KR100349058B1 - video compression and decompression Apparatus - Google Patents

Abstract

The present invention relates to an image compression restoring apparatus for compressing and restoring digital image data so as to effectively transmit digital image data and control signals through a line having a low transmission rate such as a public switched telephone network (PSTN). First storage means for storing the input image data; Compression restoring means for encoding, decoding and compressing and restoring the input image data by performing motion estimation operation, motion compensation operation, and data conversion on the image data input from the first storage means; Second storage means for storing the resultant data according to the compression mode determined by the motion estimation operation of the compression restoring means and the quantized and dequantized data by the compression restoring means for compressing the input image data; Third storage means for storing variable length decoded data and inverse discrete cosine transformed data in the compression restoring means for restoring the input image data; And control means for controlling the reading and writing of the first to third storage means and controlling the compression restoring operation of the compression restoring means by time division, wherein the first to third storage means and the compression restoring are performed. The means are characterized in that they are coupled in the form of a coprocessor around the control means.

Description

Video compression and decompression Apparatus

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image compression restoration device, and more particularly, to an image that compresses and decompresses digital image data so that digital image data and control signals can be effectively transmitted through a line having a low data rate such as a public switched telephone network (PSTN). A compression restoring apparatus.

Currently, various multimedia systems, including digital TVs, video telephones, and the like, process data in a digital manner, and thus all require the storage and processing of a significant amount of video data. Accordingly, various standards for compression and decompression have been established to efficiently store and process moving image data. In the case of still images it has been led by the Joint Picture Experts Group (JPEG) and in the case of moving pictures by the Moving Picture Experts Group (MPEG).

The MPEG standard is a standard that requires considerable image data processing capability and is not suitable for transmitting video using a low data rate line such as a public switched telephone network (PSTN). H.263 and H.261, recommended by the Telecommunication Standardization Sector, are applied as coding standards.

The H.263 and H.261 standards target communication media that require real-time encoding and decoding. Motion Estimation, Motion Compensation, Quantization, and Discrete Cosine Transformation A method of compressing and restoring moving images by block-based data transformation based on DCT: Discrete Cosine Transform is presented.

Generally, a motion compensation, motion estimation, and data conversion method is basically required to implement an algorithm for processing a video, and in order to implement it in hardware, a separate functional block can be appropriately combined or separated as described above. After the configuration is controlled by a program is used.

In the algorithm for processing a video, a representative video decoder according to the standard of H.261 has a structure as shown in FIG.

As shown in the figure, the image decoder includes a buffer 10, a variable length decoder 11, an inverse quantizer 12, an inverse discrete cosine transformer 14, a prediction memory 16, The adder 18 is comprised.

The image data processing of the decoder configured as described above is as follows.

First, the bitstream BS input from the outside, that is, image data encoded at a predetermined predetermined ratio, is temporarily stored in the buffer 10. In this case, since the buffer 10 has a predetermined size and cannot receive the bit stream continuously from the outside, the buffer 10 stores the bit stream in accordance with the rate at which image data is decoded.

The image data stored in the buffer 10 is decoded by the variable length decoder 11 and restored to data before being encoded, and then dequantized by the inverse quantizer 12.

Then, the data dequantized by the inverse quantizer 12 is subjected to inverse discrete cosine transform (IDCT) through the inverse discrete cosine transformer 14, thereby obtaining actual image data.

The data obtained by the inverse discrete cosine transform (IDCT) has two forms as follows.

First, if the current frame is intra, that is, intra-frame encoding, there is no relationship with the previous frame, and thus the present frame is reproduced as it is. In this case, data that has undergone inverse discrete cosine transform (IDCT) becomes data of the current frame.

Secondly, if the current frame is Inter, that is, inter-frame encoding, the current frame includes a constant motion vector (MV: Motion Vector) according to the relationship with the previous frame, and the previous frame to improve the compression efficiency. Only the difference value is passed.

Therefore, when the current frame is Inter, the result of the inverse discrete cosine transform (IDCT) is not the data of the current frame, but the motion compensation process with the previous frame data previously stored in the prediction memory 16. do. To this end, the data obtained by the inverse discrete cosine transform (IDCT) is compensated by the difference value with the previous frame stored in the prediction memory 18 in the adder 16.

All of these processes are performed based on unit blocks by dividing the entire screen into blocks that are basic units. Where 1 block is 8 It consists of eight pixels.

Meanwhile, encoding for compressing and transmitting a video undergoes a process opposite to that of the above-described decoding, which will be described below.

First, a current frame to be compressed and transmitted is divided into a macro block and motion prediction is performed using image data of a previous frame. Here, one macro block is one unit composed of six blocks, that is, four blocks including brightness components and two blocks including color components.

When the motion vector for the corresponding macroblock of the current frame is obtained through the motion prediction, it is determined whether the current macroblock to be transmitted is intra or inter.

As a result of the above determination, since the current macro block has no direct relation with the image data of the previous frame in case of Intra, the current macro block data is transmitted as it is, and in the case of Inter, the corresponding macro in the previous frame Only the difference value of the image data is transmitted based on the block. In this case, the transmitted beak string is sent with a motion vector that knows the position of the macroblock in the previous frame so that it can be used during the encoding process.

Then, the image data obtained as the difference value from the current frame itself or the previous frame is converted from the time domain to the frequency domain through a discrete cosine transform (DCT) process, and the converted data is quantized through a quantizer and finally The compressed data is transmitted through an external output terminal in a compressed form through a variable length encoder (not shown), thereby completing the compression process of the video data.

As described above, since the current frame data to be compressed requires data of a previous frame, the data to be currently encoded is not completed by compression, but the current frame should be restored using the compressed and transmitted data. This is done for two purposes:

First, by reconstructing the current frame using the current compressed data, it can be used as the data referenced in the next frame,

Secondly, the process performed by the decoder that receives and restores the compressed data is reproduced as it is, thereby minimizing an error between the receiver and the data exchanged.

In order to perform the video processing by decoding and encoding as described above, the following two methods can be used.

First, it is a software method for programming both the decoding and encoding processes to be performed in a processor including a function of a high speed digital signal processor (DSP),

Second, a hardware method of increasing the processing speed by implementing specific blocks in hardware during the decoding and encoding processes.

This application method can be suitably used depending on the system to which the above algorithm is applied.

However, the following problems occur in implementing the actual system by applying the video processing algorithm as described above.

First, when implemented by the video processing algorithm program, a high performance processor is required to process in real time, and unnecessary resources are used to overcome the limitation of real time processing, which leads to a cost increase of the system.

Next, when the video processing algorithm is implemented in hardware, a lot of hardware resources are required for the design of dedicated functional blocks such as motion estimation, motion compensation, discrete cosine transform, inverse discrete cosine transform, and so on. The efficiency of the processing performance is lowered.

In order to solve the above problems, an object of the present invention is to connect an image data processor and related memories in a coprocessor format based on a micro sequencer, thereby effectively moving video data by effectively operating individual functional blocks while having a simple hardware structure. An object of the present invention is to provide an image compression restoring apparatus that can be compressed and restored.

In order to achieve the above object, the image compression restoring apparatus of the present invention comprises: first storage means for storing image data input from the outside; Compression restoring means for encoding, decoding and compressing and restoring the input image data by performing motion estimation operation, motion compensation operation, and data conversion on the image data input from the first storage means; Second storage means for storing the resultant data according to the compression mode determined by the motion estimation operation of the compression restoring means and the quantized and dequantized data by the compression restoring means for compressing the input image data; Third storage means for storing variable length decoded data and inverse discrete cosine transformed data in the compression restoring means for restoring the input image data; And control means for controlling the reading and writing of the first to third storage means and controlling the compression restoring operation of the compression restoring means by time division, wherein the first to third storage means and the compression restoring are performed. The means are characterized in that they are coupled in the form of a coprocessor around the control means.

In order to achieve the above object, the image compression restoration method of the present invention comprises the steps of: determining whether to compress or restore the input image data; Performing step compression of motion estimation, motion compensation, discrete cosine transform, quantization, and variable length coding when it is determined that the result of the determination step is compression; Performing step reconstruction of variable length decoding, inverse quantization, inverse discrete cosine transform, and motion compensation when it is determined that the result of the determination step is restored; When performing the compression process, it is checked whether the restoration is useful for each compression process, and if the restoration is useful as a result of the check, the next compression process is suspended, and the useful restoration process is first performed. Performing the next compression process that is suspended; And checking whether compression is useful for each restoration process when performing the restoration process, and if the compression is useful as a result of the check, suspend the next restoration process and perform the useful compression process first, and when compression is not useful. And performing the next restoring process held by the block, and compressing and restoring image data by time division.

1 is a block diagram showing a conventional image data decoder.

Figure 2 is a block diagram showing an image compression restoration apparatus according to an embodiment of the present invention.

Figure 3 is a flow chart for explaining a compression restoration method of the image compression restoration apparatus according to an embodiment of the present invention.

4 is a flowchart illustrating a motion estimation algorithm according to an embodiment of the present invention.

* Explanation of symbols for the main parts of the drawings

100: first storage unit 110: compression recovery unit

120: second storage unit 140: third storage unit

160: control unit DB1, DB2: data bus

AB: address bus

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

2 is a block diagram illustrating an image compression restoration apparatus according to an embodiment of the present invention. As shown in the drawing, the first storage unit 100, the compression restoration unit 110, and the second storage unit 120 are largely shown. And a third storage unit 140 and a control unit 160, each block, that is, the first storage unit 100, the compression restorer 110, the second storage unit 120, The third storage unit 140 is superimposed in the form of a separate coprocessor around the control unit 160 to perform functions corresponding independently and in parallel as a separate block, and receives a command from the control unit 160. Processing relating to the start and end of all operations that are performed.

The first storage unit 100 is composed of a dynamic random access memory (DRAM), which is a volatile memory, according to an embodiment of the present invention, video data input from an external camera, video data output to an external display device, and compression. Stores all video data processed in real time, such as video data and compressed video data before it is received, receives a read and write control signal from the control terminal ctrl4 of the controller 160, and accesses the read control signal. The stored video data is transmitted to the controller 160 through the data bus DB1.

The compression restoring unit 110, that is, the image data processor performs various image processing algorithms satisfying the H.263 standard as a calculation block, and the second storage unit 120 through the data bus DB2 and the address bus AB. And access the third storage unit 140 and receive video data stored in the first storage unit 100 through the control terminal ctrl2 of the controller 160.

In general, since a common operation block is used to decompress a video, the compression restoration unit 110 according to an embodiment of the present invention effectively operates a common block without separately designing a dedicated block for encoding and decoding. A motion estimation calculation block (not shown), a motion compensation calculation block (not shown), and a data conversion block (not shown) are included to perform compression restoration in real time.

The motion estimation arithmetic block (not shown) includes a first memory for storing image data of a macroblock being processed by an image compression process in a current frame, and a top, bottom, center of the same position of a previous frame for motion estimation. A second memory for storing image data of an area extended by 8 pixels each for left and right, a multiplexer for selecting a relative position of each pixel of current and previous data, a data of the current frame and a previous frame for motion estimation Comparator for making the final decision about the difference data calculated using the current and previous frame, that is, the calculation element to subtract and accumulate using the data of And find the most similar part of the macro block to be processed in the current frame from the previous frame. Calculates a motion vector and makes the resulting data available to the motion compensation calculation block. However, the motion estimation arithmetic block is used only for video encoding, so it is not used in the decoding process.

The motion estimation process of the motion estimation calculation block configured as described above is as follows.

First, image data of a macro block to be processed in a current frame is read from an external first storage unit 100 and stored in a first memory (not shown), which is a local memory in a motion estimation calculation block, and The corresponding area of the previous frame data required for comparison is read from the external first storage unit 100 and stored in a second memory (not shown) which is a local memory in the motion estimation calculation block.

In this state, a motion estimation operation of an integer pixel unit is performed using data stored in the first and second memories (not shown).

As a result of the motion estimation operation, a motion vector of an integer pixel unit having the least error is obtained. The motion estimation operation of a half pel unit is performed again based on this value.

When the motion estimation operation in units of half pels is completed, the operation result is used to determine whether the mode of the macroblock currently being processed is intra or inter. The motion vector and the mode of the current macroblock according to the motion estimation operation result of the half pel unit are transferred to the motion compensation calculation block so that the compression vector can be used in the subsequent compression process.

The motion compensation calculation block (not shown) is a block that performs motion compensation on a previous frame in the process of compressing and restoring an image, and performs three functions as follows by a program.

First, when the data input from the outside is subjected to variable length coding (VLC) and inverse discrete cosine transform (IDCT) when the compressed image data is restored, all coefficients and motion vectors of the corresponding block are obtained. When the data are in the inter mode, they correspond to difference values from the previous frame, and thus, motion compensation using the data of the previous frame is performed again to restore an image.

Second, when the corresponding macroblock of the current frame is to be processed in the inter mode during image compression, the corresponding frame of the previous frame from the macroblock of the current frame is obtained by using the motion vector according to the operation result of the motion estimation calculation block. The difference value is obtained by subtracting the macroblock data value.

Third, after all compression is completed in the image compression process, inverse quantization and inverse discrete cosine transform (IDCT) of the data after the final operation is performed, and the current frame data before compression is restored by using the data obtained therefrom. It can be used when compressing frames.

The data conversion block (not shown) includes a discrete cosine transformer and an inverse discrete cosine transformer, a quantizer and an inverse quantizer, a variable length encoder and a variable length decoder, and have undergone pre- or post-processing to compress and decompress an image. It is responsible for converting the data in the time domain to the data in the frequency domain and vice versa, and finally compressing or restoring the converted data.

The discrete cosine transformer converts the data in the time domain input from an external camera to the frequency domain data, and the inverse discrete cosine transformer converts the compressed data input from the outside, that is, the data of frequency domain Performs a function that converts data into data.

The quantizer performs a function of quantizing by a specific value in order to increase the transmission efficiency of the data converted through the discrete cosine transformer, and the inverse quantizer is used to restore the original value of the data converted by the inverse discrete cosine transformer. Dequantizes by a specific value.

The variable length coder performs a function of finally compressing data on the quantized data, and the variable length decoder performs a function of finally restoring the data on the dequantized data. Since these variable length encoders and variable length decoder blocks operate all operations using a code table already defined in H.263 or H.261, it is possible to compress and restore them in various ways by changing only these code tables.

On the other hand, the second storage unit 120, that is, the encoding memory is a common memory for image compression used when the compression storage unit 110 performs encoding, and is provided on the 4-bit address bus AB and the 32-bit data bus DB2. Is accessed by. Therefore, the size is 16 words 36 bits.

In general, all image compression in a moving picture processing algorithm is performed based on a unit block or a macro block. In more detail, the entire frame is 8 It is divided into a group of 8 blocks, which is used to perform all data processing. In the case of motion estimation, four unit blocks are performed in macroblock units, and motion compensation is performed based on unit blocks.

Therefore, since the second storage unit 120 only needs to compensate for the motion of the unit block, 16 words as described above. It is preferably composed of dual port memory having a size of about 36 bits and having two read and two write ports for performance.

Meanwhile, the third storage unit 140, that is, the decoding memory, is a common memory for image restoration used when the compression storage unit 110 performs encoding, and is provided on the 4-bit address bus AB and the 32-bit data bus DB2. Accessed by the 16 words, similar to the decoding memory It is preferably composed of dual port memory having a size of about 36 bits and having two read and two write ports for performance.

Meanwhile, the controller 160, that is, the micro sequencer, controls reading and writing of the first to third storage units 100, 120, and 140 and the internal memory, and specifies an operation mode for each block of the compression restoring unit 110. Control the data flow, perform calculations on the bit rate of the image data being compressed or reconstructed, process interrupts generated in each block of the compression restorer 110, calculate motion vectors, and perform compression restorer 110. Image mapping of the data calculated in each block of) is performed, and the coding and decoding of the compression storage unit 110 are controlled by time division.

Referring to the operation of the compression restorer according to an embodiment of the present invention.

First, the compression process according to an embodiment of the present invention, video data input from an external device such as a camera is copied to the designated area of the first storage unit 100 before the compression process. The reason for this is that the difference between the transmission rate of the data input from the outside and the rate of performing the compression process inside the processor, it is not possible to process the input video data immediately.

At this time, the form of the image data stored in the first storage unit 100 is in the form of QCIF (Quarter Common Intermediate Format) or CIF (Common Intermediate Format) according to the picture format defined in H.263.

The image data stored in the first storage unit 100 is provided to the compression restoring unit 110 through the control terminal ctrl2 of the controller 160, where the shape of the frame is determined. To this end, the compression storage unit 110 performs a determination process based on motion estimation.

As is known, motion estimation is performed using a macroblock to be compressed in the current frame and a macroblock at the same position of the previous frame. That is, a difference calculation is performed for each pixel unit for regions that are made up by -8-+8 pixels based on the data of the macroblock being processed in the current frame and the position thereof, based on the same position of the previous frame. This is the current macro block of 16 Each of the 16 pixels contains 32 values of the same-position macro block of the previous frame. The difference between the pixel values and the difference value of the region extended to 32 is obtained.

By this method 32 The position having the smallest difference value in the 32 pixel area is obtained, and the motion vector is obtained for the macro block currently being processed in the current frame. Here, the motion vector is a value used for motion compensation later.

Next, it is determined in which mode to transmit the current frame based on the motion vector. That is, it is determined whether to compress the current macro block in the form of intra or in the form of inter, and the shape of data stored in the second storage unit 120 is changed according to the result.

If it is to be compressed in the form of intra, since the current frame has no association with the previous frame, the motion vector has a value of "0". In this case, the input image data is stored in the second storage unit 120 without any manipulation.

If it is to be compressed in the form of an inter, the motion vector has a constant value since the macroblock at the current position plus the motion vector has a very similar correlation with the macroblock to be processed. In this case, in order to increase the compression efficiency, a method of compressing all the data of the current macro block as it is, instead of compressing only the difference value with the macro block of the position according to the motion vector is used, and the second storage unit 120 The difference value between the macroblock of and the corresponding macroblock of the previous frame is stored.

Next, since the data in the second storage unit 120 stored in the form of intra or inter according to the compression mode of the macro block is data in the time domain, the frequency is determined by a discrete cosine transformer, which is an internal block of the compression restoring unit 110. Converted to data in the area. The converted data is stored in the second storage unit 120 after being quantized by the quantizer, which is an internal block of the compression restoring unit 110, and is input to the variable length encoder which is an internal block of the compression restoring unit 110. After variable length coding, it is transmitted to the outside.

If the compression mode of the currently processed macro block is intra, all the operations are completed by the compression process as described above, but in the inter mode, an additional process is required. As is well known, in case of inter, the current compressed and transmitted data is not the complete data of the current frame but only the difference value from the previous frame. Therefore, the data for the previous frame is needed to restore the original frame. Done.

Therefore, after the above compression process is completed, a process of restoring the compressed data again is necessary.

To this end, in the embodiment of the present invention, a reconstruction process is performed once again using data that has undergone only discrete cosine transform and quantization stored in the second storage unit 120.

In order to restore the compressed data again, the inverse quantizer and the inverse discrete cosine transformer, which are internal blocks of the compression restoring unit 110, access the second storage unit 120 to recover data before the final compression, that is, data in the frequency domain. The data is restored to the time domain, and the restored data is stored in the second storage unit 120 again.

Next, the motion compensation block inside the compression restoring unit 110 accesses the data converted into the time domain stored in the second storage unit 120 and restores the original image data. The previous frame data is found using the motion vector obtained from the motion estimation block inside the compression restoring unit 110, and the current data uses the data converted into the time domain stored in the second storage unit 120.

The image data reconstructed in the motion compensation block is stored in a specific region of the first storage unit 100 after all compression on the current frame is completed and used as frame data for reference for compression of the next frame.

Referring to the restoration process according to an embodiment of the present invention, first, the compressed video data input from the outside is stored in a specific area of the first storage unit 100, and the control terminal (ctrl2) of the control unit 160 during restoration The data is input to the compression restoring unit 110 and decoded by an internal variable length decoder and stored in a designated area of the third storage unit 140.

Next, the compression restoring unit 110 accesses the third storage unit 140 to read the stored encoded data, and inversely discretely decodes the read data, that is, data in the frequency domain, as an internal block of the compression restoring unit 110. In the cosine transformer and the inverse quantizer, the inverse discrete cosine transform and inverse quantization are converted into data in the time domain. The converted data is again stored in the designated area of the second storage unit 140.

The inverse discrete cosine transformed video data undergoes a corresponding motion compensation process according to the mode of the frame and the mode of the macroblock currently being processed in the motion compensation block inside the compression restoring unit 110.

If the macro block currently being processed is in the form of intra, the previous frame is not necessary separately for the final restoration of the current macro block. therefore. Since the macro block currently being processed contains a complete video by itself, the data stored in the third storage unit 140 through the inverse discrete cosine transform and inverse quantization without any motion compensation is required. The data is transmitted to and stored in the first storage unit 100.

In the compression and reconstruction of an image according to an embodiment of the present invention, a method of controlling and separating the image into a compression process and a reconstruction process is basically used. Motion compression, motion compensation, discrete cosine transform, quantization, Variable length coding is performed, and variable length decoding, inverse quantization, inverse discrete cosine transform, and motion compensation are performed for image reconstruction.

3 is a flowchart illustrating a compression restoration method of an image compression restoration apparatus according to an embodiment of the present invention. In the figure, a portion ST200 denoted encoding represents a flow for compressing input video data, and a portion ST300 denoted decoding represents a flow for restoring compressed image data, and according to an embodiment of the present invention, compression is performed. The and recovery process can only be transitioned through the steps of ENC pole (ST400) and DEC pole (ST500), respectively.

The ENC pole performs each stage of compression and is always performed after the current stage is completed before moving on to the next compression stage. The ENC pole first checks whether the reconstruction step is useful or not useful, i.e. whether the decoder is enabled or disabled, transitions to the reconstruction step if available, and continues with the next compression step if not available. Let's proceed.

According to an embodiment of the present invention, there is a flag indicating whether the restoration is useful in the controller 160, and whether or not the transition to the restoration step is determined by examining the state of the flag. In addition, the current compression state is stored in the control register R1 before the transition to the restoration step, which allows the user to return to the compression step that is currently in progress after the specific restoration step is completed.

The DEC pole is a step that is always performed after each step of recovery and before the current step is completed, before moving on to the next recovery step. The DEC pole first checks whether the compression step is useful or not useful, i.e. whether the encoder is enabled or disabled, transitions to the corresponding compression step if it is available, and if not available, continues with the next reconstruction step. Let's proceed.

According to an embodiment of the present invention, there is a flag indicating whether compression is useful in the control unit 160. The presence or absence of a transition to the compression step is determined by checking the state of the flag. Then, before the transition to the compression step, the current recovery state is stored in the control register R2, which allows the user to return to the recovery step exactly after the specific compression step is completed.

Hereinafter, a compression restoration method of an image compression restoration apparatus according to an embodiment of the present invention will be described with reference to the flowchart of FIG. 3.

First, the idle loop (idle_loop) determines whether to compress or restore the input image data (ST100).

In more detail, the idle loop is a part that determines whether the hardware is ready to perform compression or restoration and notifies the start of data processing. That is, by using the control registers R1 and R2 inside the control unit 160, it is possible to forcibly control compression and restoration internally or externally. In this step, the values of the control registers R1 and R2 are continuously maintained. In this case, it determines whether to compress or restore and proceeds according to the judgment.

When it is determined that the compression results in the 100th step (ST100), the value of the control register (R) is set to "0", and the compression process of the motion estimation, the motion compensation, the discrete cosine transform, the quantization, and the variable length coding is phased. It is performed (ST200).

According to an embodiment of the present invention, in step 200, steps of ENC0, Wait CCD, ENC1, ENC2, ENC3, and Encode reset are sequentially performed as the compression process.

In the ENC0 step, the control register R1 is set to "4" as a starting point when encoding is useful after determining whether to perform compression or restoration in an idle loop as a part of initial compression. Therefore, by performing the restoration and returning to the control register R1, it is possible to perform the next Wait CCD step.

The wait CCD step is a step of determining whether the compression restoring unit 110 can process a signal input from a CCD which is an image display device connected to the outside of the system. If the CCD is valid, the compression process continues, otherwise it waits until the CCD is valid.

The ENC1 step performs the following functions.

First, compression is performed through a variable length encoder to encode only a portion corresponding to a picture header of a corresponding frame, a total number of bit strings encoded to date for quantization is calculated, and a quantization value is calculated based on this.

Next, a motion estimation operation is performed to determine a mode, i.e., intra or inter for the currently input image data. In this case, the motion estimation calculation block inside the compression restoring unit 100 is used. Since the motion calculation process is a time-consuming part, it is inefficient to perform this process every time each macro block is compressed. In particular, since the motion estimation process is used only during compression, there is no problem of time division with the reconstruction process, such as a motion compensation block, a discrete cosine transform block, and an inverse discrete cosine transform block. Considering that this is only once, this process can be performed in parallel with the operation with other functional blocks.

Therefore, the above process differentially processes all the steps of the above-mentioned motion estimation according to the position of the macro block and proceeds to the next step. Once the motion estimation for the first macro block is completed, the data of the macro block that comes after the macro block currently being processed in parallel is processed in parallel during the subsequent compression process for the next six blocks of the macro block. Reads from the storage unit 100 and stores in a first memory (not shown) which is a local memory in the motion estimation calculation block.

Next, using the motion vector calculated by the motion estimation block, the portion having the most similar difference value in the macro block of the current frame is found in the previous frame, and the motion compensation process is performed using these two data.

When the motion compensation process is completed, discrete cosine transform and quantization are performed by using the difference output between the data of the current frame and the previous frame data.

When the above process is completed, it is determined whether the restoration process is performed through the ENC pole, and if the transition to the restoration step is performed, the control register R1 is set to "2" to perform the restoration process. Allow it to start again.

In step ENC2, a compressed code is obtained through a variable length encoder inside the compression restoring unit 110 for data that has undergone a discrete cosine transform and quantization process in step ENC1.

The compressed code thus obtained is data based on the difference value with the previous frame, and thus needs to be restored to the original frame data.

In addition, in the ENC2 step, when the macroblock to be processed is not the first of the frame, each of 8 blocks in the up, down, left, and right directions based on the position of the macroblock next to the macroblock currently being processed in the previous frame is used. The expanded area is calculated and the data of the area is read from the external first storage unit 100 and stored in a second memory (not shown) which is a local memory in the motion estimation calculation block.

When all the above processes are completed, it is determined whether the restoration process is performed through the ENC pole. If the transition to the restoration step is performed, the control register R1 is set to "3" to perform the restoration process. Allow it to start again.

In step ENC3, a motion compensation operation is performed using the image data and previous frame data reconstructed in step ENC2, and when the macroblock to be processed is not the first, the macroblock data that comes after the current macroblock is processed. The motion estimation operation is performed by using In this case, the resulting various data, that is, the motion vector in the form of a macro block, can be used immediately without time delay when performing the process of the steps ENC1 to ENC3 for the next macro block.

When the above process is completed, it is determined whether the restoration process is performed through the ENC pole, and if the transition to the restoration phase is performed, the control register (R1) is set to "4" to perform the restoration process and return to Encoder reset. Allow you to start over.

The encoder reset step is a step performed after the encoding for one frame is completed or when all the sequences to be compressed are terminated and the compression process is no longer valid. The compression process is enabled and the process proceeds to the next step. When this step is finished, it is determined whether to perform compression through the ENC pole step. If the transition to the recovery step is performed at this time, the control register R1 is set to "0" to restore and then return from the ENC0 step. Get started.

On the other hand, if it is necessary to restore the determination result of the 100th step (ST100), the variable length decoding, inverse quantization, inverse discrete cosine transform, and the restoration of motion compensation are performed step by step (ST200).

According to an embodiment of the present invention, in step 300, the steps of DEC0, Wait LCD, DEC prestage1, DEC prestage2, and DEC prestage3 are sequentially performed as the restoration process, and DEC1, Perform the steps of DEC2, DEC3, and DEC4, and finally perform the steps of Decoder reset.

The DEC0 step is a step in which an initial restoration process is started. If the restoration is useful after determining whether to process compression or restoration in the idle loop, it starts. In this step, the control register R5 is set to " 5 " so that compression can be performed after completing the current step, that is, the DEC0 step, and then resumed from the next Wait LCD step. do.

The wait LCD step is a step of determining whether it is possible to process a signal restored from the LCD which is an image display device connected to the outside of the system. If the LCD is valid, the restoration process will continue. If not, the LCD will continue to wait until the LCD is valid.

In the DEC prestage1 step, the compressed image data received from the external first storage unit 100 is received and searched in units of 1 bit, and the first start code of the frame, that is, PSC (Picture Start Code) is found. In this case, a variable length decoder inside the compression restoring unit 110 is used, and the compressed data is continuously read until the PSC is found.

In the DEC prestage step 1, if the compressed image data read from the outside is insufficient, that is, if the input buffer is empty or one of the desired PSC codes is found, the current step is terminated. It is determined whether the compression process is performed. If the transition to the compression process occurs, the former sets the control register (R2) to "6" to perform compression and, upon returning, to start the current step again to find the PSC code, and in the latter case the control register (R2). ) Is set to "7" so that compression can be performed consecutively from the next DEC prestage2 on return.

In the DEC prestage2 step, the compressed image data input from the external first storage unit 100 is received and searched in units of 1 bit, and the GOB (Group of Block) or MB (Macro Block) header of the frame is found. In this case, the variable length decoder inside the compression restoring unit 110 is used, and the compressed image data is continuously read until the header portion is found.

In the DEC prestage2 step, if the compressed image data read from the outside is insufficient, that is, if the input buffer is empty or one of the desired header parts is satisfied, the current step is terminated and the DEC pole is terminated. It is determined whether the compression process is performed. At this time, if the transition to the compression process, in the former case the control register (R2) is set to "7" to perform the compression and return to start again from the current step, and in the latter case the control register (R2) " Set it to 8 "so that compression can be performed sequentially when the next time it comes back.

In the DEC prestage 3 step, the compressed image data input from the external first storage unit 100 is received and searched in units of 1 bit, and the data for the block that is the minimum unit of the frame, that is, 64 pixels in each block The coefficient values are found and stored in the third storage unit 140. In this process, the variable length encoder inside the compression restoring unit 110 is used, and a block unit search is performed.

Next, in the DEC prestage 3 step, it is determined in which mode the block data received from the variable length decoder is compressed. In other words, it is possible to apply the corresponding reconstruction algorithm by judging which category among the types of macro blocks specified in H.263, and performing inverse discrete cosine transform and inverse quantization according to the determination result and then variable length decoding. The final motion vector is calculated using the motion vectors interpreted and accepted by the machine.

The operation in the case where it is possible to proceed to the next step in the DEC prestage 3 step is as follows.

First, when the compressed image data read from the outside is insufficient, that is, the input buffer is empty, the current step is terminated, and it is determined whether the compression process is performed by the DEC pole. At this time, if the transition to the compression process, the control register (R2) is set to "8" to perform the compression and return to continue from the current step again when returning.

Second, when the information on the block is analyzed and the form of the macro block is not coded, that is, when there is no count value for the block currently being processed, the macro block data at the same position of the previous frame is retained. In addition, the current step is terminated, and the DEC pole determines whether the compression process is performed. At this time, if the transition to the compression process, the control register (R2) is set to "1" to perform the compression and return to start from the next DEC1 step in succession.

Third, when the information on the block is analyzed and the macroblock has no coefficient, that is, when the macroblock of the part designated by the motion vector in the previous frame is used as it is, the current step ends, and the DEC The pole determines whether the compression process is performed. At this time, if the transition to the compression process, the control register (R2) is set to "2" to perform the compression and return from the next DEC2 step to start continuously.

Fourth, if the result of analyzing the information on the block is that all the coefficient values are present and the inverse discrete cosine transform is required, data conversion is performed using an inverse discrete cosine transformer and inverse quantizer, which are blocks inside the compression restoring unit 110. After the current step, it is determined whether the compression process is performed by the DEC pole. At this time, if the transition to the compression process, the control register (R2) is set to 3 so that the compression can be started from the next DEC3 step when returning.

Meanwhile, the DEC1 step compensates for the motion of each block by using the motion vector obtained through the variable length decoder. At this time, since the motion vector to be used has already been obtained through the DEC prestage 3 step, in this step, the position of the block corresponding to the previous frame is found and read from the external first storage unit 100 using the calculated motion vector. The final current frame data is restored through the motion compensation block inside the compression restoring unit 110.

The case where the DEC1 step can be performed is when the current block is in an inter form and is not coded.

On the contrary, when the DEC1 step is completed, it is divided into when all six blocks included in one macro block are processed or when all six blocks are not processed. In the former case, the process proceeds to the DEC prestage2 step so that the data for the next macro block can be input through the variable length encoder, and in the latter case, the remaining blocks must be processed. After completion, the compression process is performed by the DEC pole. In this case, if a transition to the compression process occurs, the control register R2 is set to "1" so that the current step can be restarted when the compression is performed and returned.

The DEC2 step performs the same function as the DEC1 step, but can only perform this step, when the macroblock type interpreted in the DEC prestage3 is inter and the encoded form is "0". The coded block pattern (CBP) is "0". If this step can be terminated is the same as the DEC prestage 3 will be omitted below.

The DEC3 step performs the same function as the DEC1 step, but if only this step can be performed, the coded block pattern (CBP: Coded Block Pattern (CBP)) is an intra or inter form of the macroblock interpreted in the DEC prestage3 step. Is "1". When this step is finished, it is determined whether or not the compression process is performed by the DEC pole step. At this time, if it is transitioned to the compression process, the control register R2 is set to "4" to perform compression and return. Allow you to start over.

The DEC4 step finally performs motion compensation with the result of the DEC3 step. The end of this step is divided into when all six blocks are processed or when all six blocks are not processed. In the former case, the process proceeds to the DEC prestage2 step so that the data for the next macro block can be input through the variable length decoder, and in the latter case, the remaining blocks must be processed. To be input through the variable-length decoder.

Decoder reset step according to an embodiment of the present invention is a step performed after the completion of the compression for one frame, or when all of the sequence to be restored is finished and the restoration process is no longer used. After the completion of this step, it is determined whether the compression is performed through the DEC pole. If the transition to the compression step occurs, the control register R2 is set to "0" to perform compression and return. When starting again from the DEC0 step can be continued.

As described above, when performing the 200th step (ST200), it is checked whether the restoration is useful through the ENC pole step every compression process, and if the restoration is useful as the result of the inspection, the next compression process is suspended and the restoration is performed. The process is first performed, and if the restoration is not useful, the next compression process is performed (ST400).

On the other hand, as described above when performing the 300 step (ST300), it is checked whether the compression is useful through the DEC pole step in each restoration process, and if the compression is useful as a result of the inspection, the next restoration process is suspended. The corresponding compression process is first performed, and if the restoration is not useful, the next restoration process is held (ST400).

In accordance with an embodiment of the present invention, the compression and decompression processes as described above are each pipelined to speed up the processing of the data.

According to an embodiment of the present invention, the pipeline of the compression process may include performing variable length coding on a motion estimation operation and a picture header, performing motion compensation and discrete cosine transform operations, and coefficients of each macro block. Performing variable length coding and inverse discrete cosine transform operation, and performing motion compensation and variable length coding.

Meanwhile, according to an embodiment of the present invention, the pipeline of the reconstruction process may include: finding an initial start code of a frame, restoring a picture header, restoring a mode of a macro block, and restoring each block in the macro block. Restoring information about the current macroblock; processing a case in which the current macroblock is inter-shaped and not coded; processing a case in which the current macroblock is inter-shaped and the coded block pattern is "0"; It is divided into a process of processing a case in which the block is intra or inter and the code block pattern is "1", and checking whether the external display device can be used.

4 is a flowchart illustrating a motion estimation algorithm according to an embodiment of the present invention.

A commonly used motion estimation block is characterized in that it is performed only once every macroblock is processed in the compression process, and is not used until the next macroblock is processed. That is, the motion estimation block performs motion estimation on the next macro block after processing all six blocks inside the macro block currently being processed.

Therefore, in the embodiment of the present invention, the motion estimation block is designed to operate in parallel with other blocks in order to overcome the above limitations and to increase the efficiency of use of the hardware, and also the entire process of the motion estimation block in three steps. By performing separate, it was possible to parallel operation in each step.

Referring to FIG. 4, the motion estimation process will be described below.

First, if the current macroblock to be compressed is the first macroblock of the frame, all the motion estimation processes for the current macroblock are performed without the parallel execution part (ST600). The motion estimation process performed at this time is performed in the manner used in the detailed description of the motion estimation block, and thus the detailed description thereof will be omitted.

When the 600th step ST600 is completed, a motion compensation process is performed on the image data to be compressed (ST610). In operation 610, the difference value between the macroblock to be processed in the current frame and the corresponding macroblock of the previous frame obtained based on the vector obtained through the motion estimation operation is calculated.

A discrete cosine transform is performed to convert the image data obtained by the difference value in step 610 into the frequency domain (ST640). At this time, in the motion estimation block, the position on the first storage unit 100 is found in parallel with respect to the macroblock next to the macroblock being processed in the current frame, and the corresponding image data is read into the memory inside the motion block. The operation is performed (ST620), and then it is determined whether the restoration process is performed (ST630).

In step 640, the variable length coding is performed on the data on which the discrete cosine transform is completed to generate compressed data (ST650), and then the inverse discrete cosine transform is performed again to restore the current frame. To perform (ST660). At this time, the motion estimation block reads data of a previous frame from the first storage unit 100 in parallel (ST670). That is, based on the macro block to be processed, the position of the region extended by 8 pixels each up, down, left, and right is found from the previous frame, and the corresponding data is stored from the first storage unit 100 using the same. The process of reading into the internal memory of the motion estimation block is performed.

Upon completion of step 670 (ST670), the motion estimation block performs a final motion estimation operation using current macroblock data read into the internal memory and macroblock data of a previous frame (ST680). This step ST680 is performed in parallel with the 660th step ST660.

Upon completion of the step 680 (ST680), it is determined whether the restoration process is useful to determine whether the transition is useful (ST690). When the compression process is useful, the data estimated by the motion estimation operation in the step 680 (ST680), ie, inverse Data reconstructed in the time domain by the discrete cosine transform operation undergoes a motion compensation process using previous frame data in order to be finally reconstructed into the current frame data (ST700). In this case, the motion estimation block performs a final motion estimation operation by using the current macro block that has already been processed in parallel and read into the internal memory and the macro block data of the previous frame, and calculates a motion vector.

Next, in step 710 (ST710) it is determined whether the final calculation of the motion vector being performed in the current motion estimation block by the number of blocks. If the number of blocks is six, the motion vector is calculated as described above. If the number of blocks is 6 or less, the process returns to the motion compensation step of step 610 (ST610).

The motion compensated data through the step 710 (ST710) is finally compressed through a variable length encoding step (ST720).

By doing so, in the embodiment of the present invention, the current macroblock is compressed and simultaneously the motion estimation process for the next macroblock is completed together. When the next macroblock is compressed, the motion estimation process is performed again. It is not necessary to just use the data resulting from the previous motion estimation process.

As described above, in the present invention, by connecting the image data processor and related memories in the form of a coprocessor around the micro sequencer, individual functional blocks can be operated in parallel with an independent processor separate from the micro sequencer. It is easy to add and modify future functional blocks, and all functional blocks can be operated by a program of a micro sequencer, thereby ensuring the flexibility of hardware according to an application.

In addition, in the present invention, by performing the encoding and decoding in a time division manner, there is another effect that can increase the hardware utilization of the blocks used for encoding and decoding.

In addition, in the present invention, the motion estimation block is performed in parallel with other functional blocks, thereby enabling another function of maximizing the function of the motion estimation block to be performed only once for each macroblock during encoding.

Claims (12)

First storage means for storing externally input image data; Compression restoring means for encoding, decoding and compressing and restoring the input image data by performing motion estimation operation, motion compensation operation, and data conversion on the image data input from the first storage means; Second storage means for storing the resultant data according to the compression mode determined by the motion estimation operation of the compression restoring means and the quantized and dequantized data by the compression restoring means for compressing the input image data; Third storage means for storing variable length decoded data and inverse discrete cosine transformed data in the compression restoring means for restoring the input image data; And Control means for controlling the read and write of the first to third storage means and controlling the compression restore operation of the compression restore means by time division, wherein the first to third storage means and the compression restore means Image compression restoring apparatus, characterized in that coupled to the coprocessor form around the control means. The method according to claim 1, And the first storage means comprises a dynamic RAM. The method according to claim 1, The compression restoring means includes a motion estimation calculation block for calculating a motion vector between a current frame and a previous frame, a motion compensation calculation block for performing motion compensation with a previous frame, a discrete cosine and inverse discrete cosine, and a quantization unit. And a data conversion block including a quantization unit, a variable length encoder, and a variable length decoder to perform a compression or reconstruction process. The method according to claim 3, The second storage means comprises a dual port memory having two read ports and two write ports, wherein the motion estimation arithmetic block, the discrete cosine transform unit of the data transform block, the variable length encoder of the data transform block, and the motion An image compression restoring apparatus, which is commonly used by a compensation operation block and the control means. The method according to claim 1, The third storage means comprises a dual port memory having two read ports and two write ports, an inverse discrete cosine of the motion compensation operation block and the data conversion block, a variable length decoder of the data conversion block, and the control. An image compression restoring apparatus which is commonly used by means. Determining whether to compress or restore the input image data; Performing step compression of motion estimation, motion compensation, discrete cosine transform, quantization, and variable length coding when it is determined that the result of the determination step is compression; Performing step reconstruction of variable length decoding, inverse quantization, inverse discrete cosine transform, and motion compensation when it is determined that the result of the determination step is restored; When performing the compression process, it is checked whether the restoration is useful for each compression process, and if the restoration is useful as a result of the check, the next compression process is suspended, and the useful restoration process is first performed. Performing the next compression process that is suspended; And When performing the restoration process, it is checked whether compression is useful for every restoration process. If the compression is useful as a result of the check, the next restoration process is suspended, and the useful compression process is first performed. And restoring the suspended next restoring process, and compressing and restoring the image data by time division. The method according to claim 6, The compression process further comprises the step of determining whether the external image output device can process the restored signal. The method according to claim 6, The restoring process may further include determining whether a signal input from an external image input apparatus can be compressed. The method according to claim 6, And the motion estimation, motion compensation, discrete cosine transform and inverse discrete cosine transform are performed in parallel. The method according to claim 6, And the compression process and the restoration process are each pipelined to speed up the processing of the data. The method of claim 10, wherein the pipeline of the compression process Performing variable length coding on the motion estimation operation and the picture header; Performing motion compensation and discrete cosine transform operations; Performing variable length coding and inverse discrete cosine transform on the coefficients of each macro block; And a method for performing motion compensation and variable length coding. The method of claim 10, wherein the pipeline of the restoration process Finding the initial start code of the frame, Restoring the picture header; Restoring the mode of the macroblock; Restoring information about each block in the macro block; Dealing with the case where the current macroblock is intertype and not coded, Processing a case in which the current macroblock is an intertype and the coded block pattern is "0"; Processing a case in which the current macroblock is intra or inter and the code block pattern is "1"; And determining whether or not the external display device is usable.