JP2011035682A - Method and device for coding/decoding data - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To solve the problem that decoding becomes difficult since even a code of what data for coding is lost is not clear anymore on the decoding side resulting from buffer overflow. <P>SOLUTION: In this data coding device, a Golomb-Rice coding means 134 performs variable length coding of a conversion coefficient which is the data for coding, and stores generated codes in a coding FIFO memory 123. A code output abbreviation means 136 counts one or more numbers of continuation of codes which can not be stored in the coding FIFO memory 123 due to a lack of free space among the generated codes as a skip number. A coefficient skip number coding means 137 codes the skip number counted by the coding output abbreviation means 136 to be stored in the coding FIFO memory 123 when vacancy is generated in the coding FIFO memory 123. <P>COPYRIGHT: (C)2011,JPO&INPIT

Description

  The present invention relates to a method and apparatus for encoding data, and more particularly to a method and apparatus for variable-length encoding data.

  One of the methods for variable-length encoding fixed-length encoding target data is Golomb-Rice coding. The Golomb-Rice code outputs the data to be encoded as a “single decimal code” + “k-bit fixed length code (k is an estimated value of the effective number of digits)”. The closer the value is to 0, the shorter the code length. Therefore, it is used as simple encoding such as prediction residual.

  In Golombrice coding, a binary number to be encoded is divided into lower k bits and upper remaining bits, and the upper remaining bits are replaced with a hexadecimal code. For example, when the signal 00010011 (binary representation) is encoded with k = 3, the lower 5 bits are output as they are after the upper 5 bits (00010) excluding the lower 3 bits are unicode encoded. When using a code in which "1" bits are added to the number of "0" bits as many as the number represented by the bit string as the unicode code, the upper 5 bits in the above example are 2. "0" is output and terminated with "1". Together with the lower 3 bits, “001011” is finally output as a code.

  On the decoding side, using the same k value as that used for encoding, the binary number to be decoded is divided into lower k bits and upper remaining bits, and the upper remaining bits are divided into those before the unicode code. Decrypt into state. For example, in the decoding of the code “001011” in the above example encoded as k = 3, first, the numerical value of 2 from the unicode code “001” is decoded and output as the upper 5 bits “00010”. Subsequent “011” is output as the lower 3 bits. As a result, a signal 00010011 that completely matches the original signal is obtained. As described above, the Golomb-Rice code is a kind of lossless compression code.

  The Golomb-Rice code has the shortest code length when the significant digit of the signal value to be encoded matches the k value. In the example of the signal 00010011 described above, the number of significant digits is 5. Therefore, when k = 5, the output code is 110011 and the code length is the shortest. Therefore, an adaptive Golomb-Rice code that predicts an optimal k value has been devised.

  An example of the adaptive Golombrice code is described in Patent Document 1. The adaptive Golomb-Rice coding method described in Patent Document 1 is based on the statistics of the signal value d (the prediction residual is the target of Golomb-Rice coding in Patent Document 1), which is the target of Golomb-Rice coding. Is to predict. Specifically, the absolute value of the signal value d that has appeared so far is accumulated, and the average number of significant digits k is estimated from the sum a and the accumulated number n. When the cumulative number reaches a predetermined upper limit value (Reset), a rescale process is performed to halve the cumulative value and the cumulative number. On the other hand, in Golomurice decoding, in order to obtain the same k value as that on the encoding side, the absolute value of the signal value d that has been Golomurice decoded so far is accumulated, and from the sum a and the accumulated number n, it was used in Golomurice coding. The same number of significant digits k is estimated. Furthermore, when the cumulative number reaches a predetermined upper limit (Reset), a rescale process is performed to halve the cumulative value and the cumulative number.

  In the Golomb-Rice code, when the number of significant digits k of the signal value to be encoded is erroneously estimated, the code length may be much longer than the number of bits of the signal value to be encoded. For example, if a 16-bit signal value of 001011001101101 is encoded with k = 5, the upper 11 bits excluding the lower 5 bits are 179, so 179-bit “0” is output and then terminated with “1”. Is done. Therefore, when the lower 5 bits are combined, the code length is 185 bits.

  In order to prevent such a long Golomb-Rice code from being generated, when the code length of a unicode code exceeds the preset upper limit length max, a sequence of bits "0" equal to the upper limit length max is encoded. Encoding is performed by another method of adding to the data to be encoded. For example, when the upper limit length max is 32, if the above-mentioned 001011001101101 is encoded with k = 5, the upper 11 bits excluding the lower 5 bits will be more than 32 and 179, so 32-bit "0" will be consecutive Is encoded by a method of adding to the head of 0001011001101101. As a result, the code amount is 32 + 16 = 48 bits, which is longer than the 16-bit length of the original data to be encoded, but shorter than 185 bits in the case of Golombrice encoding. Such an encoding method is called an encoding method in a raw mode.

US Pat. No. 5,763,474 WO2005 / 081539

  As described above, in Golombrice coding, a code having a code length longer than the data to be encoded may be temporarily generated. For this reason, in an encoding device that encodes and outputs continuously encoded data to be encoded in real time, if the capacity of the buffer that temporarily accumulates until the generated code is output is small, it is buffered with a code that waits for output. Becomes full and a part of codes of data to be encoded is lost due to overflow of the buffer. The data to be encoded corresponding to the code lost due to the buffer overflow cannot be decoded on the decoding side, and cannot be referred to for estimation of the decoding parameter k used in the subsequent decoding. To make matters worse, since it is not known which encoding target data has lost its code, subsequent decoding becomes difficult. Ensuring a buffer with sufficient capacity prevents loss of code due to buffer overflow, but it is wasteful in terms of cost to secure a large capacity buffer due to buffer overflow that rarely occurs. .

  Accordingly, an object of the present invention is to provide a data encoding / decoding method and a data encoding / decoding method that solves the above-described problem that even the code of which encoding target data is lost due to buffer overflow is not understood on the decoding side and decoding becomes difficult. To provide an apparatus.

  The data encoding apparatus according to the present invention includes a buffer for temporarily storing generated codes, variable length encoding means for variable length encoding of encoding target data, and codes generated by the variable length encoding means. Code output omitting means for counting, as a skip number, one or more consecutive codes that could not be stored in the buffer due to a lack of free space in the buffer, and code output omitting means when the buffer is empty And a skip number encoding means for encoding the skip number counted in step (b) and storing it in the buffer.

  The data decoding apparatus of the present invention also includes a code including a code obtained by variable-length coding the data to be coded and a code obtained by coding the number of skips indicating how many continuous data to be coded are not coded. A decoding device having a sequence as input, decoding means for decoding the encoding target data and the skip number from the code string, and encoding target data in advance for the number of skips decoded by the decoding means Skip means for handling the data encoded as a predetermined value.

  In the data encoding method of the present invention, a code generated by variable-length encoding the encoding target data is stored in a buffer, and one or more consecutive numbers of codes that could not be stored in the buffer due to insufficient free space in the buffer Is counted as a skip number, and when the buffer is empty, the counted skip number is encoded and stored in the buffer.

  Also, the data decoding method of the present invention is a code including a code obtained by variable-length coding the data to be coded and a code obtained by coding the number of skips indicating how many continuous data to be coded are not coded. A sequence is input, the encoding target data and the skip count are decoded from the code sequence, and the encoding target data is encoded with a predetermined value by the decoded skip count. The subsequent decoding is continued.

  According to the present invention, it is possible to eliminate a situation in which decoding becomes difficult because the decoding side does not even know which encoding target data code has been lost due to buffer overflow.

1 is a block diagram of a data encoding apparatus according to a first embodiment of the present invention. It is explanatory drawing of wavelet transformation of 3 layers. It is a flowchart which shows operation | movement of the data coding apparatus which concerns on the 1st Embodiment of this invention. It is a flowchart which shows operation | movement of the data coding apparatus which concerns on the 1st Embodiment of this invention. It is a block diagram of the data decoding apparatus which concerns on the 2nd Embodiment of this invention. It is a flowchart which shows operation | movement of the data decoding apparatus which concerns on the 2nd Embodiment of this invention. It is a block diagram of the data coding apparatus which concerns on the 3rd Embodiment of this invention. It is a flowchart which shows operation | movement of the data coding apparatus based on the 3rd Embodiment of this invention. It is a flowchart which shows operation | movement of the data coding apparatus based on the 3rd Embodiment of this invention. It is a block diagram of the data decoding apparatus which concerns on the 4th Embodiment of this invention. It is a flowchart which shows operation | movement of the data decoding apparatus which concerns on the 4th Embodiment of this invention.

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

[First embodiment]
Referring to FIG. 1, in the first embodiment of the present invention, the present invention is applied to a wavelet transform coding device, and an image input device 11, a data storage device 12, and data operated by program control. It comprises a processing device 13 and a code output device 14.

  The image input device 11 is a device for inputting an image as a two-dimensional signal, and is composed of, for example, a camera or a communication device.

  The data storage device 12 includes an image memory 121, a conversion coefficient memory 122, a code FIFO memory 123, a code FIFO free capacity memory 124, and a coefficient skip number memory 125.

  The image memory 121 stores an image input from the image input device 11.

  The transform coefficient memory 122 stores wavelet transform coefficients obtained by performing wavelet transform on the input image stored in the image memory 121.

  The code FIFO memory 123 is a buffer that stores, in a first-in first-out manner, a code obtained by performing Golomb-Rice coding on the wavelet transform coefficient stored in the transform coefficient memory 122.

  The code FIFO free capacity memory 124 stores the free capacity of the code FIFO memory 123.

  The coefficient skip number memory 125 stores the number of wavelet transform coefficients whose code has not been output to the code FIFO memory 123.

  The data processing device 13 includes an image converting unit 131, a coefficient reference value changing unit 132, an encoding parameter determining unit 133, a Golomb-Rice encoding unit 134, a code output determining unit 135, a code output omitting unit 136, a coefficient Skip number encoding means 137 and code output means 138 are provided.

  The image conversion unit 131 generates wavelet transform coefficients by performing wavelet transform on the input image stored in the image memory 121, and stores the wavelet transform coefficients in the transform coefficient memory 122. The wavelet transform is a kind of subband coding, and performs band division of N layers (N is an integer of 1 or more) by repeatedly performing subband division for dividing the band in the horizontal and vertical directions on the low frequency side. . Such band division is called octave division. When dividing up to three layers, ten subbands as shown in FIG. 2 are obtained.

  In FIG. 2, F0 to F3 are subbands in the lowest layer, F4 to F6 are subbands in the upper layer, and F7 to F9 are subbands in the highest layer. F1, F4, and F7 are subbands that have a low-pass filter in the horizontal direction and a high-pass filter in the vertical direction. F2, F5, and F8 are subbands that have a high-pass filter in the horizontal direction and a low-pass filter in the vertical direction. F3, F6, and F9 are subbands to which a high-pass filter is applied in both the horizontal and vertical directions. Further, F0 is referred to as the lowest frequency subband, and F1 to F9 other than F0 are referred to as the high frequency subband.

  Each subband F0 to F9 includes a wavelet transform coefficient. In this specification, wavelet transform coefficients included in subbands such as F1, F4, and F7 that have been subjected to a low-pass filter in the horizontal direction and a high-pass filter in the vertical direction are referred to as LH. Further, wavelet transform coefficients included in subbands such as F2, F5, and F8 that have been subjected to a high-pass filter in the horizontal direction and a low-pass filter in the vertical direction are referred to as HL. A wavelet transform coefficient included in a subband to which a high-pass filter is applied in both the horizontal direction and the vertical direction, such as F3, F6, and F9, is referred to as HH. The wavelet transform coefficient included in the lowest band subband F0 is referred to as LL. Furthermore, for example, LH (i, j), HL (i, j), and HH (i, j) are included in the wavelet transform coefficients LH, HL, and HH of the same spatial coordinates in a plurality of high frequency subbands in the same layer. In this way, the same coordinates are added after them. It is assumed that the horizontal direction is x, the vertical direction is y, and the coordinates are represented by (x, y).

  When the wavelet transform as described above is performed, the power is biased toward the wavelet transform coefficient LL included in the lowest band subband, and the wavelet transform coefficients LH, HL, and HH included in the high band subband become values close to zero. For this reason, it is possible to compress the amount of information by encoding the wavelet transform coefficients LH, HL, and HH included in the high frequency sub-band using a variable length code in which the code length is shorter as the value is closer to 0. become. Further, if the wavelet transform coefficients LH, HL, and HH are quantized, the number of wavelet transform coefficients close to 0 increases, so that the information amount can be further compressed.

  Here, as a method for encoding the wavelet transform coefficient included in the high frequency sub-band, a method for encoding LH, HL, and HH separately as in JPEG2000, for example, the same as shown in Patent Document 2, for example. There is a method in which LH, HL, and HH at the same spatial position in a plurality of subbands belonging to a hierarchy are encoded together. The present invention may use either method.

  The encoding parameter determination unit 133 sets the encoding parameter k of the wavelet transform coefficient to be encoded based on one or more wavelet transform coefficients already encoded among the wavelet transform coefficients stored in the transform coefficient memory 122. To calculate. Various wavelet transform coefficients to be referred to for calculating the encoding parameter k can be considered. For example, when the wavelet transform coefficient is encoded for each subband, the encoding parameter determination unit 133 uses the encoding parameter k based on one or more already encoded wavelet transforms of the same type as the wavelet transform coefficient to be encoded. Is calculated. That is, for LH, k is calculated based on the already encoded LH. Also, when LH, HL, and HH that are spatially at the same position in the high frequency subband belonging to the same layer are encoded together, the encoding parameter determination unit 133, for example, uses the wavelet transform coefficient to be encoded. Is calculated based on a plurality of already encoded nearby wavelet transform coefficients in a plurality of high frequency subbands belonging to the same layer. The encoding parameter k calculated as described above is transmitted from the encoding parameter determination unit 133 to the Golomb-Rice encoding unit 134.

The Golomb-Rice encoding means 134 reads out the wavelet transform coefficient to be encoded from the transform coefficient memory 122 and encodes it. Encoding is performed by the following first encoding method when the code length of the unicode code does not exceed the preset upper limit length max, and is encoded by the following second encoding method when exceeding the preset upper limit length max. . The upper limit length max can be set to 32, for example, when the encoding target data is 32 bits.
(1) First encoding method:
The data to be encoded is encoded as “uniary code” + “k-bit fixed length code (k is an encoding parameter)”. This is Golombrice coding itself.
(2) Second encoding method:
Encoding is performed by adding a series of bits “0” equal to a predetermined upper limit length max to the data to be encoded. This is encoding in the above-described raw mode.

  The Golomb-Rice encoding unit 134 transmits the code length of the generated code to the code output determination unit 135 and receives from the code output determination unit 135 the determination result as to whether the code can be output. When the determination result indicating that the code can be output is received, the Golomb-Rice encoding means 134 stores the generated code in the code FIFO 123 and uses the free capacity stored in the code FIFO free capacity memory 124 as the code of the code output this time. Subtract by amount. On the other hand, when the Golomb-Rice encoding unit 134 receives the determination result indicating that the code cannot be output, the generated code is discarded without being stored in the code FIFO 123.

  The code output determination means 135 is stored in the code length of the code generated by the Golomb-Rice encoding means 134, the free capacity of the code FIFO memory 123 stored in the code FIFO free capacity memory 124, and the coefficient skip number memory 125. Based on the code length of the code obtained by encoding the coefficient skip number and the coefficient skip number generated by the coefficient skip number encoding unit 137, whether or not each code is output to the code FIFO memory 123 is determined. Control.

  When the code generated by the Golomb-Rice encoding unit 134 is not stored in the code FIFO 123 for the wavelet transform coefficient to be encoded, the code output omitting unit 136 follows the instruction from the code output determining unit 135 and stores the coefficient skip count memory. The number of coefficient skips stored in 125 is incremented by one. The code output omitting unit 136 notifies the coefficient reference value changing unit 132 to change the value of the wavelet transform coefficient to be encoded that has not been encoded.

  The coefficient reference value changing unit 132 changes the value of the wavelet transform coefficient that has not been encoded among the wavelet transform coefficients stored in the transform coefficient memory 122 to a predetermined value in accordance with the instruction from the code output omitting unit 136. To do. In the present embodiment, 0 is used as a predetermined value. Of course, values other than 0 may be used.

  The coefficient skip number encoding unit 137 encodes the coefficient skip number stored in the coefficient skip number memory 125 by a predetermined method in accordance with an instruction from the code output determination unit 135 and stores the coefficient skip number in the code FIFO memory 123, and the code FIFO free space The free space stored in the memory 124 is subtracted by the code amount of the code output this time. Hereinafter, a method of encoding the coefficient skip number will be described.

  A code in which when the Golomb-Rice encoding means 134 performs encoding by the first and second encoding methods described above, binary notation indicating a value from 0 to 31 immediately follows 32 consecutive bits “0”. No pattern is generated. The reason is that a code pattern of 32 consecutive bits “0” appears only when encoding by the second encoding method is performed, but bit “0” is encoded by encoding by the second encoding method. This is because the binary representation immediately after “32” is never less than 32 (if it is less than 32, even if k = 0, it does not enter the raw mode). Therefore, in the present embodiment, the coefficient skip count 2 is added to the binary notation from 0 to 31 in the code pattern in which the binary notation from 0 to 31 immediately follows 32 consecutive bits “0”. Coding of the number of coefficient skips is performed by putting a hexadecimal notation. Specifically, if the data to be encoded is a fixed length of m bits, the number of skips from 1 to 32 is expressed by m bits immediately after 32 consecutive bits “0”. More specifically, for example, m-bit binary notation 0 is a code with a skip number 1, m-bit binary notation 1 is a skip-number 2 code,..., M-bit binary notation 31 is The number of skips is 32.

  For example, when encoding a 16-bit coefficient value, if encoding of 32 coefficients is skipped, a total of 512-bit coefficients are skipped. Since the coefficient skip number is encoded in the low mode, the output code amount during this period is 48 bits. Further, if the coefficient code cannot be output even after skipping 32 coefficients, if the count skip number is continuously output again in the low mode, 33 or more coefficient skips can be realized. In any case, according to the above encoding method, the band of the output signal can be suppressed to 48/512 = about 9.4% or less with respect to the coefficient band. In general, the wavelet transform coefficient has a larger dynamic range than the pixel value (8 bits to 16 bits, etc.), but the above method has a bandwidth reduction width (about 1/10) larger than such a bandwidth increase width (twice). Therefore, even if the band of the code FIFO memory is equal to or less than the memory band of the image signal, the code can be sufficiently output.

  The code output means 138 reads the code stored in the code FIFO memory 123 and outputs it to the code output device 14. Further, the code output unit 138 increases the free capacity stored in the code FIFO free capacity memory 124 in accordance with the code amount of the code read from the code FIFO memory 123.

  Next, the overall operation of the present embodiment will be described with reference to the flowcharts of FIGS.

  The data processing device 13 executes the operation shown in FIG. 3 and the operation shown in FIG. 4 in parallel. In the operation shown in FIG. 3, first, the stored contents of the code FIFO memory 123 are cleared (S101). Thereafter, the code output unit 138 checks whether there is untransmitted data in the code FIFO memory 123 at every signal transmission timing to the code output device 14, and if there is, transmits the untransmitted data to the code output device 14, The operation of updating the free space stored in the code FIFO free space memory 124 is repeated (S102 to S106).

  On the other hand, in the operation shown in FIG. 4, the wavelet transform encoding is performed on the image input from the image input device 11 and the operation stored in the code FIFO memory 123 is performed. Specifically, the following operations are performed.

  First, the image conversion means 131 performs wavelet transform of N layers (N is an integer of 1 or more) on the image input from the image input device 11 to the image memory 121, and the obtained wavelet transform coefficients of each subband. Is stored in the conversion coefficient memory 122 (S111).

  Next, the coefficient extraction coordinates are initialized (S112). Next, the code output omission unit 136 clears the coefficient skip count stored in the coefficient skip count memory 125 to 0 (S113).

  Next, the encoding parameter determining unit 133 converts one or more encoded wavelet transform coefficients to be referred to in order to determine the encoding parameter k for performing Golomb-Rice encoding of the wavelet transform coefficients of the coefficient extraction coordinates to the transform coefficient memory. Read from 122 (S114). The encoded one or more wavelet transform coefficients to be referred to may be arbitrary as long as they are the same on the encoding device side and the decoding device side. The encoding parameter determining unit 133 estimates the number of significant digits of the wavelet transform coefficient of the coefficient extraction coordinates from the read one or more encoded wavelet transform coefficient values, and determines the encoding parameter k (S115).

  Next, the Golomb-Rice encoding unit 134 encodes the wavelet transform coefficient of the coefficient extraction coordinates and transmits the code length to the code output determination unit 135 (S116). As described above, when the code length of the unicode code does not exceed the upper limit length max, the Golomb-Rice encoding unit 134 uses the encoding parameter k determined by the encoding parameter determination unit 133 to perform the first encoding. Encoding is performed by the method (Golomrice encoding). Further, when the code length of the unicode code exceeds the upper limit length max, the Golomb-Rice encoding means 134 performs encoding by the second encoding method (encoding in the low mode).

  The code output determination unit 135 reads the free capacity of the code FIFO memory 123 from the code FIFO free capacity memory 124 (S117). Further, the code output determination unit 135 reads the coefficient skip number from the coefficient skip number memory 125 and determines whether the coefficient skip number is 0 or 1 or more (S118). If the number of coefficient skips is 0, the code output determination unit 135 compares the code length of the code generated by the Golomb-Rice encoding unit 134 with the free capacity of the code FIFO memory 123 and encodes the code generated this time. It is determined whether or not the data can be output to the FIFO memory 123 (S123). If output is possible, the code output determination unit 135 notifies the Golomb-Rice encoding unit 134 that code output is possible. The Golomb-Rice encoding means 134 outputs the code generated this time to the code FIFO memory 123, and subtracts the free capacity stored in the code FIFO free capacity memory 124 by the code length (S124). Then, the process proceeds to step S127.

  On the other hand, when the code FIFO memory 123 has less free space than the code length of the code generated by the Golomb-Rice encoding unit 134, the code output determination unit 135 prohibits the code output from the Golomb-Rice encoding unit 134. Notice. Further, the code output determination unit 135 operates the code output omitting unit 136. The code output omitting unit 136 adds 1 to the coefficient skip count stored in the coefficient skip count memory 125 (S125). Further, the code output omitting unit 136 operates the coefficient reference value changing unit 132. The coefficient reference value changing unit 132 rewrites the value of the wavelet transform coefficient of the coefficient extraction coordinate in the transform coefficient memory 122 to 0 (S126). Then, the process proceeds to step S127.

  If the code output determination unit 135 determines in step S118 that the coefficient skip number stored in the coefficient skip number memory 125 is 1 or more, the code output determination unit 135 sets the free space and the coefficient skip number of the code FIFO 123 read in step S117. The code length of the code is compared, and it is determined whether or not the code of the coefficient skip count can be output to the code FIFO memory 123 (S119). If the output is impossible, the code output determination unit 135 notifies the Golomb-Rice encoding unit 134 that the code output is impossible in order to prevent the reverse of the code output order. Then, the code output determination unit 135 operates the code output omitting unit 136. The code output omitting unit 136 adds 1 to the coefficient skip count stored in the coefficient skip count memory 125 (S125). Further, the code output omitting unit 136 operates the coefficient reference value changing unit 132. The coefficient reference value changing unit 132 rewrites the value of the wavelet transform coefficient of the coefficient extraction coordinate in the transform coefficient memory 122 to 0 (S126). Then, the process proceeds to step S127.

  On the other hand, the code output determination unit 135 operates the coefficient skip number encoding unit 137 if the free capacity of the code FIFO memory 123 is equal to or larger than the code length of the code of the coefficient skip number. The coefficient skip number encoding means 137 reads the coefficient skip number from the coefficient skip number memory 125, encodes it, outputs it to the code FIFO memory 123, and converts the free capacity stored in the code FIFO free capacity memory 124 to the coefficient skip number. Only the code length is subtracted (S120). Further, the coefficient skip number encoding unit 137 operates the code output omitting unit 136. The code output omitting means 136 clears the coefficient skip count stored in the coefficient skip count memory 125 to 0 (S121). Subsequently, the code output determination means 135 reads the free capacity from the code FIFO free capacity memory 124 again (S122), and compares the code length of the code generated by the Golomb-rice coding means 134 with the free capacity of the code FIFO memory 123. Then, it is determined whether or not the code generated this time can be output to the code FIFO memory 123 (S123). If output is possible, the code output determination unit 135 notifies the Golomb-Rice encoding unit 134 that code output is possible. The Golomb-Rice encoding means 134 outputs the code generated this time to the code FIFO memory 123, and subtracts the free capacity stored in the code FIFO free capacity memory 124 by the code length (S124). Then, the process proceeds to step S127. On the other hand, when the code FIFO memory 123 has less free space than the code length of the code generated by the Golomb-Rice encoding unit 134, the code output determination unit 135 prohibits the code output from the Golomb-Rice encoding unit 134. Notice. Further, the code output determination unit 135 operates the code output omitting unit 136. The code output omitting unit 136 adds 1 to the coefficient skip count stored in the coefficient skip count memory 125 (S125). Further, the code output omitting unit 136 operates the coefficient reference value changing unit 132. The coefficient reference value changing unit 132 rewrites the value of the wavelet transform coefficient of the coefficient extraction coordinate in the transform coefficient memory 122 to 0 (S126). Then, the process proceeds to step S127.

  In step S127, it is determined whether or not the processing for all wavelet transform coefficients has been completed. If unprocessed wavelet transform coefficients remain, the coefficient extraction coordinates are shifted (S128), the process returns to step S114, and the same process as described above is repeated.

  If the processing for all the wavelet transform coefficients has been completed, the code output determination unit 135 determines whether the coefficient skip count stored in the coefficient skip count memory 125 is 0, 1 or more (S129). If the number of coefficient skips is 0, the process of FIG. If the number of coefficient skips is 1 or more, the code output determination unit 135 immediately waits for a free space if there is enough free space in the code FIFO memory 123 to store the code for the number of coefficient skips ( S130), the coefficient skip number encoding means 137 is operated. The coefficient skip number encoding means 137 reads the coefficient skip number from the coefficient skip number memory 125, encodes it, and outputs it to the code FIFO memory 123 (S131). Then, the process of FIG. 4 is finished.

  As described above, according to the present embodiment, when the code generated by the Golomb-Rice encoding means 134 cannot be stored in the code FIFO memory 123 due to a lack of free space in the code FIFO memory 123, the code FIFO memory 123 has sufficient capacity. The operation of outputting the code generated by the Golomb-Rice encoding means 134 to the code FIFO memory 123 is stopped until a space is available. Instead, the number of wavelet transform coefficients that could not be output as a code is counted as the number of coefficient skips, and when the code FIFO memory 123 is empty, the coefficient skip number is encoded and output through the code FIFO memory 123. Output to the device 14. Thereby, the decoding side can identify how many wavelet transform coefficients have not been encoded by decoding the coefficient skip number code. For this reason, it is possible to avoid a situation in which even the code of which wavelet transform coefficient has disappeared is not known on the decoding side and decoding becomes difficult.

  Further, according to the present embodiment, the wavelet transform coefficient for which the code could not be output is changed to a predetermined value (0 in the present embodiment) by the coefficient reference value changing unit 132. Thereby, when the encoding parameter k is determined by referring to another wavelet coefficient that has already been encoded, the encoding side and the decoding side can refer to the wavelet transform coefficient having the same value, and the parameter k It is possible to continue encoding and decoding using the Golomb-Rice code using.

  In this embodiment, the wavelet transform is taken up as the conversion process, but the data encoding apparatus of the present invention is also applicable to an encoding apparatus that performs other types of conversion processes such as DCT (discrete cosine transform). Is possible.

[Second Embodiment]
Next, a second embodiment of the present invention will be described with reference to the drawings.

  Referring to FIG. 5, in the second embodiment of the present invention, the present invention is applied to a wavelet transform decoding device, which includes a code input device 21, a data processing device 22 operated by program control, and a data storage. The apparatus 23 and the image output apparatus 24 are comprised.

  The code input device 21 is a device for inputting a code to be decoded, and is configured by, for example, a magnetic disk device or a communication device.

  The data storage device 23 includes a conversion coefficient memory 231 and an image memory 232.

  The transform coefficient memory 231 stores the wavelet transform coefficients decoded by the data processing device 22.

  The image memory 232 stores the image decoded by the data processing device 22.

  The data processing device 22 includes a decoding parameter determination unit 221, a Golomurice decoding unit 222, a coefficient skip unit 223, and an image inverse conversion unit 224.

  The decoding parameter determination unit 221 uses, for each wavelet transform coefficient to be decoded, a coding parameter used for encoding based on one or more already decoded other wavelet transform coefficients stored in the transform coefficient memory 231. A decoding parameter having the same value as k is calculated.

  The Golomb-rice decoding means 222 decodes the wavelet transform coefficient to be decoded from the code input from the code input device 21. Specifically, the Golomb-Rice decoding means 222 determines whether the input code includes the continuation of the bit “0” equal to the predetermined upper limit length max. ) Or the code according to the above-described second encoding method (encoding in the low mode). If the code is based on the first encoding method, the Golomb-Rice decoding unit 222 uses the decoding parameter k determined by the decoding parameter determination unit 221 to decode the Golom-Rice code, which is the code, and obtains the obtained wavelet transform coefficient. Stored in the conversion coefficient memory 231.

  If the code is based on the second encoding method, the Golomb-Rice decoding means 222 uses the binary representation of a predetermined bit length immediately after the continuation of the bit “0” equal to the upper limit length max as the decoding result. If the value of the decoding result is 32 or more, it is determined that the wavelet transform coefficient has been successfully decoded, and the decoding result is stored in the transform coefficient memory 231 as a decoded value of the wavelet transform coefficient. On the other hand, if the decoded value is less than 32, it is assumed that the decoding of the wavelet transform coefficient has failed, and the decoded value of the coefficient skip count is obtained from the decoding result and transmitted to the coefficient skip means 223. For example, as described above, m-bit binary notation 0 is used as a skip number 1 code, m-bit binary notation 1 is used as a skip number 2 code,..., M-bit binary notation 31 is skipped. When the code of Equation 32 is used, a value obtained by adding 1 to the decoding result is set as a decoded value of the coefficient skip number.

  The coefficient skip means 223 generates wavelet transform coefficients having a value of 0 for the number of coefficient skips received from the Golomb-rice decoding means 222 and stores it in the transform coefficient memory 231.

  The image inverse transform unit 224 reads the wavelet transform coefficient stored in the transform coefficient memory 231, performs wavelet inverse transform, and stores the obtained pixel data in the image memory 232.

  The image output device 24 is configured by a display device, for example, and reads and outputs pixel data stored in the image memory 232.

  Next, the overall operation of the present embodiment will be described with reference to the flowchart of FIG.

  First, the coefficient extraction coordinates are initialized (S201).

  Next, the decoding parameter determination means 221 reads out from the transform coefficient memory 231 as another reference wavelet the already decoded wavelet transform coefficient used for decoding the wavelet transform coefficient of the coefficient extraction coordinates (S202). When there is no other applicable wavelet transform coefficient, a predetermined value (for example, 0) is used as the value of the other wavelet transform coefficient that does not exist.

  Next, the Golombrice decoding means 222 decodes the wavelet transform coefficient of the coefficient extraction coordinates from the code input from the code input device 21 (S204). When the Golombrice decoding means 222 succeeds in decoding the wavelet transform coefficient, the Golombrice decoding means 222 stores the decoded value in the transform coefficient memory 231 as the wavelet transform coefficient of the coefficient extraction coordinates (S205, S206). On the other hand, when the decoding of the wavelet transform coefficient fails, the coefficient skip number is obtained from the obtained decoded value and output to the coefficient skip means 223. The coefficient skip means 223 generates a wavelet transform coefficient of value 0 as the wavelet transform coefficient of each coordinate corresponding to the number of coefficient skips from the coefficient extraction coordinates, and stores it in the transform coefficient memory 231 (S207).

  Next, it is determined whether or not the decoding of all coefficients has been completed (S208). If not completed, the coefficient extraction coordinates are shifted to the next coordinates (S209), and the process returns to step S202 to perform the same processing as described above. repeat.

  If the decoding of all the coefficients has been completed, the image inverse transform unit 224 performs wavelet inverse transform on the wavelet transform coefficients stored in the transform coefficient memory 231 and stores the obtained pixel data in the image memory 232. (S210).

  The image stored in the image memory 232 is output to the image output device 24.

  This embodiment corresponds to a data decoding apparatus encoded in the first embodiment. By decoding the number of coefficient skips and generating wavelet transform coefficients having a value of 0 by the number of decoded coefficient skips, there is an advantage that decoding can be continued without failure.

  Further, by decoding the number of coefficient skips, the decoding side can detect how many wavelet transform coefficients have been transmitted without being encoded. For this reason, if necessary, it is possible to take measures such as making a retransmission request to the code side again.

  In this embodiment, the wavelet transform is taken up as the conversion process, but the data decoding apparatus of the present invention can be applied to other types of conversion processes such as DCT (discrete cosine transform).

[Third embodiment]
Referring to FIG. 7, in the third embodiment of the present invention, the present invention is applied to a predictive encoding device, and an image input device 31, a data storage device 32, and data processing that operates by program control. The apparatus 33 and the code output apparatus 34 are comprised.

  The image input device 31 is the same as the image input device 11 of FIG. 1 in the first embodiment.

  The data storage device 32 includes an image memory 321, a pixel prediction value memory 322, a prediction error memory 323, a prediction error statistical memory 324, a code FIFO memory 325, a code FIFO free capacity memory 326, and a pixel skip count memory 327. Including.

  The image memory 321 stores an image input from the image input device 31.

  The predicted pixel value memory 322 stores the predicted pixel value generated by the data processing device 33.

  The prediction error memory 323 stores the prediction error generated by the data processing device 33.

  The prediction error statistical memory 324 stores statistical values of prediction errors.

  The code FIFO memory 325 is a buffer that stores, in a first-in first-out manner, a code obtained by Golomu-Rice encoding the prediction error stored in the prediction error memory 323.

  The code FIFO free capacity memory 326 stores the free capacity of the code FIFO memory 325.

  The pixel skip number memory 327 stores the number of pixels for which the sign of the prediction error has not been output to the code FIFO memory 325.

  The data processing device 33 includes a pixel reference value changing unit 331, a pixel predicting unit 332, a prediction error calculating unit 333, an encoding parameter determining unit 334, a Golomurice encoding unit 335, a code output determining unit 336, a code An output omission unit 337, a pixel skip number encoding unit 338, and a code output unit 339 are provided.

  For each pixel stored in the image memory 321, the pixel prediction unit 332 predicts the value of the pixel from a value of another pixel that has already been encoded, and the predicted value is stored in the pixel predicted value memory. Store at 322. The prediction method is arbitrary. For example, assuming that the horizontal direction is x, the vertical direction is y, and the pixel coordinates are expressed by (x, y), the pixel value at coordinates (i, j) is expressed as coordinates (i-1, j-1). It is also possible to predict from the value of the pixel, the value of the pixel at the coordinates (i, j−1), and the value of the pixel at the coordinates (i−1, j).

  The prediction error calculation unit 333 calculates a prediction error of each pixel stored in the image memory 321 and stores the prediction error in the prediction error memory 323. The prediction error of a certain pixel is calculated as the difference between the value of that pixel stored in the image memory 321 and the prediction value of that pixel stored in the pixel prediction value memory 322.

  The encoding parameter determination unit 334 calculates the encoding parameter of the prediction error to be encoded based on one or more other already encoded prediction errors among the prediction errors stored in the prediction error memory 323. To do. The method for determining the encoding parameter is arbitrary. For example, a method similar to the method described in Patent Document 1 can be used. In that case, the encoding parameter determination unit 334 stores statistical values such as the cumulative value of the absolute value and the cumulative number of prediction errors that have appeared so far in the prediction error statistical memory 324.

  The calculated encoding parameter k is transmitted from the encoding parameter determination unit 334 to the Golomb-Rice encoding unit 335.

  The Golomb-Rice encoding means 335 uses the same method (first encoding method and second encoding method) as the Golomb-Rice encoding means 134 of FIG. 1 of the first embodiment for the data to be encoded. Encoding is performed. The Golomb-Rice encoding unit 335 transmits the code length of the generated code to the code output determination unit 336 and receives a determination result as to whether or not the code can be output from the code output determination unit 336. When the determination result indicating that the code can be output is received, the Golomb-Rice encoding means 335 stores the generated code in the code FIFO 325 and uses the free capacity stored in the code FIFO free capacity memory 326 as the code of the code output this time. Subtract by amount. On the other hand, when the Golomb-Rice encoding means 335 receives the determination result indicating that the code cannot be output, the generated code is discarded without being stored in the code FIFO 325.

  The code output determination unit 336 stores the code length of the code generated by the Golomb-Rice encoding unit 335, the free space of the code FIFO memory 325 stored in the code FIFO free space memory 326, and the pixel skip count memory 327. Based on the code length of the code obtained by encoding the number of pixel skips and the number of pixel skips generated by the pixel skip number encoding unit 338, whether or not each code is output to the code FIFO memory 325 is determined. Control.

  When the code generated by the Golomb-Rice encoding unit 335 is not stored in the code FIFO 325 for the prediction error of the pixel to be encoded, the code output omitting unit 337 follows the instruction from the code output determining unit 336 and the number of pixel skips. The number of pixel skips stored in the memory 327 is incremented by one. Further, the code output omitting unit 337 notifies the pixel reference value changing unit 331 to change the value of the pixel for which the prediction error is not encoded.

  In response to the notification from the code output omitting unit 337, the pixel reference value changing unit 331 changes the value of the pixel in which the predicted pixel is not encoded among the pixel values stored in the image memory 321. In the present embodiment, the prediction error that has not been encoded is determined to be regarded as 0 on both the encoding side and the decoding side. Therefore, the pixel reference value changing unit 331 changes the pixel value so that the prediction error becomes zero. Specifically, the pixel reference value changing unit 331 reads the predicted value of the pixel in which the predicted pixel has not been encoded from the pixel predicted value memory 322 and overwrites the value of the pixel in the image memory 321. When it is determined that the prediction error is regarded as a value D other than 0, the pixel reference value changing unit 331 rewrites the value of the pixel with a value obtained by adding D to the predicted value of the pixel.

  The pixel skip number encoding unit 338 receives the notification from the code output determination unit 336 and converts the pixel skip number stored in the pixel skip number memory 327 into the coefficient skip number encoding of FIG. 1 of the first embodiment. The data is encoded by the same method as that of the means 137, stored in the code FIFO memory 325, and the free space stored in the code FIFO free space memory 326 is subtracted by the code amount of the code output this time.

  The code output means 339 reads the code stored in the code FIFO memory 325 and outputs it to the code output device 34. The code output means 339 increases the free capacity stored in the code FIFO free capacity memory 326 in accordance with the code amount of the code read from the code FIFO memory 325.

  Next, the overall operation of the present embodiment will be described with reference to the flowcharts of FIGS.

  The data processing device 33 executes the operation shown in FIG. 8 and the operation shown in FIG. 9 in parallel. In the operation shown in FIG. 8, first, the stored contents of the code FIFO memory 325 are cleared (S301). Thereafter, the code output means 339 checks whether there is untransmitted data in the code FIFO memory 325 at every signal transmission timing to the code output device 34, and if there is, transmits the untransmitted data to the code output device 34, The operation of updating the free space stored in the code FIFO free space memory 326 is repeated (S302 to S306).

  On the other hand, in the operation illustrated in FIG. 9, the prediction error is encoded for each pixel of the image input from the image input device 31, and the operation stored in the code FIFO memory 325 is performed. Specifically, the following operations are performed.

  First, pixel extraction coordinates are initialized (S311). Next, the code output omitting unit 337 clears the number of pixel skips stored in the pixel skip number memory 327 to 0 (S312).

  Next, the pixel predicting unit 332 reads other encoded pixels for obtaining the predicted value of the pixel at the pixel extraction coordinate from the image memory 321 (S313). Next, the pixel prediction unit 332 calculates a pixel predicted value of the pixel extraction coordinates from the other encoded pixels that have been read out and stores them in the pixel predicted value memory 322, and the prediction error calculation unit 333 calculates the pixel extraction coordinates. The difference between the pixel value and the predicted value is obtained as a prediction error and stored in the prediction error memory 323 (S314).

  Next, the encoding parameter determination means 334 determines an encoding parameter k for performing Golomb-Rice encoding on the predicted value of the pixel at the pixel extraction coordinate from the statistical value stored in the prediction error statistical memory 324 (S315).

  Next, the Golomb-Rice encoding unit 335 encodes the prediction error of the pixel at the pixel extraction coordinate, and transmits the code length to the code output determination unit 336 (S316). If the code length of the unicode code does not exceed the upper limit length max, the Golomb-Rice encoding unit 335 uses the encoding parameter k determined by the encoding parameter determination unit 334 to perform the first encoding method (Golom-Rice encoding). ). Further, when the code length of the unicode code exceeds the upper limit length max, the Golomb-Rice encoding means 335 performs encoding by the second encoding method (encoding in the low mode).

  The code output determination unit 336 reads the free capacity of the code FIFO memory 325 from the code FIFO free capacity memory 326 (S317). Further, the code output determination unit 336 reads the pixel skip number from the pixel skip number memory 327 and determines whether the pixel skip number is 0 or 1 (S319). If the number of pixel skips is 0, the code output determination unit 336 compares the code length of the code generated by the Golomb-Rice encoding unit 335 with the free capacity of the code FIFO memory 325 and encodes the code generated this time. It is determined whether or not the data can be output to the FIFO memory 325 (S323). If output is possible, the code output determination unit 336 notifies the Golomb-Rice encoding unit 335 that code output is possible. The Golomb-Rice encoding means 335 outputs the code generated this time to the code FIFO memory 325, and subtracts the free capacity stored in the code FIFO free capacity memory 326 by the code length (S324). When the code generated by the Golomb-Rice encoding unit 335 is output to the code FIFO memory 325, the encoding parameter determining unit 334 updates the prediction error statistical memory 324. That is, the prediction value of the pixel at the pixel extraction coordinate is read from the prediction error memory 323, and a new statistical value is calculated from the read prediction value and the statistical value stored in the prediction error statistical memory 324, and the prediction error statistical memory Stored in 324. Then, the process proceeds to step S327.

  On the other hand, the code output determination unit 336 prohibits code output from the Golomb-Rice encoding unit 335 when the free space in the code FIFO memory 325 is smaller than the code length of the code generated by the Golo-Murice encoding unit 335. Notice. Further, the code output determining unit 336 operates the code output omitting unit 337. The code output omission unit 337 adds 1 to the number of pixel skips stored in the pixel skip number memory 327 (S325). Further, the code output omitting unit 337 operates the pixel reference value changing unit 331. The pixel reference value changing unit 331 rewrites the pixel value of the coefficient extraction coordinate in the image memory 321 with the predicted value (S326). Then, the process proceeds to step S327.

  If the code output determination unit 336 determines in step S318 that the number of pixel skips stored in the pixel skip number memory 327 is 1 or more, the code output determination unit 336 sets the free space and the number of pixel skips of the code FIFO 325 read in step S317. The code length of the code is compared to determine whether or not the code of the pixel skip count can be output to the code FIFO memory 325 (S319). If the output is impossible, the code output determination unit 336 notifies the Golomb-Rice encoding unit 335 that the code output is impossible in order to prevent the reverse of the code output order. Then, the code output determining unit 336 operates the code output omitting unit 337. The code output omission unit 337 adds 1 to the number of pixel skips stored in the pixel skip number memory 327 (S325). Further, the code output omitting unit 337 operates the pixel reference value changing unit 331. The pixel reference value changing unit 331 replaces the pixel value of the pixel extraction coordinate in the image memory 321 with the predicted value of the pixel (S326). Then, the process proceeds to step S327.

  On the other hand, the code output determination unit 336 operates the pixel skip number encoding unit 338 if the free capacity of the code FIFO memory 325 is greater than or equal to the code length of the code of the pixel skip number. The pixel skip number encoding means 338 reads out the pixel skip number from the pixel skip number memory 327, encodes it, outputs it to the code FIFO memory 325, and sets the free capacity stored in the code FIFO free capacity memory 326 as the pixel skip number. Only the code length is subtracted (S320). Further, the pixel skip number encoding unit 338 operates the code output omitting unit 337. The code output omitting unit 337 clears the pixel skip number stored in the pixel skip number memory 327 to 0 (S321). Subsequently, the code output determination means 336 reads the free capacity from the code FIFO free capacity memory 326 again (S322), and compares the code length of the code generated by the Golomb-rice coding means 335 with the free capacity of the code FIFO memory 325. Then, it is determined whether or not the code generated this time can be output to the code FIFO memory 325 (S323). If output is possible, the code output determination unit 336 notifies the Golomb-Rice encoding unit 335 that code output is possible. The Golomb-Rice encoding means 335 outputs the code generated this time to the code FIFO memory 325, and subtracts the free capacity stored in the code FIFO free capacity memory 326 by the code length (S324). Then, the process proceeds to step S327. On the other hand, the code output determination unit 336 prohibits code output from the Golomb-Rice encoding unit 335 when the free space in the code FIFO memory 325 is smaller than the code length of the code generated by the Golo-Murice encoding unit 335. Notice. Further, the code output determining unit 336 operates the code output omitting unit 337. The code output omitting unit 337 adds 1 to the coefficient skip count stored in the pixel skip count memory 327 (S325). Further, the code output omitting unit 337 operates the pixel reference value changing unit 331. The pixel reference value changing unit 331 replaces the pixel value of the image extraction coordinate in the image memory 321 with the predicted value (S326). Then, the process proceeds to step S327.

  In step S327, it is determined whether or not processing for all pixels has been completed. If unprocessed pixels remain, the image extraction coordinates are shifted (S328), the process returns to step S313, and the same process as described above is repeated.

  If the processing for all the pixels has been completed, the code output determination unit 336 determines whether the number of pixel skips stored in the pixel skip number memory 327 is 0, 1 or more (S329). If the number of pixel skips is 0, the process in FIG. If the number of pixel skips is 1 or more, the code output determination unit 336 immediately waits if there is enough free space in the code FIFO memory 325 to store the code for the number of pixel skips. S330), the pixel skip number encoding means 338 is operated. The pixel skip number encoding means 338 reads out the pixel skip number from the pixel skip number memory 327, encodes it, and outputs it to the code FIFO memory 325 (S331). Then, the process of FIG. 9 is finished.

  As described above, according to the present embodiment, when the code generated by the Golomb-Rice encoding means 335 cannot be stored in the code FIFO memory 325 due to a lack of free space in the code FIFO memory 325, the code FIFO memory 325 is sufficient. The operation of outputting the code generated by the Golomb-rice coding means 335 to the code FIFO memory 325 is stopped until a space is available. Instead, the number of pixels for which the code of the prediction error could not be output is counted as the number of pixel skips, and when the code FIFO memory 325 is vacant, the number of pixel skips is encoded and the code FIFO memory 325 is encoded. To the code output device 34. Accordingly, the decoding side can identify how many prediction errors of the pixels have not been encoded by decoding the pixel skip number code. For this reason, it is possible to avoid a situation in which decoding is difficult because even the prediction error code of which pixel has disappeared is unknown on the decoding side.

  Further, according to the present embodiment, a pixel for which a sign of the prediction error could not be output is set to a predetermined value (the same value as the prediction error of the pixel in the present embodiment) by the pixel reference value changing unit 331. change. As a result, when the encoding parameter k is determined with reference to the values of other already encoded pixels, both the encoding side and the decoding side can refer to the same pixel value, and the parameter k It is possible to continue encoding and decoding using the Golomb-Rice code using.

[Fourth embodiment]
Next, a fourth embodiment of the present invention will be described with reference to the drawings.

  Referring to FIG. 10, in the fourth embodiment of the present invention, the present invention is applied to a prediction error decoding device, which includes a code input device 41, a data processing device 42 operated by program control, and a data storage. The apparatus 43 and the image output apparatus 44 are comprised.

  The code input device 41 is the same as the code input device 21 of FIG. 5 of the second embodiment.

  The data storage device 43 includes a prediction error statistical memory 431, a prediction error memory 432, a pixel prediction value memory 433, and an image memory 434.

  The prediction error statistical memory 431 stores the statistical value of the prediction error calculated by the data processing device 42.

  The prediction error memory 432 stores the prediction error decoded by the data processing device 42.

  The predicted pixel value memory 433 stores the predicted value of the pixel decoded by the data processing device 42.

  The image memory 434 stores the pixel value decoded by the data processing device 42.

  The data processing device 42 includes a decoding parameter determination unit 421, a Golomurice decoding unit 422, a pixel value restoration unit 423, a pixel skip unit 424, and a pixel prediction unit 425.

  For each prediction error to be decoded, the decoding parameter determination unit 421 uses a decoding parameter having the same value as the encoding parameter k used for encoding based on the prediction error statistical value stored in the prediction error statistical memory 431. Is calculated.

  The Golombrice decoding means 422 decodes the prediction error of the decoding target pixel from the code input from the code input device 41. More specifically, the Golomb-Rice decoding means 422 determines whether the input code includes the continuation of the bit “0” equal to the predetermined upper limit length max, depending on whether or not the first encoding method (Golom-Rice encoding) is performed. ) Or the code according to the above-described second encoding method (encoding in the low mode). If the code is based on the first encoding method, the Golomb-Rice decoding unit 222 uses the decoding parameter k determined by the decoding parameter determination unit 421 to decode the Golomb-Rice code, which is the code, and predicts the obtained prediction error. Store in the error memory 432.

  In the case of a code according to the second encoding method, the Golombrice decoding means 422 uses a binary representation of a predetermined bit length immediately after the continuation of the bit “0” equal to the upper limit length max as the decoding result. If the decoding result value is 32 or more, a prediction error decoding value is obtained from the decoding result and stored in the prediction error memory 432. On the other hand, if the decoding result is less than 32, the decoding result is transmitted to the pixel skip means 424 as a decoded value of the number of pixel skips.

  The pixel skip unit 424 generates a pixel having the pixel predicted value stored in the pixel predicted value memory 433 as a pixel value for the pixels corresponding to the number of pixel skips received from the Golomb-rice decoding unit 422 and stores the generated pixel in the image memory 434. .

  The pixel value restoration unit 423 restores the pixel value from the prediction error stored in the prediction error memory 432 and the pixel prediction value stored in the pixel prediction value memory 433, and stores the pixel value in the image memory 434.

  The pixel prediction unit 425 calculates a predicted value of a pixel to be decoded from the values of other already decoded pixels stored in the image memory 434, and stores them in the pixel predicted value memory 433.

  The image output device 44 is the same as the image output device 24 of FIG. 5 of the second embodiment.

  Next, the overall operation of the present embodiment will be described with reference to the flowchart of FIG.

  First, pixel extraction coordinates are initialized (S401).

  Next, the pixel predicting unit 425 reads other decoded pixels that are referred to in order to calculate the predicted value of the pixel of the pixel extraction coordinates from the image memory 434 (S402). Next, the pixel predicting means 425 predicts the pixel value of the pixel extraction coordinate from the read values of the other pixels, and stores it in the pixel predicted value memory 433 (S403). When there is no other applicable pixel, a predetermined value (for example, 0) is used as the value of the other pixel that does not exist.

  Next, the decoding parameter determination unit 421 determines a decoding parameter k from the statistical value of the prediction error stored in the prediction error statistical memory 431 (S404). Next, the Golomb-rice decoding means 422 decodes the prediction error of the pixel at the pixel extraction coordinate from the code input from the code input device 41 (S405).

  When the Golomb-rice decoding means 422 succeeds in decoding the prediction error, the Golombrice decoding means 422 stores the decoded value in the prediction error memory 432 as the prediction error of the pixel at the pixel extraction coordinate. The pixel value restoration unit 423 calculates the pixel value of the pixel extraction coordinate from the prediction error of the pixel of the pixel extraction coordinate read from the prediction error memory 432 and the prediction value of the pixel of the pixel extraction coordinate read from the pixel prediction value memory 433. Is stored in the image memory 434 (S406 to S408). Next, the Golomb-Rice decoding means 422 updates the statistical value of the prediction error in the prediction error statistical memory 431 using the prediction error decoded this time (S409).

  On the other hand, the Golomb-rice decoding unit 422 obtains the number of pixel skips from the obtained decoded value and outputs it to the pixel skip unit 424 when decoding of the pixel prediction error fails. The pixel skip means 424 reads the predicted value of each coordinate for the number of pixel skips from the pixel extraction coordinate from the pixel predicted value memory 433 and stores it in the image memory 434 as the value of the pixel for each number of pixel skip from the image extraction coordinate. (S410).

  Next, it is determined whether or not the decoding of all the pixels has been completed (S411). If not, the pixel extraction coordinates are shifted to the next coordinates (S412), and the process returns to step S402 to perform the same processing as described above. repeat.

  If the decoding of all the pixels has been completed, the processing in FIG. 11 ends. The pixel data stored in the image memory 434 is then output to the image output device 44.

  This embodiment corresponds to a data decoding device encoded in the second embodiment. By decoding the number of pixel skips and using the predicted value of the pixel as the value of the pixel by the number of decoded pixel skips, there is an advantage that decoding can be continued without failure.

  Further, by decoding the number of pixel skips, the decoding side can detect how many prediction values of pixels have been transmitted without being encoded. For this reason, if necessary, it is possible to take measures such as making a retransmission request to the code side again.

  Although the embodiment of the present invention has been described above, the present invention is not limited to the above embodiment, and various other additions and modifications can be made. For example, the variable length code is not limited to the Golomb-Rice code. In addition, the data encoding / decoding device of the present invention can be realized by a computer and a program, as well as by hardware. The program is provided by being recorded on a computer-readable recording medium such as a magnetic disk or a semiconductor memory, and is read by the computer at the time of starting up the computer, etc. It functions as a data encoding / decoding device in the form.

11 Image input device 12 Data storage device 13 Data processing device 14 Code output device 21 Code input device 22 Data processing device 23 Data storage device 24 Image output device 32 Data storage device 33 Data processing device 34 Code output device 41 Code input device 42 Data Processing device 43 Data storage device 44 Image output device 121 Image memory 122 Conversion coefficient memory 123 Code FIFO
124 Code FIFO capacity memory 125 Coefficient skip number memory 131 Image converting means 132 Coefficient reference value changing means 133 Coding parameter determining means 134 Golomurice coding means 135 Code output determining means 136 Code output omitting means 137 Coefficient skip number coding means 138 Code Output means 221 Decoding parameter determination means 222 Golomurice decoding means 223 Coefficient skip means 224 Image inverse transform means 231 Transformation coefficient memory 232 Image memory 321 Image memory 322 Pixel prediction value memory 323 Prediction error memory 324 Prediction error statistics memory 325 Code FIFO
326 Code FIFO capacity memory 327 Pixel skip number memory 331 Pixel reference value change means 332 Pixel prediction means 333 Prediction error calculation means 334 Coding parameter determination means 335 Golomurice coding means 336 Code output determination means 337 Code output omission means 338 Number of pixel skips Coding means 339 Code output means 421 Decoding parameter determination means 422 Golomurice decoding means 423 Pixel value restoration means 424 Pixel skip means 425 Pixel prediction means 431 Prediction error statistical memory 432 Prediction error memory 433 Pixel prediction value memory 434 Image memory

Claims (35)

A buffer for temporarily storing the generated code;
Variable length encoding means for variable length encoding the encoding target data;
Code output omitting means for counting, as a skip number, one or more consecutive codes of codes that could not be stored in the buffer due to a lack of free space in the buffer among the codes generated by the variable length encoding means;
A skip number encoding means for encoding the skip number counted by the code output omitting means when the buffer is empty and storing it in the buffer;
A data encoding device comprising:   Of the codes generated by the variable-length coding means, the codes that could not be stored in the buffer can continue the encoding process as if the data to be encoded was encoded with a predetermined value. The data encoding apparatus according to claim 1, further comprising a changing unit configured to change to   The data encoding apparatus according to claim 2, wherein the encoding target data is conversion data obtained by performing predetermined conversion processing on input data.   4. The data encoding according to claim 3, wherein the changing unit changes the conversion data, which is data to be encoded, to the predetermined value for a code that could not be stored in the buffer. apparatus.   3. The encoding target data is a prediction error indicating a difference between a predicted value predicted from the value of another input data that has already encoded the value of the input data and the value of the input data. The data encoding device described in 1.   For the code that could not be stored in the buffer, the changing unit regards the prediction error that is the encoding target data as the predetermined value, and the prediction error that is the encoding target data. 6. The data code according to claim 5, wherein the value of the input data used for calculating the prediction error is rewritten with a value obtained by adding the predicted value used for calculating the predetermined value and the predetermined value. Device. A code output judging means;
The code output determination means includes
Each time a code is generated by the variable length encoding means, it is determined whether or not the skip number is a predetermined value or more,
If it is greater than or equal to a predetermined value, it is determined whether or not the buffer has enough free space to store the code of the current skip number,
If there is no free space, the skip number is updated by the code output omitting means,
If there is free space, a code obtained by encoding the current skip number by the skip number encoding unit is stored in the buffer, the skip number is initialized, and the code generated this time by the variable length encoding unit It is determined again whether or not the buffer has enough free space to store the code. If there is free space, the code generated this time is stored in the buffer, and if there is no free space, the code output omitting means Update the number of skips,
If the number of skips is not greater than or equal to a predetermined value, it is determined whether or not there is enough free space in the buffer to store the code generated this time by the variable length coding means. The code is stored in the buffer, and if there is no free space, the skip number is updated by the code output omitting means.
The data encoding device according to claim 1, wherein the data encoding device is a data encoding device. The variable length encoding means, when the encoding parameter is k, the data to be encoded is divided into two, lower k bits and upper remaining bits, and the upper remaining bits are replaced with a hexadecimal code If the code length of the unicode code does not exceed a preset upper limit length max, the encoding target data is encoded by first encoding that replaces the higher-order remaining bits with the unicode code, and the unicode If the code length of the code exceeds the upper limit length max, the encoding target data is obtained by second encoding that adds a sequence of bits having a predetermined value equal to the number of the upper limit length max before the encoding target data. Encoding,
The skip number encoding means adds a continuation of bits having a predetermined value equal to the number of the upper limit length max before a fixed-length binary notation indicating the number of skips to be encoded. Encoding,
The data encoding apparatus according to claim 1, wherein the data encoding apparatus is a data encoding apparatus. A decoding device that receives as input a code string that includes a variable-length encoded code of encoding target data and a code that encodes a skip number indicating how many consecutive encoding target data are not encoded. And
Decoding means for decoding the encoding target data and the skip number from the code string;
The skip means that treats the data to be encoded as being encoded with a predetermined value by the number of skips decoded by the decoding means,
A data decoding apparatus comprising:   The data decoding apparatus according to claim 9, wherein the encoding target data is conversion data obtained by performing predetermined conversion processing on input data.   11. The data decoding apparatus according to claim 10, wherein the skip means generates the converted data by the skip count for the predetermined value.   10. The encoding target data is a prediction error indicating a difference between a predicted value predicted from a value of another input data that has already encoded the value of the input data and the value of the input data. The data decoding device described in 1.   The skip means regards the prediction error that is the encoding target data as the predetermined value, and the prediction value used for calculating the prediction error that is the encoding target data 13. The data decoding apparatus according to claim 12, wherein input data having a value obtained by adding a predetermined value is generated for the number of skips. The code generated by variable length encoding of the data to be encoded is stored in the buffer,
The number of consecutive ones or more of codes that could not be stored in the buffer due to insufficient free space in the buffer is counted as a skip number,
When the buffer is vacant, the counted skip number is encoded and stored in the buffer.
A data encoding method characterized by the above.   15. The data encoding method according to claim 14, wherein the encoding process is continued with respect to a code that cannot be stored in the buffer, assuming that the encoding target data is encoded with a predetermined value. .   The data encoding method according to claim 15, wherein the encoding target data is conversion data obtained by performing predetermined conversion processing on input data.   17. The data encoding method according to claim 16, wherein for the code that could not be stored in the buffer, the converted data that is the data to be encoded is changed to the predetermined value.   16. The encoding target data is a prediction error indicating a difference between a predicted value predicted from the value of another input data that has already encoded the value of the input data and the value of the input data. The data encoding method described in 1.   For the code that could not be stored in the buffer, the prediction error that is the data to be encoded is considered to be the predetermined value, and was used to calculate the prediction error that is the data to be encoded. 19. The data encoding method according to claim 18, wherein the value of the input data used for calculation of the prediction error is rewritten with a value obtained by adding the prediction value and the predetermined value. A code string including a code obtained by variable-length coding the data to be encoded and a code obtained by encoding a number of skips indicating how many continuous data to be encoded are not encoded,
Decoding the encoding target data and the skip number from the code string,
Continue decoding thereafter, assuming that the data to be encoded is encoded with a predetermined value by the number of skips that have been decoded.
The data decoding method characterized by the above-mentioned.   The data decoding method according to claim 20, wherein the encoding target data is conversion data obtained by performing predetermined conversion processing on input data.   The data decoding method according to claim 21, wherein the predetermined value is generated by the number of skips for the converted data.   21. The encoding target data is a prediction error indicating a difference between a predicted value predicted from a value of another input data that has already encoded the value of the input data and the value of the input data. The data decoding method described in 1.   Assuming that the prediction error that is the encoding target data is the predetermined value, the prediction value used for the calculation of the prediction error that is the encoding target data and the predetermined value 24. The data decoding method according to claim 23, wherein input data having a value obtained by adding is generated for the number of skips. A computer having a buffer for temporarily storing the generated code;
Variable length encoding means for variable length encoding the encoding target data;
Code output omitting means for counting, as a skip number, one or more consecutive codes of codes that could not be stored in the buffer due to a lack of free space in the buffer among the codes generated by the variable length encoding means;
A skip number encoding means for encoding the skip number counted by the code output omitting means when the buffer is empty and storing it in the buffer;
Program to make it function. Said computer further
Of the codes generated by the variable-length coding means, the codes that could not be stored in the buffer can continue the encoding process as if the data to be encoded was encoded with a predetermined value. Change means,
The program according to claim 25, for causing the program to function as:   26. The program according to claim 25, wherein the encoding target data is conversion data obtained by performing predetermined conversion processing on input data.   28. The program according to claim 27, wherein the changing unit changes the conversion data, which is data to be encoded, to the predetermined value for a code that could not be stored in the buffer.   27. The encoding target data is a prediction error indicating a difference between a predicted value predicted from the value of another input data that has already encoded the value of the input data and the value of the input data. The program described in.   For the code that could not be stored in the buffer, the changing unit regards the prediction error that is the encoding target data as the predetermined value, and the prediction error that is the encoding target data. 30. The program according to claim 29, wherein the value of the input data used for calculation of the prediction error is rewritten with a value obtained by adding the prediction value used for calculation of the value and the predetermined value. Configuring a decoding apparatus that receives as input a code string that includes a variable-length encoded code of the data to be encoded and a code that encodes the number of skips indicating how many consecutive data to be encoded are not encoded Computer
Decoding means for decoding the encoding target data and the skip number from the code string;
The skip means that treats the data to be encoded as being encoded with a predetermined value by the number of skips decoded by the decoding means,
Program to make it function.   32. The program according to claim 31, wherein the encoding target data is conversion data obtained by performing predetermined conversion processing on input data.   33. The program according to claim 32, wherein the skip means generates the conversion data by the skip count for the predetermined value.   32. The encoding target data is a prediction error indicating a difference between a predicted value predicted from the value of another input data that has already encoded the value of the input data and the value of the input data. The program described in.   The skip means regards the prediction error that is the encoding target data as the predetermined value, and the prediction value used for calculating the prediction error that is the encoding target data 35. The program according to claim 34, wherein input data having a value obtained by adding a predetermined value is generated for the number of skips.

Images