KR19990080690A - Fast Compression Algorithm for Video Transmission - Google Patents

Abstract

The present invention relates to an image compression algorithm having a high speed algorithm that can be implemented in a low speed CPU by optimizing a compression algorithm of a transmission image.

Conventionally, there are many compression algorithms for transmitting multimedia in a high speed transmission path, but it is configured to repeat a very complex operation, it takes a long time to compress the image and the system is complicated, there are many problems. The present invention solves this problem, thereby minimizing the amount of computation and maximizing the amount of compression, thereby enabling image compression that cannot be implemented with conventional algorithms.

As shown in FIG. 2, the fast compression algorithm of the transmission image of the present invention includes a video input unit 1 receiving a signal from a video, a HADAMARD operator 14 and a quantizer 17 which compress image data. A format generator 18 for creating a format of compressed information, a variable length encoder 21 for representing the formed format with a small number of bits, an error corrector 25 for correcting errors in transmission, and compressed information. Compressed information transmitter 27 for transmitting a.

Description

Fast Compression Algorithm for Video Transmission

The present invention relates to an image compression algorithm having a high speed algorithm that can be implemented in a low speed CPU by optimizing a compression algorithm of a transmission image.

Due to the rapid development of transmission paths and computers, the demand for multimedia is increasing rapidly. Accordingly, many researches on compression algorithms for multimedia transmission on high speed transmission paths are underway. ITU-T H.261, the international standard for narrowband video telephony or video conferencing, ISO MPEG-1, a video compression algorithm of VTR level 3, ISO MPEG-2, a video compression algorithm of high definition, and still image compression algorithm of high definition. ITU-T JPEG.

In the conventional transmission image compression algorithm, as shown in FIG. 1, an image is input from a video input unit 1, and an arithmetic processing for compression is performed by an information source encoder 2, and the data compressed by the multiplex encoder 6 is shown. It consists of an algorithm that creates a format and inserts code and transmits compressed data to cope with errors during transmission.

The configuration of a conventional video compression algorithm is described in detail with reference to FIG. 1 as follows.

The transmission image compression algorithm of FIG. 1 receives an image to be compressed by the video input unit 1 and performs an arithmetic process for compressing the amount of information in the information source encoder 2. The computation process includes the DCT (DISCRETE COSINE TRANSFORM) (3), quantization (4), motion compensation (5), and the like. The multiplexed encoder 6 also generates and processes the data format after compression. The hierarchical structure 7 and variable length coding 8 are used for this data format. In addition, a transmission buffer (BUFFER) 9 is provided to constantly transmit the data, and the amount of data compressed through the encoding controller 10 is adjusted. The transmission encoder 11 corrects an error that may occur during transmission. For this purpose, it consists of a process of inserting and transmitting an error correction code.

In addition, the transmission image compression process of the prior art will be described in detail as follows.

Conventional image compression algorithms use DCT (3), quantization (4), motion compensation (5), etc. to receive the image data from the video input unit 1 and to compress the data from the information source encoder (2). DCT (DISCRETE COSINE TRANSFORM) (3) is a COSINE orthogonal transformation, which allows the input image to be resolved into spatial frequency components by dividing the pixels irregularly spread into horizontal and vertical pixels through DCT (3). Through the DCT (3) decomposing into components, irregular pixels tend to concentrate toward the low frequency term, so that information can be compressed with little information loss by discarding the high frequency term. Quantization (QUANTIZATION) (4) is to divide each orthogonal transformed frequency component value by a certain value called "quantization step size" in order to discard the high frequency component of the information transformed through the above DCT (3) It means to get rid of high frequency terms by removing them. With motion compensation (5), moving images are displayed as successive images by displaying different still images one after another. In this case, neighboring images are very similar, so when a portion is moved from the previous image, the movement amount of the previous portion is moved. It is a process of bringing a certain area of screen from the previous image with the amount of movement by transmitting the vector. The multiplexed encoder 6 compresses the information compressed by the information source encoder 2 by using the variable length encoding 8 to generate and process the format of the data. In variable length coding (8), the probability of occurrence of data generated when compressed by the foregoing method is noticeably biased, so that information generated by the information source encoder (2) is assigned a long code to a value having a low probability of occurrence. In other words, a short code is assigned to a high probability of occurrence to reduce the average code.

In addition, in order to make the data rate for transmission constant, a method of reducing the amount of generation of data through the encoding controller 10 is used when the amount of data compressed by the transmission buffer 9 as a memory is filled in the transmission buffer 9. When data enters the transmission buffer 9 as before, the transmission encoder 11 inserts an error correction code 12 that can correct an error in preparation for an error that may occur during transmission, and transmits the data. The error correcting code 12 carries on the data a complex error correcting code 12 defined in various conventional international standards mentioned above.

As described above, the conventional image compression algorithm compresses an image using various various compression algorithms. The DCT (3) transform used in the information source coder 2 of FIG. 1 is an orthogonal transform using cosine and converts an input image into a frequency band with a floating point value that is a value of a cosine function. However, since this operation has floating point and includes addition and multiplication operations, it takes much time to perform the actual operation. In addition, the quantization step size (QUANTIZATION STEP SIZE) is not constant in the quantization (4), so multiple divisions must be performed to perform quantization. In the multiplex encoder 6, HUFMAN TABLE is used to allocate the information compressed by the information source encoder 2 according to the probability of occurrence. The HUFMAN table has various lengths of codes, which is suitable for compressed data. The Huffman table values that match the compressed data can be assigned only by searching the Huffman tables one by one for many hours. Since the transmission encoder 11 always has a probability that an error occurs in data during transmission, an error correcting code 12, "BCH (511, 493): BoseChaudhuri-Hocquenghem Code", is used to prevent such an error. The error correcting code 12 can be generated only by performing a large number of operations to generate the correcting code 12.

Therefore, as mentioned above, in order to compress an image for transmission by a conventional method, very many operations must be performed to obtain compressed image data.

If the above conventional algorithms are implemented using an 8BIT CPU or a 16BIT CPU or the like, it takes a very long operation to compress an image due to the complexity of the algorithm's calculation process. Pentium-class personal computers (PCs), for example, take a long time to compress. When implemented on an 8-bit CPU or a 16-bit CPU, it takes too long computation time, so if you use a DSP (DIGITAL SIGNAL PROCESSOR) or ASIC (APPLICATION SPECIFIC IC) It can have a lot of effect on the improvement, but the price of DSP and ASIC is quite high compared to CPU for actual application and even if DSP and ASIC are used, CPU must be used for transmission and control. As the price increases, the price competitiveness is deteriorated. Even if the equipment is made using a DSP or ASIC dedicated to transmission video compression, the DSP or ASIC dedicated to image compression cannot be applied to industrial applications. I have a problem.

In order to solve the above problems, the present invention has been developed to minimize the amount of computation and to implement a high speed compression algorithm of a transmission system of a new system that can be used in various applications while enabling implementation in an existing low speed CPU.

The present invention is a conventional transmission video compression algorithm is configured to repeat a very complex operation takes a long time to perform the video compression and solve the problem that the system is more complicated than necessary and minimizes the amount of computation is possible in the existing algorithm It is a high-speed compression algorithm of transmission video of a new system that can be implemented in various applications by enabling the implementation even in a low speed CPU.

The present invention is to perform quantization (4) and DCT (3) in order to compress the information in the conventional information source encoder (2), to perform variable length coding (8) in the multiplex encoder (6) and to adjust the amount of data generation By using the HADAMARD transform, which is an orthogonal transform with a very small amount of computation in the HADAMARD operator 14, compared to using the coding controller 10 and inserting the error correcting code 12 to correct the error. If the same image block is compared with the previous screen, the format is generated separately from DC, which is a DC component, and AC, which is generated by performing a transformation by a method that does not transmit content, and the variable length encoder 21 can quickly distinguish the decoder. Express the compressed data in several kinds of lengths, and the error corrector 23 collects the headers and performs a checksum. The acid was 25, and inserts the error correction code is the Hamming cord (HAMMING CODE) (26) to correct an error during the transmission, and transmitted.

As described above, the transmission image algorithm of the present invention is characterized by a DCT (3) operation consisting of multiplication and addition of a conventional complex floating-point operation, quantization (4), which is a division operation of a floating point, and BIT of various lengths. Unlike the variable length coding with numbers (8) and the error correcting code (12), which repeats the complex calculation process, HADAMARD operator 14 with very small amount of information is compared with the previous screen. STO (SAME TO OLD BLOCK) to reduce the amount of data, and variable length code calculation (22) that can be quickly calculated by placing several BlT numbers that are easy to distinguish. It is possible to implement image compression that can not be implemented in a very small operation.

1 is a fast compression algorithm of a conventional transmission image

2 is a fast compression algorithm of a transmission video of the present invention.

3 is a HADAMARD matrix of the present invention.

3-1 shows the HADAMARD matrix

3-2 is OPTIONAL HADAMARD matrix

Figure 3-3 is a vertical matrix

3-4 shows the horizontal matrix

4 is a quantization table (QUANTIZATlON TABLE) of the present invention.

4-1 shows a quantization table relating to brightness

4-2 shows a quantization table for color

5 is a zigzag scan flow chart of the present invention.

6 is a variable length code table of the present invention

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

1: video input unit 2: information source encoder

3: DCT 4: quantization

5: motion compensation 6: multiplexing encoder

7: Hierarchy 8: Variable length coding

9: transmission buffer 10: encoding controller

11: transmission encoder 12: error correction code

13: Compressed information transmitter 14: HADAMARD operator

15: OPTIONAL HADAMARD 16: STO Search

17: quantizer 18: format generator

19: DC (DPCM) 20: AC (RUN LENTH CODING)

21: variable length encoder 22: variable length code calculation (DC, AC)

23: error corrector 24: header checksum (HEADER CHECK SUM)

25: CCITT CRC-16 calculation of data 26: Hamming coding

27: compression information transmitter 28: HADAMARD matrix

29: OPTIONAL HADAMARD matrix 29-1: vertical matrix

29-2: Horizontal matrix

30: QUANTIZATION TABLE

30-1: Quantization Table for Brightness

30-2: Quantization Table for Color

31: Zigzag Scan Sequence (ZIG-ZAG SCAN)

32: VATIABLE LENGTH TABLE

As shown in FIG. 2, the fast compression algorithm of the transmission image of the present invention includes a video input unit 1 receiving a signal from a video, a HADAMARD operator 14 and a quantizer 17 which compress image data. A format generator 18 for creating a format of compressed information, a variable length encoder 21 for representing the formed format with a small number of bits, an error corrector 25 for correcting errors occurring in transmission, and compressed information. Compressed information transmitter 27 for transmitting a.

The process of the fast compression algorithm of the transmission video of the present invention will be described in detail with reference to FIG.

The video input unit 1 sends the input video signal to the HADAMARD operator 14 in units of 8x8 pixel blocks so as to compress the input video signal.

In the HADAMARD operator 14, the HADAMARD matrix shown in Fig. 3 is a simple addition and subtraction of 8 × 8 pixels sent from the video input unit 1, rather than a complicated floating-point operation as in the conventional algorithm. 28) multiply horizontally and vertically to convert the input image to a frequency band. In addition, to increase the compression rate, OPTIONAL HADAMARD, a variation of the HADAMARD operation, is made so that the same data is compared with the previous image so that the same data can be restored without transmitting the data. STO search that calculates horizontal matrix (29-1) and vertical matrix (29-2) and then sends only the indication STO (SAME TO OLD BLOCK) After performing (16), if it is different from the previous image, the remaining HADAMARD operation is performed.

In the quantizer 17, a quantization table (QUANTIZATION TABLE) 30-1 for brightness, which is two quantization tables for brightness and color, as shown in FIG. 4, and a quantization table for color (QUANTIZATION TABLE) 30-2) to eliminate high frequency components of the image data.

In the format generator 18, among the 8 × 8 blocks calculated by the HADAMARD operator, the value at the top left is DC, and in the previous 8 × 8 block using DPCM (DIFFERRENTIAL PULSE CODE MODULATION) (19). After making the difference value of DC, RUN LENGTH CODING (20) of the scanned data using the ZIGZAG-SCAN method shown in FIG. Using the method, we express how many "0" s come out after "0".

In the variable length encoder 21, DC and AC are separately calculated. In order to distinguish the length of the code in 4 bit units with the DPCM 19 value of DC calculated by the format generator 18, the code length is changed to FIG. As shown in the figure, "0" and "1" are inserted in front of the code, S (SIGN) to put "+" and "-", and the actual data is put in D (DATA).

In addition, AC has the run length coding (20) value calculated by the former format generator 18 and changes the length of the sign in 4 bit units so that the separator "O" and "1" are inserted and the RUN value is set. Put in R (RUN) and put S and D. Thus, the amount of data is reduced by inserting a variable variable according to the probability of occurrence of the code. In addition, by incrementing by 4BIT, it is much easier to check by 1BIT when verifying by 4BIT when restoring compression.

The error corrector 25 performs a checksum 24 by collecting only the header HEADER having data generated by the variable length encoder 21 described above and having important information about compression. Then add CHECK SUM 24 for header to the data and add 2BYTE by performing CCITT CRC-16 (CYCLIC REDUNDANCY CHECK) 25 for error checking. Finally, a Hamming Code 26 is inserted to correct an error in the data to be transmitted. A typical Hamming Code 26 can correct errors by adding a Parity Bit in BIT units, but expanding the Hamming Code 26 in Bytes results in bytes (Byte). BYTE) errors can be corrected in units.

In the present invention, unlike the conventional transmission image compression algorithm, the conventional transmission image compression algorithm is configured to repeat a very complicated operation, it takes a long time to perform the image compression and the system becomes more complicated than necessary By solving the problem and minimizing the amount of computation and the amount of compression, it is possible to implement a low cost image transmission device such as 8bit CPU and 16bit CPU, which was not possible in the existing algorithm, and to implement a low cost image transmission device. It is a new fast compression algorithm for transmission video.

Claims (4)

In the method of compressing the transmission video at high speed, the HADAMARD calculator 14 performs HADAMARD transform, which is an orthogonal transformation with very small amount of calculation, and compares the content if it is the same image block compared to the previous screen. A process of reducing the amount of computation and compression by not transmitting, removing a high frequency component with two QUANTIZATION TABLEs 30-1 and 30-2 of brightness and color in the quantizer 17, Process of forming format separately from DC which is a DC component and AC component which is generated by performing HADAMARD conversion in format generator 18, and several kinds to quickly distinguish from decoder in variable length encoder 21 Expressing compressed data with the length of, and collecting the headers in the error corrector 23 to perform a check sum (Calculate the CRC with the generated data and to correct the error during transmission) And inserting and transmitting the Hamming code 26, which is an error correction code. The HADAMARD operation 14 according to claim 1, wherein the content is not transmitted if the same image block is compared with the previous screen by using the HADAMARD transform, which is an orthogonal transformation with a very small amount of calculation, and thus the amount of operation and the amount of compression In the process of reducing the value, we compute the horizontal matrix (29-1) and the vertical matrix (29-2) from the OPTIONAL HADAMARD matrix, which is a variation of the HADAMARD operation, and then the 4 × 4 value. And a STO search (16) which transmits only an indication of STO (SAME TO OLD BLOCK) if compared with the previous image. [4] The 4BIT according to claim 1, wherein the variable length encoder 21 has a DPCM 19 value of DC calculated by the format generator 18 in the course of expressing compressed data in several kinds of lengths so as to be quickly distinguished by the decoder. In order to distinguish by changing the length of the sign in units, as shown in Fig. 6, put "0" and "1" in front of the sign and put S (SIGN) to put "+" and "-". Put the data into D (DATA) and AC has the RUN LENGTH CODING (20) value calculated by the format generator 18 and changes the length of the sign in 4bit units so that the separators "0" and "1" , Put RUN value into R (RUN), and put S and D. The method of claim 1, wherein the error corrector 23 collects headers, performs a checksum, inserts a hamming code 26, which is an error correction code, to correct CRCs and correct errors during transmission. In the process of transmitting and transmitting the data, the data generated by the variable length encoder 21 are collected separately, and a header (HEADER) having important information about compression is separately collected to perform a checksum (24). 24) to the data and CCITT CRC-16 (CYCLIC REDUNDANCY CHECK) (25) to check the error, add 2BYTE, and finally correct the error to the data to be transmitted. ) Inserting the same.