JPH06334987A - Picture encoding device - Google Patents

Abstract

PURPOSE:To reduce the degradation of the picture quality due to transmission error without degrading the encoding efficiency in the encoding device which compresses moving picture data by an orthogonal transformation and variable length encoding. CONSTITUTION:The device is provided with a fixed length encoding means 108 which encodes the orthogonally transformed coefficients of picture data subjected to an orthogonal transformation to a fixed length code consisting of the absolute coordinate values and the level, a variable length encoding means 107 which encodes the combination of the relative coordinate values and the level from the coefficients of the nonzero level variable-length encoded in the preceding step, and the level to a two-dimensional variable length code, an absolute coordinate counting means 105, a relative coordinate counting means 104, a level comparing means 106 which makes the variable length encoding means active in the case that the level is equal to or lower than a threshold and is nonzero, and a frame assembling means 109 which assembles fixed length encoded data and variable length encoded data so that they can be discriminated from each other.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image coding apparatus which employs image coding resistant to transmission path errors.

[0002]

2. Description of the Related Art Conventionally, in video conferencing and video telephone signal encoding, hybrid encoding is generally used which combines interframe prediction utilizing intraframe correlation and intraframe orthogonal transformation in the frame direction. Target. 1
A television image composed of 30 images (frames) per second has a large correlation in the time axis direction, and it is possible to use the pixel at the same position in the screen one frame before in the prediction by utilizing the interframe correlation. For example, the most ideal prediction can be made when the screen is stationary. However, in the inter-frame coding, when there is a motion in the screen, the correlation between the frames is low, and the correlation between adjacent pixels in the field is low. On the other hand, each pixel of the image signal for each frame has a small level change with an adjacent pixel and has a strong correlation. It is said that the autocorrelation function can be approximated by a negative exponential function. At this time, the power spectrum density, which is the Fourier transform of the autocorrelation function, has a maximum at zero frequency component (DC) and has the property of monotonically decreasing as the frequency component increases. The most well-known orthogonal transform to the frequency domain is the Fourier transform, but since it involves complex number operations and complicates the structure, two-dimensional DCT (discrete cosine) is used as an alternative orthogonal transform in image coding. (Conversion) is generally used. The transform coefficient decomposed into frequency components by DCT is quantized from level ± 1 which is a transform coefficient (insignificant coefficient) which is not coded and level ± 1 which is a significant coefficient having a discrete quantized representative value to level ± K. After that, the image data is compressed by the run length code that encodes the continuity of the insignificant coefficients and the Huffman encoding that assigns the variable length code according to the occurrence probability of the level of the significant coefficient.

For example, CCITT Recommendation H.264. In H.261, inter-frame prediction is applied to an image with little motion,
The following coding is performed on the prediction error between frames. In addition, inter-frame prediction is not applied to images with large motion, and the following encoding is directly performed on frame pixels. FIG. 261 illustrates encoding of image data in H.261. In FIG. 9, 101 is an orthogonal transform unit that performs two-dimensional DCT, 102 is a quantization unit,
103 is a scanning unit, and 904 is a run length level 2
It is a dimensional Huffman coding unit. A prediction error between frames or frame pixels is divided into blocks of 8 pixels × 8 lines, and the orthogonal transform unit 101 performs two-dimensional DCT on each block. The pixels of each block are converted into frequency components by DCT. The obtained transform coefficient is used by the quantization unit 102.
Each transform coefficient quantized by is represented by the level of the significant coefficient which is an integer from level 0 of the insignificant coefficient to level ± 127. Example of quantized transform coefficient
Shown in. In FIG. 2, the upper left is a DC component, and the other is an AC component, and the spatial frequency increases toward the lower right. Since image data has a high correlation between adjacent pixels, power is generally concentrated and distributed in a low frequency region by orthogonal transformation. The conversion unit performs a zigzag scan as indicated by an arrow in FIG. 3 by the scan unit 103 to rearrange the low-frequency components to the high-frequency components in a one-dimensional array. The rearranged transform coefficients are encoded by the run-length two-dimensional Huffman encoder 904. In the run-length level two-dimensional Huffman encoding unit 904, among the combinations of runs (RUN) indicating the number of consecutive insignificant coefficients having a level of 0 and subsequent significant coefficient levels, a combination having a high occurrence frequency is illustrated. Variable length coding is performed as shown in Table 10. Other combinations of RUN and level are encoded into a 20-bit fixed-length code consisting of a 6-bit ESCAPE code for identifying a variable-length code, a 6-bit RUN and an 8-bit level. For example, when 1 appears after 2 consecutive 0s, the code is "01010" and the code length is 5 because the RUN is 2 and the level is 1.
By orthogonally transforming the frame pixels, power is generally concentrated and distributed in the low frequency region. Therefore, in most areas except the low frequency area, the coefficient with a small level or the level is 0.
The frequency of the insignificant coefficient of is increased. A zigzag scan and run-length coding of this code can reduce the number of codes to be transmitted, and a Huffman coding can reduce the code length. H. 261 compresses image data by taking advantage of such a property.

[0004]

On the other hand, when the prediction error signal between frames is orthogonally transformed, the power is rarely concentrated in the low frequency region, and the significant coefficient is also distributed in the high frequency region. This is because a high frequency component is generated by calculating the difference between frames. The correlation coefficient in the space is about 0.95 in the image of one frame, but it decreases to about 0.5 to 0.6 in the inter-frame difference signal. That is, when there is no motion between frames, invalid coefficients frequently appear, and when the amount of motion is small, significant coefficients of a small level frequently occur. The run-length Huffman coding of H.261 is effective. However, when there is a large amount of movement between frames, a significant level of significant coefficients frequently appears and is distributed over a wide area. Most of them are coded into the fixed-length code of H.261. In this case, the advantage of run-length Slevel Huffman coding for reducing the code amount is not exhibited at all. Next, consider a case where the quality of the transmission path is poor, such as a wireless transmission path, and an error occurs in a part of the transmitted image compression data. As a property of run-length encoding, the position of the coefficient on the array is represented by the relative position from the coefficient whose previous level is not 0. Therefore, if one coefficient is incorrect, the positions of all the coefficients after that are incorrect. . This state is shown in FIG. In FIG. 11, (a) is a part of the orthogonal transform coefficient after quantization. This is sequentially scanned and the result of run-length level Huffman coding is shown in (b).
Shown in. Significant coefficients with a large level are coded into a 6-bit escape code, a 6-bit RUN, and an 8-bit level with a total of 20-bit fixed length code, and the first 2 are RU.
It can be converted into a 7-bit variable length code in which N is 2 and the level is 2, but for convenience sake, the notation is as shown in the figure. In (b), a transmission error occurs in the second fixed length data,
The case where N is wrong from 0 to 5 is shown in (c). When the data string of (c) is decoded, the orthogonal transformation coefficient of (d) is obtained, but this is all different from (a) except the first coefficient. If the RUN of one coefficient is wrong, the coefficient transmitted without error thereafter shifts to a different position. Further, as a property of the variable-length code, when an error occurs, the code can be changed into a code having a different code length, so that the break of the code that follows it cannot be seen (e). As a result, as shown in (f), the coefficient after that is decoded into an incorrect coefficient including the fixed length data. Since all the prediction error signals are error signals with respect to the already transmitted previous frame, the significance of a significant coefficient having a small level is low. On the other hand, the significance coefficient, which is a prediction error and has a large level, is important because it is information that conveys a large difference from the previous frame. H. 261
However, since they are all treated in the same manner, a 1-bit transmission error spreads over the entire block, and the image is greatly deteriorated. Further, since the deteriorated image is used as the original data for inter-frame prediction, the deterioration of the image remains until the next frame pixel data is sent. In other words, the conventional method was able to efficiently transmit the data of blocks that are relatively insignificant because the image change between frames is relatively small, but compress the data of important blocks where the image change between frames is large in compression efficiency. Without being affected by the reduction, it was affected only by the shortcomings of run-length coding, and was coded into a code that was very vulnerable to a transmission error despite being important data.

The present invention has been made in view of the above circumstances, and limits the influence of transmission errors to erroneous data without impairing the compression efficiency of important block data in which image changes between frames are large. It is an object of the present invention to provide an image encoding device that does not spread within a block.

[0006]

According to a first aspect of the present invention, a fixed-length code for encoding an orthogonal transform coefficient of image data converted by an orthogonal transform unit into a fixed-length code composed of absolute coordinate values of coefficients and levels. The encoding means and the combination of the relative coordinate value and the level from the coefficient of which the variable length encoding is performed before the nonzero level are 2
Dimension variable length coding means for variable length coding, absolute coordinate counting means for counting absolute coordinate values of coefficients to be coded and shown in fixed length coding means, and the level previously variable length coded is non-zero. The relative coordinate counting means for counting the relative coordinates from the coefficient and showing the variable length coding means and the given threshold value and the coefficient level are compared, and the absolute value of the coefficient level exceeds the threshold value. In the case where the fixed length coding means is activated, and when the absolute value of the coefficient level is equal to or less than the threshold value and non-zero, the variable length coding means is activated. The configuration includes a frame assembling unit that realizes a frame assembling method for assembling long coded data into an identifiable transmission frame.

According to a second aspect of the present invention, in the frame assembling means according to the first aspect of the present invention, the content of the data is fixed length data or variable length data in a data area of n times the data length m bits of the fixed length encoded data. M × n + L bits to which data identification bits L bits for identifying whether or not are used as a basic frame constituent unit, and the data identification bits are arbitrarily designated according to the frequency of occurrence of fixed length encoded data and variable length encoded data. By arranging a plurality of basic frame constituent units, a fixed length can be flexibly assembled into a transmission frame that can be distinguished from variable length data depending on the frequency of occurrence.

According to a third aspect of the present invention, the frame assembling means according to the first aspect of the present invention stores n pieces of fixed-length data having a data length of m bits and indicates a full state at the time of writing m × n bits. The buffer, the m × n bit shift register that shifts at the bit generation timing of variable-length data, and shows a full state at the time of m × n bit shift, the k-stage FIFO buffer having a width of m × n + L bits, and the shift When the register is full, the parallel data output of the m × n bit shift register and the L-bit data identification bit indicating the variable-length coded data are written to the FIFO buffer, and when the fixed-length data buffer is full Parallel data output with an m × n bit data buffer and data identification bits indicating fixed length encoded data are stored in the FIFO buffer. It is obtained by a write to the configuration.

According to a fourth aspect of the invention, when the frame assembling means according to the third aspect of the invention writes the variable length data indicating the end of the divided image area to the shift register, the data of the next image area is written. Is configured so that the shift register is brought into a full state without writing.

[0010]

In the invention of claim 1, the level comparison means performs orthogonal transformation by the orthogonal transformation section, judges whether the level of the transform coefficient quantized by the quantizing section is an important level or an unimportant level, and if the level is an important level. Uses the fixed-length coding means to code the absolute coordinates and level indicated by the absolute-coordinate coefficient means into a fixed-length code. The coefficient of the non-important level is encoded into a variable length code determined by the combination of the relative coordinate and the level from the coefficient of the preceding non-important level indicated by the relative coordinate coefficient means.

According to the second aspect of the present invention, whether to store the fixed length data or the variable length data in the basic frame constituent unit is arbitrary according to the occurrence frequency of both, and the length of the basic frame constituent unit. The length is the same for both fixed-length coded data and variable-length coded data, and because the length is always constant, even if the break in the code due to an error in the variable-length coded data is unknown. The beginning of fixed-length coded data can be found immediately. Further, since the variable length coded data and the fixed length coded data can be identified by the data identification bit, it is not necessary to add an escape code for distinguishing the fixed length coded data from the variable length coded data.

In the invention of claim 3, the data buffer efficiently stores fixed-length data, and the shift register easily stores variable-length data. If the data buffer or the shift register, whichever is full, is stored in the FIFO buffer, the frame according to claim 2 can be efficiently created with a simple configuration.

In the invention of claim 4, the shift register is brought into a full state without writing the next image data, whereby the start of the data of the next image region can be made from the start of the basic frame constituent unit, The influence of the error of the variable length data is prevented from spreading to the next image area.

[0014]

Embodiment 1 Embodiment 1 of the present invention will be described below with reference to the drawings.

FIG. 1 is a block diagram of an image coding apparatus according to an embodiment of the present invention. In FIG. 1, 101 is an orthogonal transformation unit,
Numeral 102 is a quantizer, and 103 is a scan unit, which is composed of relative coordinate counting means 104 and absolute coordinate counting means 105. 106 is a level comparison means, 107 is a variable length coding means, 108 is a fixed length coding means, and 109 is a frame assembling means.

The operation of the image coding apparatus configured as described above will be described with reference to FIGS. 2 to 6. The prediction error between frames is divided into blocks of 8 pixels × 8 lines, and the orthogonal transformation unit 101 performs two-dimensional DCT on each block. The pixels of each block are converted into frequency components by DCT. The obtained transform coefficient is used by the quantization unit 102.
Each transform coefficient quantized by is represented by the level of the significant coefficient which is an integer from level 0 of the insignificant coefficient to level ± 127. Example of quantized transform coefficient
Shown in. In FIG. 2, the upper left is a DC component, and the other is an AC component, and the spatial frequency increases toward the lower right. The conversion unit performs a zigzag scan as indicated by an arrow in FIG. 3 by the scan unit 103 to rearrange the low-frequency components to the high-frequency components in a one-dimensional array. The scanning unit 103 is composed of an absolute coordinate counting means 105 and a relative coordinate counting means 104. Among these, the absolute coordinate counting means 105 counts absolute coordinates from 0 to 63 in the zigzag scan of the transform coefficient and performs fixed length coding. Present to the means. The level comparing means 106 compares the absolute value of the level of the transform coefficient quantized by the quantizer 102 with a preset threshold value, and when the absolute value of the level exceeds the threshold value, The fixed length coding means 108 is activated. When the absolute value of the level exceeds 0 and is less than or equal to the threshold value, the variable length coding means is activated and the relative coordinate counting means 104 is notified of that fact. When the level is 0, neither the variable length coding means 107 nor the fixed length coding means 108 is activated. The relative coordinate counting means 104 counts the relative coordinates of the coefficient that previously activated the variable length coding means 107 and the coefficient that next activates the variable length coding means 107, and presents it to the variable length coding means 107. .
The variable-length coding means 107 codes a combination of relative coordinates (RUN) and levels as shown in FIG. Here, the relative coordinate is given by the relative coordinate counting means 104, and the level is given by the quantizer 102. FIG. 3 is an example when the threshold value is 3, and EOB
Indicates the end of the variable length coded data. As shown in FIG. 5, the fixed length encoding means 108 encodes fixed length data of a total of 14 bits of level 8 bits and absolute coordinates 6 bits. The frame assembling unit 109 assembles fixed-length encoded data and variable-length encoded data into identifiable transmission frames. Although there are various possible configurations of the transmission frame, the structure is not limited here. For example, a fixed-length area and a variable-length area unique to the first half and the second half of the frame have fixed-length data and variable-length data, respectively. Alternatively, each area may have a variable length and a flag indicating the size of the area may be included in the frame configuration. Further, the frame assembling method described in claim 2 may be used. In FIG. 6, (a) is a part of the orthogonal transform coefficient after quantization. The result of scanning and coding this with the image coding apparatus of this system is shown in (b). The significant coefficient whose absolute value of level exceeds 4 is 14 in total of 8-bit level and 6-bit absolute coordinate.
It is encoded into a fixed length code of bits and transmitted in the fixed length area. Significant coefficients whose absolute value of level exceeds 0 and is 4 or less are coded into a two-dimensional variable length code of RUN and level of relative coordinates and transmitted in the variable length area. However, here, for convenience, 1
The level and absolute coordinates or RUN are displayed in 0-base. A case where a transmission error occurs in the second fixed-length data in (b) and the absolute coordinates are erroneous from 1 to 5 is shown in (c). When the data sequence of (c) is decoded, the orthogonal transform coefficient of (d) is obtained. The second and sixth coefficients are wrong, but all other coefficients are the same as in (a). .
Since the coordinates of the fixed length code are absolute coordinates, even if an error occurs in one coefficient, the coefficient and the coefficient resulting from the error cannot be decoded, but all other coefficients can be decoded. In addition, H. In H.261, it is necessary to add a 6-bit escape code to each fixed-length data in order to distinguish between the variable-length data and the fixed-length data, and the fixed-length code has a code length of 20 bits. As far as data is concerned, the fixed-length code has a code length of 14 bits, which contributes to improvement of compression efficiency. Further, if limited to the inter-frame difference signal in which the spatial correlation decreases, the frequency of occurrence of a code whose level is small together with RUN decreases, so that the H.264 standard is reduced. The Huffman code table used by 261 is not very effective. That is, the coding efficiency of this method is H.264 as a whole. The 261 is comparable. Next, when an error occurs in the variable-length code, the code length can be changed to a different code, and the break in the code that follows it cannot be seen (e) because the variable-length code and the fixed-length code are transmitted in a distinguishable manner. All variable length data after the incorrect coefficient is wrong, but all fixed length data can be correctly decoded (d). Here, the variable length data is data in which the absolute value of the level is equal to or less than the threshold value, and even if all erroneous data is set as the insignificant coefficient (level 0), the influence on the image quality is small.

As described above, according to this embodiment, important data in the inter-frame difference error signal whose level absolute value exceeds the threshold value is coded so that the error does not propagate without impairing the coding efficiency. It is possible to change.

(Second Embodiment) A second embodiment of the present invention will be described below with reference to the drawings.

FIG. 7 is a configuration diagram of a transmission frame by a frame assembling method for assembling fixed length coded data and variable length coded data into identifiable transmission frames according to an embodiment of the present invention. In FIG. 7, 701 is a data identification bit, 702 is a fixed length data part, 703 is a variable length data part, and 704 is a basic frame constituent unit. The data identification bit 701 may be 1 bit, for example, 1 when the data part is fixed length data and 0 when it is variable length data. The fixed-length data unit 702 has a length of m × n bits capable of storing n pieces of fixed-length data to be handled, assuming that the fixed-length data to be handled has an m-bit length. For example, the fixed length code of the first embodiment is 14 bits, which is 14 × n bits. The basic frame structural unit 704 in this case has a length of 1 + 14 × n bits. The variable length data section 703 also has the same length as the fixed length data section 702, and the variable length data is packed in order from the front. If the variable length data section is just full, there is no problem. However, if the variable length data section is full, it is stored up to the middle.
The rest is stored in the next variable length data section. If the data ends before the variable-length data section is full, it is detected by EOB, which is a variable-length code.

Next, FIG. 8 shows an embodiment of a frame assembling means for realizing the above method. In FIG. 8, reference numeral 801 is a data buffer, 802 is a shift register, 803 is a selector, and 804 is a FIFO buffer. The data buffer 801 has a length of m × n bits, and indicates a full state when n pieces of fixed-length data of m bits are written. The shift register 802 is an m × n-bit shift register, and
It indicates a full state when it is shifted by × n bits. The selector 803 is a 2-input, 1-output selector of m × n + L bit width, and when the data buffer 801 indicates a full state, an L × bit data identification indicating a parallel output of m × n bit width of the data buffer 801 and fixed length data. Bits are read in parallel and written to the FIFO buffer 804. When the shift register 802 indicates the full state, the shift register 80
2 m × n bit width parallel output and L bit data identification bit indicating variable length data are read in parallel and FIF
Write to O buffer 804. When data is sequentially read from the FIFO buffer, a frame is created by the frame assembling method of the present invention. When EOB, which is variable-length data indicating the end of an image block, is written in the shift register 802, the data of the next image block is shifted to the full state without reading the data of the next image block, and the data of the next image block is updated. It can be stored from the beginning of the basic frame constituent unit.

As described above, according to this embodiment, the fixed length coded data and the variable length coded data can be efficiently identified by flexibly responding to the frequency of occurrence of the fixed length coded data and the variable length coded data. Even if the variable length coded data becomes erroneous and the break of the data becomes unclear, the break of the next basic frame structural unit is clear and the fixed length coded data can be decoded without any problem .

[0022]

As described above, according to the present invention, data having a large inter-frame difference error and having a large influence on the image quality can be encoded as absolute coordinate fixed length data independently of other variable length data. It is possible to reduce image quality deterioration due to a transmission error without impairing the coding efficiency.

[Brief description of drawings]

FIG. 1 is a configuration diagram of an image encoding device according to a first embodiment of the present invention.

FIG. 2 is a diagram showing an example of transform coefficients quantized by a quantizer according to the first embodiment of the present invention or a conventional technique.

FIG. 3 is a diagram showing an example of a zigzag scan order of transform coefficients according to the first embodiment of the present invention or a conventional technique.

FIG. 4 is a diagram showing a coding example of a variable-length coding unit according to the first embodiment of the present invention.

FIG. 5 is a diagram showing a coding example of a fixed length coding means in the first embodiment of the present invention.

FIG. 6 is an operation explanatory diagram according to the first embodiment of the present invention.

FIG. 7 is a frame configuration diagram according to the second embodiment of the present invention.

FIG. 8 is a configuration diagram of a frame assembling unit according to a second embodiment of the present invention.

FIG. 9 is a configuration diagram of an image encoding device according to a conventional technique.

FIG. 10 is a diagram showing a variable length coding table in the conventional technique.

FIG. 11 is an operation explanatory diagram in the conventional technique.

[Explanation of symbols]

 101 Orthogonal Transform Unit 103 Scan Unit 104 Relative Coordinate Counting Means 105 Absolute Coordinate Counting Means 106 Level Comparison Means 107 Variable Length Coding Means 108 Fixed Length Coding Means 109 Frame Assembling Means 701 Data Identification Bits 702 Fixed Length Data Parts 703 Variable Length Data 704 Basic frame configuration unit 801 Data buffer 802 Shift register 803 Selector 804 FIFO buffer

Claims (4)

[Claims] 1. Fixed-length coding means for coding an orthogonal transform coefficient of image data converted by an orthogonal transform unit into a fixed-length code composed of absolute coordinate values of coefficients and levels; Variable-length coding means for two-dimensionally variable-length coding a combination of relative coordinate value and level from a coded non-zero coefficient, and fixed-length coding means for counting absolute coordinate values of the coded coefficient. Absolute coordinate counting means shown, relative coordinate counting means for indicating relative coordinates from a variable-length-encoded coefficient whose level is non-zero immediately before and shown to the variable-length encoding means, given threshold value and coefficient , The fixed-length coding means is activated when the absolute value of the coefficient level exceeds the threshold value, and variable when the absolute value of the coefficient level is less than the threshold value and non-zero. Level for activating long coding means Comparison means, the image coding apparatus that includes a frame assembly means for implementing the frame assembly method of assembling the fixed length encoded data and variable-length coded data capable of identifying transmission frame. 2. A frame assembling means has a data identification bit L bit for identifying whether the content of the data is fixed length data or variable length data in a data area of n times the data length m bits of the fixed length encoded data. By adding m × n + L bits as a basic frame constituent unit, by arranging a plurality of basic frame constituent units by arbitrarily designating the data identification bits according to the frequency of occurrence of fixed length encoded data and variable length encoded data, 2. The image coding apparatus according to claim 1, wherein a fixed length is assembled into a transmission frame that can be discriminated from variable length data flexibly according to the frequency of occurrence. 3. A frame assembling means stores n pieces of fixed-length data having a data length of m bits, and shifts at a data buffer showing a full state at the time of writing m × n bits and at a bit generation timing of variable-length data. Then, an m × n bit shift register showing a full state at the time of m × n bit shift and a k-stage FIFO having a width of m × n + L bits
When the buffer and the shift register are in the full state, the parallel data output of the m × n bit shift register and the L-bit data identification bit indicating the variable length coded data are added to the FIF.
Writing into the O buffer, and writing the parallel data output of the m × n bit data buffer and the data identification bit indicating the fixed length encoded data into the FIFO buffer when the data buffer for the fixed length data is in the full state. The image coding apparatus according to claim 1. 4. When the frame assembling means writes the variable length data indicating the end of the divided image area to the shift register, the frame register means sets the shift register to the full state without writing the data of the next image area. The image coding device according to claim 3, wherein