ₂ of FIG. 6I. On the other hand, in another embodiment of the present invention, in configuring the bounding box, the frame may be scanned in units of 8x8 blocks or 16x16 macroblocks. In this case, if a block in which at least one "1" exists in each block appears, the coordinates of the upper left pixel and the lower right pixel of the block may be used to update the corner of the bounding box. In this case, the recording of the position and size of the bounding box can also be made using the number of the block or macroblock.

In addition, when two or more shapes exist in one frame, a separate bounding box may be derived for each shape. 9A to 9C show examples of bounding boxes formed when two or more shapes are present. FIG. 9A shows a motion object frame (MOBFR), FIG. 9B shows a bounding box for a car among the motion objects present in the motion object frame of FIG. 9A, and FIG. 9C shows a bounding box for an airplane among the motion objects. .

Referring back to FIG. 2, in the preferred embodiment the motion object extractor 106 extracts the minimum bounding box for each of the individual motion objects in each bounding box. 10 shows a minimum bounding box extraction process. First, the motion object extractor 106 obtains a motion vector between a previous frame and a current frame with respect to a macroblock included in or partially overlapped with the mask MASKERi for each bounding box (step 300).

The detected motion vectors are classified into several groups, such as AA and BB, according to the similarity in magnitude and direction (step 302). If macroblocks belonging to each group are adjacent to each other, it is regarded as a single object and a minimum bounding box covering the object is extracted (steps 304 and 306). This minimum bounding box can be calculated with only the macroblock number belonging to the group. All areas within the minimum bounding box to which the object does not belong are filled with "0" or any average value. The motion object extractor 106 outputs the difference image for the minimum bounded box to the discrete wavelet transform unit 16 of FIG. 1, so that the discrete wavelet transform unit 16 outputs the difference image in units of the minimum bounded box. Lets you convert wavelets.

FIG. 11 illustrates an example of grouping shapes in a bounding box according to a motion vector in the images illustrated in FIGS. 6A to 6I and 8. In the example shown, the shape in the bounding box is divided into two single objects, AA and BB. In the preferred embodiment, the motion object extractor 106 extracts the minimum bounding boxes 340 and 342 as shown in Figs. 12A and 12B for each of these single objects, so that the discrete wavelet transform unit 16 is the least bounded. Wavelet transform the difference image for a single object in boxes 340 and 342. However, in another embodiment of the present invention, it is also possible to perform discrete wavelet transformation on all forms in the bounding box as one unit without extracting the minimum bounding box.

On the other hand, the motion object extractor 106 extracts a part of the background that is hidden from the motion object in the previous frame together with the motion object, but the background is recovered from the current frame. Such a background object extracts the background mask F _N EXT _N-1 by subtracting the mask MASKER _N pixel by pixel from the mask frame MASKER _{N, N-1} described above, and extracts the background mask F _N EXT. _N-1 ) can be extracted by logical ANDing the frame image F _N again. FIG. 13 shows a background mask for the image illustrated in FIGS. 6A to 6I. The motion object extractor 106 outputs the difference image of the background object to the discrete wavelet transform unit 16 in bounded units thereof, so that the discrete wavelet transform unit 16 converts the wavelet.

On the other hand, a moving object in an image often disappears from one edge over time and then disappears through another edge after elapse of a certain time, rather than being in full form in every frame. 14A and 14B show an example of a moving object appearing from an edge like this. Although the new motion object part DD newly shown in FIG. 14B is part of the motion object that is not present in the previous frame, it is inappropriate to encode the difference image based on the previous frame F ₁ . Accordingly, the moving object extracting unit 106 forms separate minimum bounding boxes for the portion CC existing in the previous frame and the portion DD not present in the previous frame. The motion object extractor 106 supplies the difference image signal to the discrete wavelet transform unit 16 for the portion CC existing in the previous frame, but provides a complete image for the portion DD not present in the previous frame. It is supplied to the let transform unit 16 so that the absolute pixel value can be encoded.

FIG. 15 illustrates the discrete wavelet transform unit 16 shown in FIG. 2 in detail. The discrete wavelet converter 16 includes first and second switches 400 and 402, a filter 404, a signal expander 406, and a data aligner 408. The first switch 400 receives the frame image signal as one input terminal and the output signal of the signal extension unit 406 as the other input terminal, and filters one of them in response to the first selection signal SEL1. ). The filtering unit 404 converts the received signal into wavelet transform and outputs it to the second switch 402 and the data alignment unit 408. The second switch 402 receives the output signal of the filtering unit 404 as one input terminal and the output signal of the data alignment unit 408 as the other input terminal, and responds to one of these in response to the second selection signal SEL2. Is output as the transform coefficient. The first and second selection signals SEL1 and SEL2 are supplied by a controller (not shown).

When a frame image signal representing complete pixel values is supplied, the discrete wavelet transform unit 16 performs only wavelet transform and outputs the result as a transform coefficient. However, when the object difference image signal is supplied, the discrete wavelet transform unit 16 passes the wavelet transformed through the signal extension unit 406 before wavelet transforming the signal to prevent image quality deterioration at the interface. The result is passed back through the data sorter 408. The signal extension unit 406 may be implemented by, for example, a Dowchies filter, and outputs a point-symmetric extension of a difference image signal at an interface. The data aligning unit 408 rearranges and outputs the output data of the filtering unit 404 in order to preserve the complete restoration characteristics of the filter bank in the filtering unit 404.

FIG. 16 shows the filtering unit 404 shown in FIG. 15 in more detail. The filtering unit 404 includes a multiplexer 450, a row operation unit 452, a first memory 458, a column operation unit 460, and a second memory 470. The multiplexer 450 receives an input data stream from the outside and data from the second memory 470, and selects one of them and outputs the data. The row operation unit 452 includes a low pass filter 454 and a high pass filter 456, and performs low pass filtering and high pass filtering in parallel along the row direction of the output data of the multiplexer 450. Downsamples and stores the data in the first memory 458. The thermal operation unit 460 includes low pass filters 462 and 466 and high pass filters 464 and 468, and the low pass filtering and the data stored in the first memory 458 in parallel along the column direction. Perform highpass filtering and downsample to 2. The thermal operation unit 460 stores the LL band sub-image among the four sub-images LL, LH, HL, and HH in the second memory 470, and outputs the remaining sub-images to the outside.

The apparatus of FIG. 16 operates as follows. When the image signal is newly input to the filtering unit 404, the multiplexer 450 supplies the input image signal to the row operation unit 452. As shown in FIG. 17, the input image is wavelet-transformed one level at a time by the row operation unit 452 and the thermal operation unit 460 so that four sub-images LL ₀ , LH ₀ , HL ₀ , HH Decomposes into ₀ ). Here, the sub-image LL ₀ includes low frequency horizontal direction information and low frequency vertical direction information, and the sub-image LH ₀ includes low frequency horizontal direction information and high frequency vertical direction information. In addition, the sub-image HL ₀ includes high frequency horizontal direction information and low frequency vertical direction information, and the sub-image HH ₀ includes high frequency horizontal direction information and high frequency vertical direction information. Since the double sub-image LL ₀ contains the most energy, the sub-image LL ₀ is fed back to the row operation unit 452 and the thermal operation unit 460 so that four sub-images LL ₀ are shown as shown in FIG. 18. Decompose into sub-images (LL ₁ , LH ₁ , HL ₁ , HH ₁ ). Depending on the application, the sub-image LL may be further resolved.

19A to 19G show the configuration of a bit stream output by the multiplexer / buffer 22 in the video encoding system of FIG. As shown in FIG. 19A, the video sequence bit stream is arranged with successive screen groups (GOP) 502 in the header 500. As shown in FIG. 19B, each GOP 502 includes encoded picture data 512 for a certain number of frames in the header 510. The header 510 includes information such as the number of frames in the GOP 502. As shown in FIG. 19C, each screen data 512 includes a header 520, motion object information, and still object information. In particular, the movement object information may include a header 522 indicating that the movement object data starts and a plurality of movement object data 524. Similarly, the stationary object information may include a header 526 indicating that the stationary object data starts, and a plurality of stationary object data 528. Here, the stationary object includes a background object and a moving object part newly appearing on the screen.

As shown in FIG. 19D, the motion object header 522 includes a start code 530, the number of motion objects 532 in the corresponding screen, the bounding box coordinates 534, and the individual motion object list 536. . As shown in FIG. 19E, the motion object data 524 includes a header 540, a minimum bounding box coordinate 542 for that motion object, a wavelet coefficient 544 and a motion vector 546 for that object. ). As shown in FIG. 19F, the stationary object header 526 includes a start code 550, the number of stationary objects 552 in the screen, the bounding box coordinates 554, and the individual stationary object list 556. . As shown in FIG. 19G, the stationary object data 528 includes a header 560, a minimum bounding box coordinate 562 for that stationary object, and a wavelet coefficient 544 for that object.

20 shows an embodiment of an image decoding system according to the present invention. The illustrated system includes a variable length decoder 600, an inverse quantizer 602, an inverse wavelet transform unit (IDWT) 604, an adder 606, and a memory 608. The variable length decoder 600 performs variable length decoding on an arithmetic-encoded data stream, the inverse quantizer 602 inversely quantizes the decoded data, and the inverse transformer 604 converts the output signal of the inverse quantizer 602. Inverse wavelet transform. In the case of the first frame image, that is, the reference image (I-picture), the inverse wavelet transformed and decoded data is output to the outside and stored in the memory 608. In the case of subsequent screens other than the I-screen, that is, the P-screen, the inverse wavelet transformed data is added to the previous screen data stored in the memory 608 to be synthesized into the entire data of the screen and output to the outside. 608 is stored.

The operations of the video encoding system and the decoding system according to the present invention are summarized as follows.

Referring to FIG. 21, in video encoding, first, the entire reference picture (I-picture) is wavelet-transformed and encoded (step 700). Every 5 to 30 frames are set to a GOP and the first screen in the GOP is set as the reference picture. However, when the background change is severe or the number of moving objects is large, the corresponding screen may be set as the reference screen, in which case the GOP itself may be newly set. Meanwhile, in another embodiment of the present invention, when there is almost no background change, only the background may be separately extracted and transmitted or stored, and only the motion object may be encoded for the reference picture.

Screens other than the reference picture, that is, P-picture, are always encoded in the object unit by referring to the previous picture. The object extractor 12 shown in FIG. 2 extracts a motion object, calculates a motion vector for each motion object, supplies the difference image of the motion object to the discrete wavelet transform unit 16, and supplies the motion vector to the multiplexer / The buffer 22 is supplied to the buffer 22 (step 702). The discrete wavelet transform unit 16 encodes the difference image in units of one level. Since P-picture coding is performed on a per object basis, the size of the input image for the wavelet transform is variably determined. If moving objects occur in multiple places within an image, the order of each minimum bounding box for input is determined from top to bottom and left to right.

The wavelet transformed data is sequentially quantized by the quantization unit 18 for each minimum bounding box, and the quantized value is arithmetic encoded by the entropy coder 20. The multiplexer / buffer 22 forms the bit stream in accordance with the format shown in Figs. 19D and 19E along with the motion vector and the bounding box coordinates (step 704).

Meanwhile, the object extractor 12 extracts still objects such as a new appearance part and a background object among the moving objects and obtains a minimum bounding box for each still object object (operation 706). The discrete wavelet transform unit 16 encodes a still object image, the quantization unit 18 sequentially quantizes the wavelet transformed data for each minimum bounding box, and the entropy coder 20 performs arithmetic encoding on the quantized value. do. The multiplexer / buffer 22 forms the bit stream in accordance with the format shown in Figs. 19F and 19G with the arithmetic coded value along with the minimum bounding box coordinates (step 708).

Steps 702 to 708 are performed for all screens other than the reference screen in the GOB. When encoding of each screen is finished, it is determined whether the corresponding screen is the end of the GOB (step 710). If the corresponding screen is not the end of the GOB, steps 702 to 708 are performed for the next screen again. .

Referring to FIG. 22, in image decoding, first, a reference picture is decoded and stored in the memory 608 (operation 750). After the motion object difference image data of the next screen is decoded (operation 752), a portion of the motion object corresponding to the decoded data on the previous screen is separated and moved by a motion vector, and then overwritten on the previous screen. Accordingly, updating of the information on the moving object is completed (operation 754). Next, after decoding the still object image data (step 756), the decoded data is overlaid on the corresponding position of the previous screen. If the screen is not the end of the GOB (step 758), steps 752 to 758 are performed again for the next screen.

Those skilled in the art to which the present invention pertains will understand that the present invention can be implemented in other specific forms without changing the technical spirit or essential features. Therefore, the above-described embodiments are to be understood as illustrative in all respects and not as restrictive. The scope of the present invention is shown by the following claims rather than the detailed description, and all changes or modifications derived from the meaning and scope of the claims and their equivalents should be construed as being included in the scope of the present invention. do.

As described above, since the present invention is based on a motion object and uses wavelet transform in compressing a video, it is possible to flexibly change the image size, support multiple resolutions, and at the same time increase the processing speed, Compression efficiency can be increased. The high compression ratio allows to reduce the memory size or the usage amount, and to increase the processing speed. In particular, the system of the present invention has the advantage that the hardware configuration is simple and can be used for both still and moving images. Accordingly, the present system can be applied to various fields including the surveillance system, and can allow the user to use the imaging system, which has been used individually for a limited portion, with little financial burden.

Claims (9)

An image encoding system which is imaged by an image pickup means and compressively encodes a plurality of screens including a reference screen and at least one subsequent screen disposed after the reference screen. Object extraction means for extracting a motion object included in the screen with respect to the subsequent screen, and calculating a motion vector of the motion object; Wavelet converting means for converting the reference screen into a wavelet, wavelet transforming a difference image with the previous screen of the subsequent screen with respect to a motion object in the subsequent screen, and outputting a wavelet transform value; Quantization means for quantizing the wavelet transform value and outputting a quantized value; Entropy encoding means for performing arithmetic encoding on the quantized value to output an encoded value; And Multiplexing and buffering means for receiving and temporarily storing the encoded value and the motion vector of the motion object, and forming a bit stream of a predetermined format including the encoded value and the motion vector of the motion object; Image coding system comprising a. The method of claim 1, The object extracting means additionally extracts a background object which is covered by the moving object in the previous screen but not covered by the moving object in the subsequent screen in encoding the subsequent screen, And the wavelet transform means additionally wavelet transforms the background object. The method according to claim 1 or 2, In encoding the subsequent screen, camera motion detection for detecting the amount of movement of the image pickup means moved between the subsequent screen and the previous screen, and the wavelet converting means to wavelet transform the difference image reflecting the motion amount Way; Image coding system further comprising. An image encoding method for compressing and encoding a plurality of screens which are picked up by an imaging unit and include a reference screen and at least one subsequent screen disposed after the reference screen, (a) wavelet transforming, quantizing, and arithmetically encoding the reference picture to generate coded reference picture data; (b) extracting a motion object for the subsequent screen and calculating a motion vector of the motion object; generating encoded motion object data by performing wavelet transform, quantization, and arithmetic encoding on a difference image with respect to the subsequent screen before the subsequent screen with respect to the motion object; And (d) forming a bit stream of a predetermined format including the coded reference picture data, the coded motion object data, and the motion vector; Image encoding method comprising a. The method of claim 4, wherein Step (b) is And further extracting a background object that is hidden by the moving object in the previous screen but not covered by the moving object in the subsequent screen. Step (c) is And further performing wavelet transform, quantization, and arithmetic encoding of the background object. The method according to claim 4 or 5, Step (b) is (b1) detecting an amount of movement of the imaging means moved between the subsequent screen and the previous screen; In the step (c), the video encoding method for wavelet transform the difference image reflected in the motion amount. The method of claim 4, wherein Step (b) is (b1) obtaining a bounding box surrounding the moving object and having a pixel value of zero except for the moving object; More, In the step (c), the video encoding method for performing wavelet transform, quantization and arithmetic encoding of the bounding box unit. An image decoding method for decoding an image encoded by the method of claim 4, (a) decomposing the bit stream to restore the coded reference picture data, the coded motion object data, and the motion vector; (b) variably decoding the encoded reference picture data, inverse quantization, and inverse wavelet transform to restore the reference picture; (c) restoring the motion object by variable length decoding, inverse quantization, and inverse wavelet transform of the encoded motion object data with respect to the subsequent screen; And (d) restoring the subsequent screen by moving the motion object part in the previous screen by the motion vector and then overwriting the motion object part on the reference screen; Image decoding method comprising a. The method of claim 8, wherein the bit stream further includes encoded background object data of a background object that is hidden by the moving object in the previous screen but not covered by the moving object in the subsequent screen. The step (a) further includes the step of recovering the encoded background object data from the bit stream, The method is (e) variable length decoding, inverse quantizing and inverse wavelet transforming the encoded background object data for the subsequent screen to restore the background object; And (f) overwriting the background object on the subsequent screen; Image decoding method further comprising.