JP2009177850A - Decoder, and method thereof - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To substantially equalize the quality of an obtained decoded image to the quality of an input image. <P>SOLUTION: An encoder so generates the difference-value data of image data from its input data, and so processes this data a discrete-cosine-transform-wise at an arbitrary output-bit accuracy as to generate the discrete-cosine-transform factors of the input image data. Then, the discrete-cosine-transform factors are quantized by using an expanded quantizing scale, based on at least one of the output-bit accuracy and an input-bit accuracy of the input image data. Thereafter, the obtained quantized factors are so compression-coded by using a variable-length coding table as to be able to perform the compression-coding while preventing the degrading of the image quality. Also, the compression-coded quantized factors are compression-decoded by using the variable-length coding table. Then, the obtained quantized factors are so processed inverse-quantization-wise by using the expanded quantization scale, based on at least one of the input-bit accuracy of the input image data and the output-bit accuracy of the discrete-cosine-transform factors as to be able to perform the decoding while preventing the degrading of the image quality. <P>COPYRIGHT: (C)2009,JPO&INPIT

Description

  The present invention relates to an encoding apparatus and method thereof, a decoding apparatus and method thereof, and is suitable for application to, for example, an encoder that compresses and encodes moving image data and a decoder that decodes the compressed and encoded moving image data. It is.

  Conventionally, as a system provided with this type of encoding device and decoding device, there is a system for transmitting moving image data to a remote place such as a video conference system or a video phone system.

  In such a system, in order to transmit moving image data efficiently using a transmission path, the moving image data is sequentially compressed and encoded in units of frames using line correlation or inter-frame correlation of images in units of frames. ing.

  Here, as a representative compression encoding method for encoding moving image data with high efficiency, there is an MPEG2 (Moving Picture Experts Group 2: moving image encoding for storage) method.

  This MPEG2 system was discussed in ISO-IEC / JTC1 / SC2 / WG11 (International Organization for Standardization-International Electrotechnical Commission / Joint Technical Committee 1 / Sub Committee 2 / Working Group 11) and was proposed as a standard proposal. In addition, a hybrid method in which motion compensation predictive coding and DCT (Discrete Cosine Transform) coding are combined is employed.

  In the MPEG2 system, several profiles and levels are defined to support various applications and functions. The most basic one is the main profile main level (MP @ ML: Main Profile at Main Level).

  Here, as an encoder to which the main profile main level in the MPEG2 system is applied, for example, there is an encoder configured as shown in FIG.

  That is, in the encoder 1 shown in FIG. 24, moving image data to be compressed and encoded supplied from the outside is sequentially fetched into the frame memory 2 in units of frames, and the fetched frame-unit image data (hereinafter referred to as frame images). Temporarily called data).

  In this case, the frame image data supplied to the frame memory 2 is composed of a luminance component and two color difference components in accordance with the MPEG2 system, and the bit accuracy of each pixel value (that is, the accuracy of color gradation) is 8 bits. It is defined that only 8-bit images with the accuracy of can be supplied.

  Incidentally, in the frame image data, having 8-bit accuracy means that the range of values of the luminance component and the two color difference components (that is, the gradation of the color) is 0 to 225 (256 gradations). It turns out that.

The motion vector detector 3 reads the frame image data stored in the frame memory 2 in units of macroblocks composed of 16 pixels × 16 lines , for example, and reads the read macroblock data (hereinafter referred to as a macroblock). Motion vector) (referred to as block data).

  Here, in the motion vector detector 3, each frame image data is converted into an I (Intra) picture (intra-frame encoded image), a P (Predictive) picture (forward prediction encoded image) or a B (Bidirectionally predictive) picture ( Bi-directional prediction encoded image).

  Incidentally, in the motion vector detector 3, for example, which of the I picture, the P picture, and the B picture is to be processed as the frame image data that is sequentially input is determined in advance (for example, , Continuous frame image data is processed as a sequence of I picture, B picture, P picture, B picture, P picture,..., B picture, P picture).

  In practice, the motion vector detector 3 refers to predetermined predetermined reference frame image data (hereinafter referred to as reference frame image data) in the frame image data stored in the frame memory 2. Then, pattern matching (block matching) between the reference frame image data and the 16-pixel × 16-line macroblock data of the frame image data that is currently subject to compression encoding allows movement of the macroblock data. Detect vectors.

  Here, in the MPEG system (MPEG1 system and MPEG2 system), as an encoding mode of macroblock data, an intra encoding (intraframe encoding) mode, a forward predictive encoding mode, a backward predictive encoding mode, and a bidirectional predictive code are used. Four types of conversion modes are defined.

  The macroblock data of the I picture is intra-coded according to the intra-coding mode. Further, the macroblock data of the P picture is intra-coded or forward-predicted coded according to either the intra-coding mode or the forward-predictive coding mode.

Moreover, macroblock data is intra coding mode of a B picture, a forward predictive coding mode, intra coding accordance with any of the backward prediction encoding mode or both directions predictive coding mode, forward predictive coding, backward predictive coding, or Bidirectional predictive coding is performed. Incidentally, these encoding modes are defined so that they can be selected in units of macroblock data even in the same frame image data.

  For this reason, the motion vector detector 3 sets the coding mode to the intra coding mode for the macroblock data of the I picture, and sets the intra coding mode without detecting the motion vector of the macroblock data. Coding mode information indicating this is sent to a variable length coding (VLC) unit 4 and a motion compensator 5.

  In addition, the motion vector detector 3 detects the motion vector of the macroblock data by performing forward prediction on the macroblock data of the P picture, and generates a prediction error caused by performing the forward prediction, and compression coding. For example, the variance of the target macroblock data (macroblock data of P picture) is compared.

  Then, as a result of the comparison, when the variance of the macroblock data is smaller than the prediction error, the motion vector detector 3 sets the encoding mode to the intra encoding mode, and sets the encoding mode information representing this. This is sent to the variable length encoder 4 and the motion compensator 5.

  On the other hand, when the prediction error is smaller than the variance of the macroblock data, the motion vector detector 3 sets the encoding mode to the forward predictive encoding mode, and encodes the set forward encoding mode. The encoding mode information is sent to the variable length encoder 4 and the motion compensator 5 together with the detected motion vector.

  Further, the motion vector detector 3 detects the motion vector of the macroblock data for each of the forward prediction, backward prediction and bidirectional prediction by performing forward prediction, backward prediction and bidirectional prediction for the macroblock data of the B picture. .

  In this case, the motion vector detector 3 detects a minimum prediction error (hereinafter referred to as a minimum prediction error) from prediction errors caused by forward prediction, backward prediction, and bidirectional prediction, and detects the detected minimum prediction. The error is compared with, for example, the variance of macroblock data to be compression-encoded (macroblock data of a B picture).

  Then, as a result of the comparison, when the variance of the macroblock data is smaller than the minimum prediction error, the motion vector detector 3 sets the encoding mode to the intra encoding mode, and encodes mode information indicating this. Is sent to the variable length encoder 4 and the motion compensator 5.

On the other hand, when the minimum prediction error is smaller than the variance of the macroblock data, the motion vector detector 3 determines the coding mode according to forward prediction, backward prediction, and bidirectional prediction from which the minimum prediction error is obtained. and forward predictive coding mode, is set to the backward prediction encoding mode or bidirectionally predictive encoding mode, the setting was forward predictive coding mode, the coding mode or backward predictive coding mode representing a bidirectional prediction encoding mode The information is sent to the variable length encoder 4 and the motion compensator 5 together with the corresponding motion vector.

  The motion compensator 5 is based on the coding mode information and the motion vector supplied from the motion vector detector 3 and is compression-encoded and already locally decoded frame image stored in the frame memory 6. Data (hereinafter referred to as local decoded frame image data) is read out, and the read out local decoded frame image data is used as frame image data for macroblock data prediction (hereinafter referred to as predicted frame image data). To the arithmetic units 7 and 8.

  The computing unit 7 reads the same macroblock data as the macroblock data read from the frame memory 2 by the motion vector detector 3 from the frame memory 2, and is supplied from the read macroblock data and the motion compensator 5. A difference value with the corresponding macroblock data in the predicted frame image data is calculated, and this is sent to the DCT unit 9 as difference value data.

  Incidentally, the bit accuracy of the pixel value of the output image (input image to the DCT unit 9) obtained from the computing unit 7 is a difference value between corresponding macroblock data, so that the bit value is increased by 1 bit to 9 bits, The range that the value can take is -255 to +255.

  On the other hand, when only the coding mode information is given from the motion vector detector 3, the motion compensator 5 does not transmit the predicted frame image data because the coding mode is the intra coding mode.

  Therefore, at this time, the arithmetic unit 7 (the same applies to the arithmetic unit 8) sends the macroblock data read from the frame memory 2 without any particular processing as it is to the DCT unit 9 as difference value data.

  The DCT unit 9 performs a discrete cosine transform process (hereinafter referred to as DCT process) on the difference value data provided from the arithmetic unit 7, and a discrete cosine transform coefficient (hereinafter referred to as DCT) obtained as a result. Are sent to the quantizer 10.

  Incidentally, in the DCT unit 9, the macroblock data composed of 16 pixels × 16 lines is further subdivided into blocks composed of 8 pixels × 8 lines in accordance with the MPEG standard, and the data of these blocks of 8 pixels × 8 lines is divided. On the other hand, DCT processing is performed.

  Further, the DCT unit 9 quantizes the 12-bit precision DCT coefficient obtained by performing DCT processing on the difference value data according to the MPEG standard and increasing the 9-bit precision of the difference value data by 3 bits. It is sent to the device 10. Incidentally, the possible range of the value of the DCT coefficient is -2048 to +2047.

  The quantizer 10 sets a quantization scale corresponding to the data accumulation amount (the amount of data stored in the buffer 11) given from the buffer 11, and based on the set quantization scale. The DCT coefficient given from the DCT unit 9 is quantized. Incidentally, the quantizer 10 converts the fraction generated as a result of quantizing the DCT coefficient into a whole number by performing a rounding process such as rounding to four.

  Here, in the quantizer 10, there are two methods of determining the value of the quantization scale: linear quantization and nonlinear quantization.

  As shown in FIG. 25, in the quantizer 10, when linear quantization is performed, a value twice that of the quantization scale code is used as the quantization scale value. In addition, in the quantizer 10, when nonlinear quantization is executed, a quantization scale value that is nonlinear with respect to the quantization scale code is used.

  That is, in the linear quantization, when the quantization scale code and the value of the quantization scale are graphed, the result is as shown in FIG. 26. The relationship between the quantization scale code and the value of the quantization scale is expressed by the quantization. When the scale code is qsc and the quantization scale is qs, the equation (1)

Can be expressed as

  Further, in the non-linear quantization, when the relationship between the quantization scale code and the value of the quantization scale is graphed, as shown in FIG. 27, the quantization scale code is set to qsc as in the above equation (1), When the quantization scale is qs, the relationship between the quantization scale code and the value of the quantization scale is expressed by equation (2).

It is represented by

  In the non-linear quantization, the quantization scale code is expressed by 5 bits in accordance with the MPEG2 standard. Therefore, the value consisting of the upper 2 bits of the quantization scale code is k, and the value consisting of the remaining lower 3 bits is j. Then, the above equation (2) is replaced with equation (3).

It can also be expressed instead of. Incidentally, the possible range of the value of k is 0-3, and the possible range of the value of j is 0-7. Further, in this equation (3), k = j = 0 is prohibited.

  By the way, in the quantizer 10 (FIG. 24), the quantization scale value is not directly specified, but the quantization scale code is specified in accordance with the MPEG2 standard, linear quantization, non-linear quantization, The quantization scale value corresponding to the designated quantization scale code is calculated from the corresponding expression (1), expression (2), or expression (3).

  Incidentally, in the quantizer 10, it is possible to set, in units of pictures, which one of linear quantization and nonlinear quantization is used in accordance with the MPEG2 standard.

  Further, in the quantizer 10, since the quantization scale code is represented by 5 bits in accordance with the MPEG2 standard, the range of values of the quantization scale code is 1 to 31 (0 is prohibited). When linear quantization is used, the possible range of quantization scale values is 2 to 62 (FIG. 25).

  In the quantizer 10, when linear quantization is performed to quantize the DCT coefficient, the range of values of the quantization coefficient obtained as a result is a quantization scale of 2 values from −1024 to + 1023, and a quantization scale of 62 is -33 to +33.

  By the way, in the quantizer 10, if the value of the quantization scale used for quantization is relatively small, a quantization coefficient having a relatively large value is transmitted as a result of the quantization. The amount of codes generated in the encoder 4 is relatively large, and the bit rate is increased.

  On the other hand, in the quantizer 10, if the value of the quantization scale used for quantization is relatively large, a quantization coefficient having a relatively small value is transmitted as a result of the quantization. The amount of code generated in the variable length encoder 4 is relatively small, and the bit rate is small. Thus, the value of the quantization scale is directly related to the bit rate after encoding.

  In the quantizer 10, as is clear from FIG. 25, when linear quantization is used, only even values can be used as quantization scale values. That is, it is difficult to finely adjust the bit rate when specifying a relatively small quantization scale code.

  On the other hand, in the quantizer 10, when it is desired to extremely reduce the bit rate, only the quantization scale having a value of 62 at the maximum can be used. Therefore, the bit rate cannot be extremely decreased.

  Therefore, in the MPEG2 system, it is specified that nonlinear quantization can be used in addition to linear quantization at the time of quantization. In the quantizer 10, as is clear from FIG. When performing linear quantization, since the quantization scale of an odd number can be used in the small range (1 to 8) of the value of the quantization scale code, the bit rate can be adjusted more finely.

  In the quantizer 10, when performing nonlinear quantization, a quantization scale having a value of 112 can be used corresponding to the maximum value 31 of the quantization scale code, and therefore linear quantization is executed. Compared to the case, the bit rate can be further reduced.

  Incidentally, in the quantizer 10, when nonlinear quantization is performed and the DCT coefficient is quantized, the range of the quantized coefficient value obtained as a result of the quantization scale code value is designated as 1. -2048 to +2047, which is equivalent to sending the DCT coefficient to the variable length encoder 4 without being quantized and encoding it.

  On the other hand, in the quantizer 10, when non-linear quantization is performed and a quantization scale code having a value of 31 is designated, the range of possible quantization coefficient values is -18 to +18. The minimum level of the image quality after the encoding that is prescribed is ensured.

However, the quantizer 10 uses the above-described linear quantization and nonlinear quantization because the DCT of the macroblock data when the coding mode set by the motion vector detector 3 is the intra coding mode is used. If for all of the AC component excluding the DC component coefficient, Ru Oh is for all DCT coefficients of the macroblock data of the intranet coding mode.

  Incidentally, in the quantizer 10, for the DC component of the DCT coefficient of the macroblock data in the intra coding mode, the bit accuracy after quantization of the DC component of the DCT coefficient in the intra coding mode specified in units of pictures. Quantize separately according to (quantization coefficient bit precision).

As the quantization coefficient bit precision of the DC component of the DCT coefficient in the intra coding mode, any one of 11 bits, 10 bits, 9 bits, and 8 bits is determined in units of pictures, and the quantization coefficient bit precision is In the case of 11 bits, this is equivalent to sending the DC component of the DCT coefficient supplied from the DCT unit 9 to the variable length encoder 4 as it is without being quantized by the quantizer 10 and encoding it.

  This describes the bit accuracy of the DCT coefficient defined in the MPEG2 system as 12 bits so far, but the DC component of the DCT coefficient does not take a negative value and does not require a sign bit. Because it is possible.

  In the quantizer 10, the DC component quantization scale values corresponding to the 11-bit, 10-bit, 9-bit, and 8-bit quantization coefficient bit precisions are 1, 2, 4, and 8, respectively. Using this quantization scale, linear quantization is performed on the DC component of the DCT coefficient in the intra coding mode. Incidentally, in the quantizer 10, when the direct current component of the DCT coefficient in the intra coding mode is quantized, if a fraction is generated, it is rounded by, for example, rounding to the nearest four to obtain an integer. .

  In this way, when the quantizer 10 quantizes the DCT coefficient, the quantizer 10 obtained as a result is sent to the variable length encoder 4 together with the value of the quantization scale used for the quantization.

  The variable length encoder 4 converts the quantization coefficient given from the quantizer 10 into a variable length code such as a Huffman code and sends it to the buffer 11.

  Here, the variable-length coding method for quantized coefficients defined by the MPEG2 system will be described. The variable-length encoder 4 performs one-dimensional information by zigzag scanning the quantized coefficients given from the quantizer 10. The low-frequency coefficient of the one-dimensional information is sequentially converted into a symbol string of zero run length + level value, and each symbol is encoded with reference to a variable-length code table prepared in advance from the low frequency. .

  By the way, the variable length encoder 4 does not prepare a variable length code table for all symbols in advance, and assigns an escape code of the variable length code table to a symbol with a low frequency of appearance. For a symbol to which a code is assigned, the zero run length and the level value are encoded as a fixed length code.

  Incidentally, in the variable length encoder 4, the fixed length code length for the zero run length is 6 bits, and the fixed length code length for the level value is 12 bits. This is the maximum zero run length is 63, This is because the range of values is the range of DCT coefficients, -2048 to +2047 (12-bit precision).

  The variable length encoder 4 also includes a quantization scale given from the quantizer 10 and coding mode information (intra coding (intraframe coding) mode, forward prediction coding given from the motion vector detector 3. The information indicating whether one of the mode, the backward prediction encoding mode, or the bidirectional prediction encoding mode is set) and the motion vector are also variable-length encoded, and the obtained encoded data is sent to the buffer 11.

  The buffer 11 temporarily stores the encoded data supplied from the variable-length encoder 4 to smooth the data amount, and outputs the encoded data as an encoded bit stream, for example, to a transmission path or recorded on a recording medium. To do.

  Further, the buffer 11 sends the data accumulation amount to the quantizer 10, and the quantizer 10 sets the quantization scale based on the data accumulation amount given from the buffer 11.

  That is, the quantizer 10 reduces the quantization coefficient data amount by relatively increasing the quantization scale when the buffer 11 is likely to overflow, and relatively decreasing the quantization scale when the buffer 11 is likely to underflow. Thus, the data amount of the quantization coefficient is increased. In this way, the quantizer 10 can prevent the overflow and underflow of the buffer 11 in advance.

  Incidentally, the quantizer 10 sends the quantization coefficient and the quantization scale to the inverse quantizer 12 in addition to the variable length encoder 4.

  The inverse quantizer 12 converts the quantized coefficient into a DCT coefficient by inversely quantizing the quantized coefficient given from the quantizer 10 according to the quantization scale similarly given from the quantizer 10, The DCT coefficient is sent to an IDCT (Inverse Discrete Cosine Transform) unit 13.

  The IDCT unit 13 performs inverse DCT processing on the DCT coefficient given from the inverse quantizer 12 and sends the obtained data (prediction residual (difference value data)) to the computing unit 8.

As described above, the computing unit 8 is provided with the same predicted frame image data as the predicted frame image data sent from the motion compensator 5 to the computing unit 7, and the data (prediction residual) provided from the IDCT unit 13. (Difference value data)) is added to the predicted frame image data to locally decode the original frame image data, and the obtained locally decoded frame image data is sent to the frame memory 6 for temporary storage.

  Incidentally, when the encoding mode is set to the intra encoding mode, the arithmetic unit 8 sends the data provided from the IDCT unit 13 as it is to the frame memory 6 as the locally decoded frame image data and temporarily stores it.

  In this way, the arithmetic unit 8 temporarily stores the locally decoded frame image data in the frame memory 6 so as to update it. As a result, the frame image data decoded over the entire frame (hereinafter referred to as this). (Referred to as decoded frame image data). The decoded frame image data is the same as the decoded frame image data obtained in the decoder described later.

  The decoded frame image data constructed in the frame memory 6 is then used as reference frame image data for frame image data that is inter-coded (forward prediction coding, backward prediction coding, bidirectional prediction coding). .

  Next, as a decoder to which the main profile main level of the MPEG2 system, which can decode the encoded data given from the encoder 1 described above with reference to FIG. 24, is configured as shown in FIG.

  As shown in FIG. 28, in the decoder 20, an encoded bit stream transmitted through a transmission path and received by a receiving device (not shown) or reproduced from a recording medium by a reproducing device (not shown). The encoded bit stream is taken into the buffer 21 and temporarily stored.

  An Inverse Variable Length Coding (IVLC) unit 22 reads an encoded bit stream stored in the buffer 21 as encoded data in a predetermined unit, and variable-length decodes the read encoded data. Thus, the encoded data is separated into motion vectors, encoding mode information, quantization scales, and quantization coefficients in units of macroblocks.

  Then, the variable length decoder 22 sends the motion vector and coding mode information out of the motion vector, coding mode information, quantization scale and quantization coefficient separated in macroblock units to the motion compensator 23, The quantization scale and the macroblock quantization coefficient are sent to the inverse quantizer 24.

  The inverse quantizer 24 inversely quantizes the quantization coefficient in units of macroblocks given from the variable length decoder 22 according to the quantization scale similarly given from the variable length decoder 22, and converts the obtained DCT coefficients into The data is sent to the IDCT unit 25.

  The IDCT unit 25 performs inverse DCT processing on the DCT coefficients in units of macroblocks supplied from the inverse quantizer 24 and sends the obtained data (prediction residual (difference value data)) to the calculator 26.

  The motion compensator 23 basically operates in the same manner as the motion compensator 5 of the encoder 1 described above with reference to FIG. 24. At this time, the motion compensator 23 converts the motion vector and the encoding mode information supplied from the variable length decoder 22. Based on this, the decoded frame image data that has been decoded and stored in the frame memory 27 is read out, and the read out decoded frame image data is sent to the computing unit 26 as predicted frame image data.

  Therefore, the arithmetic unit 26 adds the decoded frame image data by adding the data (prediction residual (difference value)) given from the IDCT unit 25 and the corresponding predicted frame image data given from the motion compensator 23. This is generated and sent to and stored in the frame memory 27, and as the reproduced image data of the original frame image, for example, it is sent to a predetermined display, so that a moving image based on the reproduced image data is displayed on the display. Is displayed.

  Incidentally, when the data supplied from the IDCT unit 25 is intra-encoded, the arithmetic unit 26 sends the data as it is to the frame memory 27 as decoded frame image data for storage. The frame memory 27 uses the stored decoded frame image data as reference frame image data for encoded data in units of macroblocks to be decoded thereafter.

  In the MPEG1 and MPEG2 systems, the B picture is not used as the reference frame image data. Therefore, in the encoder 1 (FIG. 24) and the decoder 20 (FIG. 28), the frame memories 6 (FIG. 24) and 27 (FIG. 28), respectively. ) Is not stored.

  Incidentally, the encoder 1 and the decoder 20 described above with reference to FIGS. 24 and 28 comply with the MPEG2 standard. At present, as a new compression and coding system that replaces the MPEG2 system, ISO 4 is called MPEG4 that compresses and encodes images in units of video objects, which are objects such as objects, according to ISO-IEC / JTC1 / SC29 / WG11. Standardization work for compression coding is underway.

  That is, in the MPEG2 system, it is stipulated that only an 8-bit precision input image is compression-encoded, and when the 8-bit precision image is used at home or the like, it is expressed with a practically sufficient image quality. can do.

  However, in the MPEG2 system, for use in a professional field such as a broadcasting station or video production, or in a field such as a digital cinema that will become widespread in the future, for example, the broadcasting station has an image quality with an accuracy of 10 bits or more. As digital cinema demands a maximum of 14-bit accuracy, it is difficult to cope with image quality that is higher than 8 bits.

  For this reason, recently, the MPEG4 system is used to input not only 8-bit accuracy but also higher N-bit precision image data such as 10-bit, 12-bit, 14-bit precision, etc. It is considered to provide a moving image having higher accuracy image quality by decoding.

  However, in the encoder and decoder to which the MPEG4 system is applied, although the accuracy of the input image can be expanded from 8 bits to N bits, the configuration of the quantizer and variable length encoder in the encoder and the inverse quantization in the decoder And the structure of the variable-length decoder are not extended to obtain high-precision data according to the N-bit precision of the input image, so a very high-quality image such as a broadcasting station or digital cinema is required. There were still insufficient problems to use it in the field.

  The present invention has been made in consideration of the above points, and intends to propose an encoding apparatus and method, a decoding apparatus, and a method thereof that can obtain an image having substantially the same image quality as the input image. It is.

  In order to solve such a problem, in the present invention, differential value data to be subjected to discrete cosine transform processing is generated from input image data having arbitrary input bit precision in a predetermined data unit, and the generated differential value data is arbitrarily set. The discrete cosine transform processing is performed according to the output bit precision, and a discrete cosine transform coefficient is generated, and a predetermined quantization scale is expanded based on at least one of the output bit precision and the input bit precision of the discrete cosine transform coefficient Then, after the discrete cosine transform coefficient is quantized using the expanded quantization scale to generate a quantized coefficient, the quantized coefficient is compressed and encoded using a predetermined variable length coding table.

  Therefore, it is possible to greatly reduce the deterioration of the image quality when the input image data is compression-encoded.

  In the present invention, when compressing and encoding a quantized coefficient, an arbitrary variable length is selected from a plurality of types of variable length encoding tables based on at least one of input bit precision, output bit precision, and quantization scale. An encoding table is selected, and the quantized coefficients are compression encoded using the selected variable length encoding table.

  Therefore, when the input image data is compression-encoded, it is possible to almost certainly prevent the image quality from deteriorating.

  Furthermore, in the present invention, the quantized coefficients that have been compression-coded are decoded using a predetermined variable-length coding table to generate quantized coefficients, and arbitrary input image data that is the basis of the quantized coefficients is generated. The predetermined quantization scale is expanded based on at least one of the input bit precision of the output and the output bit precision of the discrete cosine transform coefficient set in advance, and the quantized coefficient using the expanded quantization scale Was dequantized to generate discrete cosine transform coefficients.

  Therefore, when the original input image is generated by decoding the quantized coefficients that have been compression-encoded, it is possible to greatly reduce the deterioration of the image quality.

  Furthermore, in the present invention, when decoding the quantized coefficients that have been compression-encoded, a plurality of types of variable-length coding tables based on at least one of input bit precision, output bit precision, and quantization scale An arbitrary variable length coding table is selected from the above, and the quantized coefficients that have been compression-coded are decoded using the selected variable length coding table.

  Therefore, when the original input image is generated by decoding the quantized coefficients that have been compression-encoded, it is possible to almost certainly prevent the image quality from deteriorating.

  According to the present invention, differential value data to be subjected to discrete cosine transform processing is generated from input image data having arbitrary input bit precision in a predetermined data unit, and the generated differential value data is set to arbitrarily set output bit precision. In response, a discrete cosine transform process is performed to generate a discrete cosine transform coefficient, a predetermined quantization scale is expanded based on at least one of the output bit precision and the input bit precision of the discrete cosine transform coefficient, and the expanded After quantizing the discrete cosine transform coefficient using the quantization scale to generate the quantized coefficient, the quantized coefficient is compressed and encoded using a predetermined variable length coding table, so that the input image data It is possible to greatly reduce the deterioration of the image quality when compression encoding is performed.

  In addition, when compressing and encoding a quantization coefficient, an arbitrary variable-length encoding table is selected from a plurality of types of variable-length encoding tables based on at least one of input bit precision, output bit precision, and quantization scale. Since the quantization coefficient is compression-encoded using the selected variable-length encoding table, when the input image data is compression-encoded, it is almost certainly prevented that the image quality is deteriorated. be able to.

  Furthermore, the quantized coefficient that has been compression-coded is decoded using a predetermined variable length coding table to generate a quantized coefficient, and arbitrary input bit precision of predetermined input image data that is the basis of the quantized coefficient , Expanding a predetermined quantization scale based on at least one of the output bit precision of the discrete cosine transform coefficient set in advance, and performing an inverse quantization process on the quantization coefficient using the expanded quantization scale By generating discrete cosine transform coefficients, it is possible to greatly reduce the degradation of image quality when the original input image is generated by decoding the compression-coded quantized coefficients. Can do.

  Further, when decoding the quantized coefficient that has been compression-encoded, any variable length can be selected from a plurality of types of variable-length coding tables based on at least one of input bit precision, output bit precision, and quantization scale. By selecting an encoding table and decoding the quantized coefficient that has been compression-encoded using the selected variable-length encoding table, the quantized coefficient that has been compression-encoded is decoded and the original When the input image is generated, it is possible to almost certainly prevent the image quality from deteriorating.

  Therefore, according to the encoding device and the decoding device capable of preventing the deterioration of the image quality in this way, the image quality of the image generated by compressing and encoding the input image and then decoding is almost equal to the image quality of the input image. can do.

  In addition, when compressing and encoding an input image with any input bit precision, a quantization method and a variable length coding method that make it possible to vary the output bit precision of the discrete cosine transform coefficient according to the required image quality Can be provided.

It is a block diagram which shows one Embodiment of the structure of the encoder by this invention. It is a basic diagram which shows the structure of VOP. It is a basic diagram with which it uses for description of the division | segmentation of the quantization area with respect to the change of the bit precision of a DCT coefficient. It is a basic diagram with which it uses for description of the division | segmentation of the quantization area with respect to the change of the bit precision of a DCT coefficient. It is a basic diagram with which it uses for description of the relationship between a quantization scale code and a quantization scale in case the bit precision of a DCT coefficient is N + 1 + (alpha) bit. It is a block diagram with which it uses for description of switching of the variable-length encoding table according to the bit precision of the pixel of an input image. It is a block diagram with which it uses for description of switching of the variable-length encoding table according to a DCT coefficient. It is a block diagram with which it uses for description of the bit precision of the pixel of an input image, and switching of the variable length encoding table according to a DCT coefficient. It is a block diagram with which it uses for description of switching of the variable-length encoding table according to a quantization scale. It is a basic diagram with which it uses for description of the relationship between a quantization scale and the bit precision of a quantization coefficient. It is a block diagram with which description of switching of the variable-length encoding table according to the combination of the bit precision of the pixel of an input image, a DCT coefficient, and a quantization scale is provided. It is a block diagram which shows one Embodiment of the structure of the decoder by this invention. It is a basic diagram which shows the syntax of VOL. It is a basic diagram which shows a VOP layer. It is a basic diagram which shows a macroblock layer. It is a basic diagram with which it uses for description of the relationship between the quantization coefficient precision of the direct current | flow component of a DCT coefficient, and a quantization scale. It is a basic diagram which shows the syntax of a block layer. It is a basic diagram which shows the variable length coding table used for 3D variable length coding. It is a basic diagram which shows the variable length coding table used for 3D variable length coding. It is a basic diagram which shows the variable length coding table used for 3D variable length coding. It is a basic diagram with which it uses for description of encoding of a level length and a run length. It is a basic diagram with which it uses for description of switching of the variable-length encoding table by bits_per_pixel, dct_precision, and quantizer_scale. It is an approximate line figure used for explanation of change of fixed length code length of a level value by bits_per_pixel, dct_precision, and quantizer_scale. It is a block diagram which shows the structure of the encoder to which MPEG2 system is applied. It is a basic diagram with which it uses for description of the linear and nonlinear quantization system of MPEG2. It is a basic diagram which shows the relationship between the quantization scale code in an MPEG2 system, and a quantization scale. It is a basic diagram which shows the relationship between the quantization scale code in an MPEG2 system, and a quantization scale. It is a block diagram which shows the structure of the decoder to which the MPEG2 system is applied.

  The best mode for carrying out the invention (hereinafter also referred to as an embodiment) will be described below with reference to the drawings.

(1) Principle Currently, in the encoder, which is applied MPEG4 system, although may more accurately than bit precision (color tone) of 8-bit precision of the input image supplied, the number of bits of precision in their Assuming N, when the bit accuracy of the difference value data (that is, the output data of the arithmetic unit) input to the DCT unit is N + 1 bits, the bit accuracy of the DCT coefficient obtained from the DCT unit is 3 bits from the N + 1 bits. It is specified to increase to an accuracy of N + 4 bits.

  In the encoder, since the limit of the image quality of the moving image after compression encoding (after quantization) is determined by the bit accuracy of the DCT coefficient, the image quality of the moving image after compression encoding is determined according to the bit accuracy of the input image. In order to achieve high accuracy, it is necessary to expand the bit accuracy of the DCT coefficient obtained from the DCT unit.

  Therefore, in the encoder according to the present invention, the number of bits increased by the DCT operation is α, and the bit accuracy of the DCT coefficient obtained from the DCT unit is N + 1 + α bits. The value of α can be arbitrarily set according to the image quality. Incidentally, in the MPEG4 system, α is fixed as 3.

  In the present invention, the encoder can input an input image with N-bit accuracy, and the bit accuracy of the DCT coefficient obtained from the DCT unit can be set arbitrarily so that α can be arbitrarily set, from N + 4 bits to N + 1 + α bits. The configuration of the quantizer and variable length encoder provided in the subsequent stage of the DCT unit and the configuration of the inverse quantizer and inverse variable length encoder of the decoder paired with the encoder Extended according to bit precision.

  In practice, in the present invention, a quantization method applied to an arbitrary N + 1 + α-bit precision DCT coefficient for an input image with an arbitrary N-bit accuracy is a linear quantization method or a nonlinear quantization method defined by the MPEG2 method. It is realized by expanding.

  Also, in the present invention, when the encoding mode of macroblock data in an arbitrary N-bit precision input image is the intra coding mode, the quantization method applied to the DC component of the DCT coefficient with N + α-bit precision is MPEG2. This is realized by extending the quantization method of the DC component of the DCT coefficient of the intra coding mode defined in the method.

  By the way, in the present invention, the expansion of the quantization method is made compatible with the quantization method of the MPEG2 method so that it is the same as the MPEG2 quantization method under the conditions of N = 8 and α = 3. .

  Further, in order to adjust the bit rate relatively finely even in the case of a relatively high bit rate, the step size of the quantization scale is reduced in a range where the value of the quantization scale code is relatively small.

  Furthermore, regardless of the value of N or α, no matter how large the value of N or α, the compression encoding can be performed up to the bit rate equivalent to N = 8 and α = 3 in the MPEG2 system. In addition, when the maximum value of the quantization scale is used, the range of values that the quantization coefficient can take is made equal.

  In this way, in the present invention, even when a high-precision input image is processed to obtain a high-quality image, it is possible to prevent the encoder and the decoder from causing problems in processing. Yes.

  By the way, in an encoder to which the MPEG4 system is applied at present, a DCT coefficient with N + 4 bits precision is quantized by a quantizer, and a single variable length code is obtained by the variable length encoder for the obtained quantized coefficient. Although it is specified that variable length coding is performed using a quantization table, since the statistical property of establishment of the occurrence of a quantized coefficient varies depending on N representing the bit accuracy of the input image, all input that can be set Using the same variable length coding table for the bit accuracy (N) of the image is not appropriate in terms of compression coding efficiency.

  Further, if the bit precision of the DCT coefficient is expanded to N + 1 + α bits, the statistical property of establishment of the quantization coefficient changes depending on α, and therefore the same variable length code is set for all α that can be set. The use of a conversion table is not appropriate in terms of compression encoding efficiency.

  Furthermore, if the bit precision of the DCT coefficient is expanded to N + 1 + α bits, the quantization scale needs to be expanded accordingly. However, even if the statistical nature of the occurrence of the quantization coefficient depends on the quantization scale, Therefore, using the same variable length coding table for all quantization scales is not appropriate for the efficiency of compression coding.

  Therefore, in the present invention, the variable length encoder switches the appropriate variable length coding table according to the bit accuracy of the input image, the bit accuracy of the DCT coefficient, and the quantization scale (that is, the quantization coefficient bit accuracy). In order to be able to use it, it was extended with a plurality of types of variable length coding tables.

  Thereby, in the present invention, the variable length coding table applied to the quantization coefficient of the DCT coefficient of arbitrary bit precision in the bit precision of the arbitrary input image is converted into the bit precision of the input image and the bit precision of the DCT coefficient, It is possible to select an appropriate one according to the range that the quantization coefficient can take (that is, the value of the quantization scale). Thus, compression encoding can be performed more efficiently than when a variable-length encoding table cannot be selected. Has been made to get.

(2) Configuration of Encoder According to this Embodiment In FIG. 1, reference numeral 30 denotes an encoder to which the present invention is applied as a whole. Image data of an input image supplied from the outside is taken into the frame memory 31, and The image data is stored as a VOP (Video Object Plane).

  Here, as shown in FIG. 2, the VOP is a picture having shape information, data representing the shape information, size data FSZ_B of the VOP, offset data FPOS_B representing the position of the VOP in the frame, It consists of texture data representing VOP image data.

  Incidentally, in the present invention, since the shape information may be ignored for the VOP, the VOP is considered to be the same as a picture defined by the MPEG2 system or the like.

  In the encoder 30, the number of bits representing the accuracy (color gradation) of the pixel value of the image data stored in the frame memory 31 is N, and the possible range of the pixel value is 0 to 2N-1. Become. The values that N can take include values such as 8, 10, 12, and 14 that are normally used. In the present invention, any positive integer value can be taken.

  Then, the motion vector detector 32 (FIG. 1) detects a motion vector for each macroblock with respect to the VOP stored in the frame memory 31.

  Here, since the size and position of the VOP change depending on the time (picture), when detecting the motion vector, an absolute coordinate system serving as a reference for the detection is set, and the motion in the absolute coordinate system is determined. It needs to be detected.

  Therefore, the motion vector detector 32 uses the frame coordinate system as a reference absolute coordinate system, and based on the size data FSZ_B and the offset data FPOS_B, the absolute coordinate system uses the VOP to be encoded and the reference frame image data. A VOP is arranged and a motion vector is detected.

  Incidentally, the motion vector detector 32 includes a shape information encoder 50, a variable length encoder 36, a motion compensator 42, a quantizer 35 and an inverse quantizer together with the encoding mode information for the detected motion vector (MV). 38. Even when motion compensation is performed, since it is necessary to detect a motion vector in the reference absolute coordinate system as described above, the motion compensator 42 is supplied with size data FSZ_B and offset data FPOS_B. Yes.

  The calculator 33 has the same macro as that in the frame image data read out from the frame memory 31 by the motion vector detector 32, as in the calculator 7 (FIG. 24) of the encoder 1 (FIG. 24) described above with reference to FIG. 24. Block data is supplied, the difference between the macroblock data and the predicted frame image data given from the motion compensator 42 is calculated, and the obtained difference value is sent to the DCT unit 34 as difference value data.

  Here, the computing unit 33 computes the difference between the macroblock data read from the frame memory 31 and the corresponding predicted frame image data given from the motion compensator 42 to generate difference value data, thereby obtaining the difference. The range that the value data can take can be doubled from the range that the pixel value of the input image can take, such as-(2N-1) to 2N-1.

  Therefore, the computing unit 33 can set the bit accuracy of the difference value data sent to the DCT unit 34 to the accuracy of N + 1 bits obtained by increasing the bit accuracy of the pixel value of the input image by 1 bit.

  Similarly to the motion compensator 5 (FIG. 24) of the encoder 1 (FIG. 24) described above with reference to FIG. 24, the motion compensator 42 receives the arithmetic units 33 and 40 when the coding mode information is the intra coding mode. Do not send predicted frame image data. Accordingly, the calculator 33 (the same applies to the calculator 40) sends the macroblock data read from the frame memory 31 to the DCT unit 34 as it is without performing any particular processing.

  The DCT unit 34 performs DCT processing on the difference value data given from the computing unit 33 in units of blocks of 8 pixels × 8 lines, and sends the obtained DCT coefficients to the quantizer 35.

  Here, the DCT coefficient obtained from the DCT unit 34, that is, the bit accuracy of the input value to the quantizer 35 is N + 1 + α bits increased by α bits from N + 1 bits, and the range of values of the DCT coefficients can be −2N + α to 2N + α. -1. Incidentally, in the conventional MPEG2 system, α in the DCT coefficient is defined as 3, but in the present invention, an arbitrary positive integer value can be applied as α.

  The quantizer 35 quantizes the DCT coefficient given from the DCT unit 34 and varies the obtained quantized coefficient, similarly to the quantizer 10 (FIG. 24) of the encoder 1 (FIG. 24) described above with reference to FIG. The data is sent to the long encoder 36 and the inverse quantizer 38.

  In practice, the quantizer 35 performs linear quantization on a 12-bit precision DCT coefficient obtained from an 8-bit precision input image according to the MPEG2 standard, like the quantizer 10 described above with reference to FIG. Nonlinear quantization is extended to linear quantization for DCT coefficients with N + 1 + α-bit accuracy obtained from an input image with N-bit accuracy and nonlinear quantization.

  First, in the quantizer 10 described above with reference to FIG. 24, a quantization scale code having a 5-bit length (1 to 31) is assigned to a 12-bit precision DCT coefficient in accordance with the MPEG2 system. In the unit 35, the bit length of the quantization scale code is also increased or decreased according to the increase or decrease of the bit precision of the DCT coefficient.

  That is, in the quantizer 35, for example, when α = 5 for an 8-bit precision input image, the bit precision of the DCT coefficient is 2 bits higher than that of α = 3 in the MPEG2 system. The bit length of the generalized scale code is also increased by 2 bits to 7 bits. At this time, the range that the quantization scale code can take is 1 to 127.

  For example, when α = 6 for an input image with 12-bit precision, the quantizer 35 increases N by 4 bits from the input bit precision in the MPEG2 system, and from α = 3 in the MPEG2 system. In addition, since the bit precision of the DCT coefficient increases by 3 bits and increases by 7 bits, the bit length of the quantization scale code is increased by 7 bits to 12 bits. Incidentally, at this time, the range of values that the quantization scale code can take is 1 to 4095.

  When the variable of the bit length of the quantization scale code in the quantizer 35 is generalized, the bit length of the quantization scale code is N + α−6 for any N and α, and the possible range of the value Becomes 1-2N + α-6-1.

  Incidentally, when the bit precision of the DCT coefficient becomes smaller than 12 bits, the bit length of the quantization scale code becomes shorter than 5 bits.

  Next, a method for mapping the quantization scale code whose number of bits has been determined to the quantization scale in the quantizer 35, that is, a method of linear quantization and nonlinear quantization will be described below.

  In the quantizer 35, when executing the linear quantization, the quantization scale code is obtained by doubling the quantization scale code as in the MPEG2 standard, and the quantization scale used for the linear quantization is determined. The value will simply extend upwards.

  In the quantizer 35, the bit accuracy of the DCT coefficient obtained by executing the linear quantization, the maximum value of the DCT coefficient, the range of the quantization scale code, the range of the quantization scale, and the quantized scale An example of the range of the maximum value of the quantization coefficient is as follows.

  That is, under the conditions of N = 8 and α = 3 (similar to the MPEG2 system), the accuracy of the DCT coefficient is 12 bits, the maximum value of the DCT coefficient is 2047, the quantization scale code range is 1 to 31, and the quantization scale. Is 2 to 62, and the range of the maximum value of the quantization coefficient is 33 to 1023.

  Under the conditions of N = 8 and α = 6, the precision of the DCT coefficient is 15 bits, the maximum value of the DCT coefficient is 16383, the quantization scale code range is 1 to 255, and the quantization scale range is 2 to 510. The range of the maximum value of the quantization coefficient is 33 to 8191.

  Further, under the conditions of N = 12, α = 6, the accuracy of the DCT coefficient is 19 bits, the maximum value of the DCT coefficient is 262143, the quantization scale code range is 1 to 4095, and the quantization scale range is 2 to 8190. The range of the maximum value of the quantization coefficient is 33 to 131071.

  When this is generalized, under the conditions of N and α, the accuracy of the DCT coefficient is N + 1 + α bits, the maximum value of the DCT coefficient is 2N + α-1, the range of the quantization scale code is 1 to 2N + α-6-1, the quantum The range of the quantization scale is 2 to 2N + α-5-2, and the range of the maximum value of the quantization coefficient is 33 to 2N + α-1-1.

  In this way, in the quantizer 35, when linear quantization is extended, the quantizer 35 is compatible with the MPEG2 linear encoding method, and the maximum value of the quantization scale regardless of the values of N and α. It is clear that compression coding is possible up to a bit rate equivalent to that of the MPEG2 system.

  By the way, the generalized linear quantization method has the following equation (4) when the quantization scale code is qsc and the quantization scale is qs.

It is represented by

  On the other hand, in the quantizer 35, when performing linear quantization, the value of the quantization scale is simply increased upward by increasing the quantization scale code. If the method is not correctly defined, compatibility with the achievable bit rate and MPEG2 nonlinear quantization may not be maintained.

  Therefore, in this quantizer 35, the nonlinear quantization method is extended so as to satisfy the purpose of realizing a desired bit rate and maintaining compatibility with the nonlinear quantization of the MPEG2 method.

  That is, in the quantizer 35, when nonlinear quantization is performed, nonlinear quantization used for generating DCT coefficients with 12-bit accuracy in the MPEG2 system described above with reference to FIG. 26 in accordance with the increase in the bit length of the quantization scale code. Make the step size of the curve fine at regular intervals.

For example, in the quantizer 35, when the bit precision of the DCT coefficient becomes 13 bits, the step size of the quantization scale code is divided into 2 ¹ = 2 as shown in FIG.

Thereby, in the quantizer 35, +2 ¹ −1 is added to the doubled number of steps of the quantized scale code to obtain 63, and the smaller quantized scale code is assigned in order from the smallest, Double the value on the quantization scale axis.

  Thus, in the quantizer 35, the quantization scale corresponding to the new quantization scale code becomes the quantization scale in the nonlinear quantization for the DCT coefficient with 13-bit accuracy.

Further, in the quantizer 35, for example, when the bit accuracy of the DCT coefficient becomes 14 bits, the step size of the quantization scale code is divided into 2 ² = 4 as shown in FIG. .

Thus, in the quantizer 35, +2 ² −1 is added to 4 times the number of increments of the quantization scale code to 127, and the smaller quantization scale code is assigned in order from the smallest, and the quantum The value on the scale axis is quadrupled.

  Thus, in the quantizer 35, the quantization scale corresponding to the new quantization scale code becomes the quantization scale in the non-linear quantization for the DCT coefficient with 14-bit accuracy.

  In the quantizer 35, when the definition of the nonlinear quantization is generalized, when the bit accuracy of the DCT coefficient becomes N + 1 + α bit accuracy, the quantization scale code increment is set to 2N + α as shown in FIG. -Divide into 11 pieces at equal intervals.

  Thereby, in the quantizer 35, the number of increments of the quantization scale code is 2N + α−11 times + 2N + α−11−1 (2N + α−6-1), and the smaller quantization scale codes are assigned in order from the smallest. At the same time, the value on the quantization scale axis is multiplied by 2N + α−11.

  By the way, 2N + α-11-1 is added to the number of steps, but this is to make the total number of steps (2N + α-6-1), and this added step is the maximum of the quantization scale code. Added in the value direction.

  Thus, in the quantizer 35, the quantization scale corresponding to the new quantization scale code becomes the quantization scale in the non-linear quantization for the DCT coefficient with N + 1 + α bit accuracy.

  By the way, in the quantizer 35, when the bit accuracy of the DCT coefficient becomes smaller than 12 bits, the increments of the quantization scale code described above with reference to FIG. 26 are integrated accordingly.

  That is, when the bit precision of the DCT coefficient is 11 bits, the quantization scale code has a length of 4 bits, and two adjacent ones of the quantization scale code steps described above with reference to FIG. 26 may be integrated. Further, when the bit accuracy of the DCT coefficient is 10 bits, the quantization scale code is 3 bits long, and it is only necessary to integrate four adjacent ones of the quantization scale code steps described above with reference to FIG.

  In the quantizer 35, the bit precision of the DCT coefficient obtained by executing the nonlinear quantization, the maximum value of the DCT coefficient, the range of the quantization scale code, the range of the quantization scale, and the quantized scale An example of the range of the maximum value of the quantization coefficient is as follows.

  That is, under the conditions of N = 8 and α = 3 (similar to the MPEG2 system), the accuracy of the DCT coefficient is 12 bits, the maximum value of the DCT coefficient is 2047, the quantization scale code range is 1 to 31, and the quantization scale. 1 to 112, and the maximum value of the quantization coefficient is 18 to 2047.

  Under the conditions of N = 8 and α = 6, the accuracy of the DCT coefficient is 15 bits, the maximum value of the DCT coefficient is 16383, the quantization scale code range is 1 to 255, and the quantization scale range is 1 to 952. The range of the maximum value of the quantization coefficient is 18 to 16383.

  Further, under the conditions of N = 12, α = 6, the accuracy of the DCT coefficient is 19 bits, the maximum value of the DCT coefficient is 262143, the quantization scale code range is 1 to 4095, and the quantization scale range is 1 to 15352. The range of the maximum value of the quantization coefficient is 18 to 262143.

  When this is generalized, under the conditions of N and α, the accuracy of the DCT coefficient is N + 1 + α bits, the maximum value of the DCT coefficient is 2N + α-1, the range of the quantization scale code is 1 to 2N + α-6-1, the quantum The range of the quantization scale is 1 to 120 × 2N + α-11-8, and the range of the maximum value of the quantization coefficient is 18 to 2N + α-1.

  In this way, in the quantizer 35, when the nonlinear quantization is extended, the quantizer 35 is compatible with the MPEG2 nonlinear encoding method, and the quantization step size is increased at a relatively high bit rate. It is clear that when the maximum value of the quantization scale is used regardless of the values of N and α, compression encoding is possible up to a bit rate equivalent to that of the MPEG2 system. .

  Incidentally, the generalized non-linear quantization method has the following equation (5) when the quantization scale code is qsc and the quantization scale actually used for quantization is qs.

It is represented by

  In the quantizer 35, since the quantization scale code is represented by N + α-6 bits, assuming that the value consisting of the upper 2 bits is k and the value consisting of the remaining N + α-8 bits is j, (5 ) To (6)

Can be expressed as Incidentally, the possible range of the value of k is 0-3, and the possible range of the value of j is 0-2N + α-8-1. And k = j = 0 is prohibited.

  By the way, in the quantizer 35, when the coding mode of the macroblock data is set to the intra coding mode, the DC component of the DCT coefficient is obtained from an input image with 8-bit accuracy according to the MPEG2 system. The bit accuracy specification after quantizing the DC component of the 12-bit precision DCT coefficient obtained by quantizing the DC component of the N + 1 + α bit precision DCT coefficient obtained from the N-bit precision input image Extend to

That is, in the quantizer 35, when extending the bit precision of the quantization coefficients, 11 bits for the DC component of the DCT coefficients according the 11-bit precision M PEG2 scheme, 10 bits, 9 bits, 8-bit precision Four types of quantization coefficients could be selected for each picture.

  In the quantizer 35, the range that the quantization coefficient can take in each bit precision is 0 to 2047, 0 to 1023, 0 to 511, and 0 to 255, and the direct current is increased according to the increase or decrease of the bit precision of the DCT coefficient. The type of bit precision of the component quantization coefficient is also increased or decreased.

  In this case, in the quantizer 35, for example, when α = 5 for an input image with 8-bit accuracy, the bit accuracy of the DCT coefficient is 2 bits more than α = 3 in the MPEG2 system. The bit precision type of the quantized coefficient is also increased by two in the direction of increasing the bit precision to 13 bits, 12 bits, 11 bits, 10 bits, 9 bits, and 8 bits.

  In the quantizer 35, the possible range of the quantization coefficient with each bit precision is 0 to 8191, 0 to 4095, 0 to 2047, 0 to 1023, 0 to 511, and 0 to 255. That is, this is because the quantizer 35 can directly send the DCT coefficient as a quantized coefficient to the variable length coder 36, and even if the bit precision of the DCT coefficient increases, the bit rate is about the same as the MPEG2 system. This indicates that compression encoding is possible up to (quantization coefficient is 8 bit precision).

  In the quantizer 35, for example, when α = 6 for an input image with 12-bit precision, the bit precision of the input image in the MPEG2 system is increased by 4 bits, and α = 3 in the MPEG2 system. Since the bit precision of the DCT coefficient is increased by 3 bits to 7 bits in total, the bit precision type of the quantization coefficient is also increased by 7 in the direction of increasing the bit precision, 18 bits, 17 bits. 16 bits, 15 bits, 14 bits, 13 bits, 12 bits, 11 bits, 10 bits, 9 bits, and 8 bits.

  In the quantizer 35, the possible range of the quantization coefficient with each bit precision is 0 to 262143, 0 to 131071, 0 to 65535, 0 to 32767, 0 to 16383, 0 to 8191, and 0 to 4095. 0-2047, 0-1023, 0-511, 0-255. That is, it is possible to send the DCT coefficient directly from the quantizer 35 to the variable length coder 36 as a quantized coefficient, and even if the bit accuracy of the DCT coefficient increases, the bit rate (about the same as the MPEG2 system) This indicates that compression encoding is possible up to a quantization coefficient of 8 bits accuracy).

  By the way, in the quantizer 35, when the quantization accuracy of the DC component of the DCT coefficient is generalized, the bit accuracy of the DC component of the DCT coefficient is set to N + α bits (including the sign bit) with respect to the input image of N bit accuracy. N + 1 + α bit precision), the bit precision types of quantization coefficients are N + α bits, N + α-1 bits, N + α-2 bits,..., 9 bits, 8 bits.

  In the quantizer 35, the range that the quantization coefficient can take with each bit precision is 0-2N + α-1, 0-2N + α-1-1, 0-2N + α-2-1, ..., 0-511. 0 to 255.

  That is, it is possible to send the DCT coefficient directly from the quantizer 35 to the variable length coder 36 as a quantized coefficient, and even if the bit accuracy of the DCT coefficient increases, the bit rate (about the same as the MPEG2 system) This indicates that compression encoding is possible up to a quantization coefficient of 8 bits accuracy).

  In this way, in the quantizer 35, when the DCT coefficient DC component quantization method is expanded, the quantizer 35 is compatible with the DCT coefficient DC component quantization method according to the MPEG2 method, and has N and α. It is clear that compression coding can be performed up to a bit rate equivalent to that of the MPEG2 system when the minimum value of the quantization coefficient accuracy is used regardless of the value. At this time, the quantizer 35 does not need to take into consideration that the step size of the quantization scale is fine in the case of a relatively high bit rate.

  Incidentally, the inverse quantizer 38 (FIG. 1) inversely quantizes the DCT coefficient (that is, the quantized coefficient) quantized in units of blocks of 8 pixels × 8 lines provided from the quantizer 35, and generates an IDCT unit 39. To send.

  Here, when the quantization coefficient is inversely quantized, the inverse quantizer 38 uses a quantization scale code having the same value as that of the quantizer 35 described above and uses the same linear quantum as that of the quantizer 35. A quantization scale is calculated using the same method of quantization and nonlinear quantization, and inverse quantization is performed using the calculated quantization scale.

  Also, the DCT DC coefficient inverse quantization method when the macroblock prediction mode is the intra coding mode performs the reverse of the DCT DC coefficient quantization method.

  The IDCT unit 39 operates in the same manner as the IDCT unit 39 in FIG. 24, and performs DCT processing on the DCT coefficient inversely quantized by the inverse quantizer 38 and outputs the DCT coefficient to the computing unit 40.

In addition to the output data of the IDCT unit 39, the calculator 40 is supplied with the same data as the predicted image supplied to the calculator 33 from the motion compensator 42. The computing unit 40 locally decodes the original frame image data by adding the output data (prediction residual (difference data)) of the IDCT unit 39 and the predicted image data from the motion compensator 42, and is obtained. The local decoded frame image data (local decoded image data) is sent to the padding processor 51 as texture information.

  However, when the prediction mode is intra coding, the computing unit 40 sends the output data given from the IDCT unit 39 as it is to the padding processor 51 as texture information consisting of locally decoded frame image data.

  On the other hand, a shape signal (key signal), size data FSZ_B, and offset data FPOS_B are supplied to the shape information encoder 50 from the outside, and a motion vector and coding mode information are given from the motion vector detector 32.

  Then, the shape information encoder 50 encodes the shape information in accordance with the MPEG4 standard, and sends the encoded shape information to the shape information decoder 52 and the variable length encoder 36.

  The shape information decoder 52 performs local decoding of the encoded shape information given from the shape information encoder 50, and sends the data to the padding processor 51 and the variable length encoder 36.

  The padding processor 51 performs a padding process on the texture information in accordance with the MPEG4 standard, and sends the padded texture information to the frame memory 41 for storage.

  Thus, the texture information stored in the frame memory 41 is used as predicted frame image data for an image that is subsequently inter-coded (forward prediction coding, backward prediction coding, bidirectional prediction coding).

That is, the motion compensator 42, the image data stored in the frame memory 41, before the image data used for image or backward prediction is used in the direction prediction, the output to Rukoto.
On the other hand, the motion compensator 42 is given from the motion vector detector 32 to the image specified by the motion compensation reference image instruction signal (texture information consisting of a locally decoded image stored in the frame memory 41). based on the coding mode information and motion vector to generate a prediction frame image data by a score facilities motion compensation, and sends this to the computing unit 33 and 40.

  That is, the motion compensator 42 is the block of the block that is currently outputting the read address of the frame memory 41 to the computing unit 33 only in the forward predictive coding mode, backward predictive coding mode, and bidirectional predictive coding mode. The position is shifted from the position corresponding to the position by the amount corresponding to the motion vector, and the corresponding prediction frame image data used for forward prediction encoding or backward prediction encoding is read from the frame memory 41.

  Then, the motion compensator 42 sends the predicted frame image data to the calculators 33 and 40.

  Incidentally, in the bidirectional predictive coding mode, the motion compensator 42 reads both prediction frame image data used for forward prediction coding and backward prediction coding, for example, calculates an average value of the read prediction frame image data. The prediction data is sent to the calculators 33 and 40.

Thus, calculator 33, a predetermined block (macroblock) units of the frame image data read out from the frame memory 3 1 subtracts the predictive image data corresponding given from the motion compensator 42, thus, the differential value data Generate.

  In addition, in the forward predictive coding mode, the backward predictive coding mode, and the bidirectional predictive coding mode, the computing unit 40 is differentiated by the predicted frame image data in addition to the predicted frame image data provided from the motion compensator 42. The difference value data is given from the inverse DCT unit 39.

Thus, operator 40 adds the predicted frame image data given the difference value data from the motion compensator 42, thereby performs local decoding, generates the texture information.

  Incidentally, this texture information is locally decoded frame image data, which is exactly the same as the image data obtained by decoding with a decoder described later. As described above, the forward prediction is performed for the next processed image. It is stored in the frame memory 41 as predicted frame image data used when performing the coding mode, backward prediction coding mode, and bidirectional prediction coding mode.

  In addition, when the encoding mode is the intra encoding mode, the arithmetic unit 40 sends the image data as it is to the padding processor 51 as texture information because the image data itself is supplied from the inverse DCT unit 39.

  On the other hand, the variable length encoder 36 is supplied with the quantization coefficient, the quantization scale code, the motion vector, and the encoding mode information, as well as the size data FSZ_B and the offset data FPOS_B. Therefore, the variable length encoder 36 performs variable length encoding on all of these data.

  Here, the configuration of the variable-length encoder 36 will be described below based on the variable-length coding method of the quantization coefficient.

  In the MPEG2 system, it is defined that a 12-bit precision DCT coefficient for an 8-bit precision input image is encoded using a single variable-length coding table (combination of variable-length code codes). Yes.

  Also, when the value (or combination of values) of the quantization coefficient given to the variable length encoder 36 is outside the range of values (or combinations of values) defined in the variable length coding table, the variable length The value (or combination of values) is processed as a fixed length code using an escape code defined in the encoding table. That is, this escape code is a quantization coefficient value (or a combination of values) that has a relatively low probability of occurrence, and it is better to perform fixed-length coding than variable-length coding.

  Here, in the variable length encoder 36, the function of the MPEG2 system is expanded, and the N + 1 + α-bit precision DCT coefficient obtained from the N-bit precision input image is quantized, and the resulting quantum is obtained. The variable coefficients are variable length encoded.

  In the variable length encoder 36, the length of a fixed length code added when an escape code is used is changed according to the bit accuracy of the input image and the bit accuracy of the DCT coefficient. It is made like that. In addition, the encoding method can be changed according to the value of the quantization scale.

  First, the case where the variable length encoder 36 changes the variable length encoding table according to the bit accuracy of the input image will be described. This addresses the fact that the statistical properties of the occurrence of quantization coefficients change according to the input bit precision, and also prevents the deterioration of image quality by changing the range of values handled by escape codes from the establishment of the appearance. To do.

  Here, FIG. 6 shows a configuration in which the variable length coding table can be changed in the variable length coder 36 in accordance with the bit accuracy of the input image. In order to simplify the description, a plurality of types of variable-length coding tables stored in the memory are depicted outside the variable-length encoder 36.

  In the variable length encoder 36, for example, when the accuracy of the DCT coefficient output from the DCT unit 34 is fixed to α = 3, the bit accuracy of the DCT coefficient with respect to the bit accuracy N of the pixel of the input image is N + 4, and the quantized coefficient obtained by quantizing the DCT coefficient of N + 4 bit precision in the quantizer 35 is given.

  In this case, in the variable length encoder 36, a plurality of types of variable length encoding tables suitable for the bit accuracy of the pixels of the input image are stored in advance in the internal memory, and the bit accuracy N of the pixels of the input image is determined. Depending on the value, the length of the fixed-length code following the variable-length coding table and the escape code applied to the quantization coefficient can be changed.

  As a result, the variable length encoder 36 can compress and encode the quantized coefficient in an appropriate manner corresponding to the statistical property of the appearance establishment of the quantized coefficient that changes in accordance with the change in the input bit precision. Thus, it is possible to prevent the image quality from being deteriorated by the encoding process.

  Further, in the variable length encoder 36, the range that the level value of the quantization coefficient can take is −2N + 3 to 2N + 3-1, and it is necessary if the fixed length code length for representing the level value is N + 4 bits. Therefore, by changing the length of the fixed-length code in accordance with the value of N, it is possible to avoid wasting unnecessary bits on the fixed-length code, thus preventing deterioration in image quality. ing.

  Incidentally, FIG. 6 shows an example in which α = 3, but it is necessary and sufficient if the fixed length code length for expressing the level value is N + 1 + a bits even for α = a.

  That is, when the encoder 30 can handle three types of input images with pixel bit precision of N = 8 bits, 10 bits, and 12 bits, for example, and α = 6, the variable length encoder 36 has three types. A variable length coding table (for N = 8, 10, 12 respectively) may be provided, and the fixed length code lengths for representing level values may be 15, 17, and 19.

  Next, in the variable length encoder 36, the bit accuracy of the pixel of the input image is set to N = n, and the variable length encoding table when encoding the DCT coefficient of n + 1 + α bit accuracy is changed according to the value of α. The case will be described. This deals with the fact that the appearance of the quantization coefficient and the statistical properties change according to the bit accuracy of the DCT coefficient, and the range of values handled by the escape code is changed from the establishment of the appearance to prevent the deterioration of the image quality. It is something to prevent.

  Here, FIG. 7 shows a configuration in which the variable length encoding table can be changed in accordance with the bit accuracy α of the DCT coefficient in the variable length encoder 36. In order to simplify the description, a plurality of types of variable-length coding tables stored in the internal memory are depicted outside the variable-length encoder 36.

  In the variable length encoder 36, for example, when considering the case where the bit precision of the pixel of the input image is fixed to N = 8, the bit precision of the DCT coefficient is 9 + α, and the DCT coefficient of 9 + α bit precision is converted into the quantizer. The quantization coefficient obtained by the quantization at 35 is given.

  In this case, the variable length coder 36 follows the variable length coding table applied to the quantized coefficient and the escape code according to the bit precision of the DCT coefficient output from the DCT unit 34, in this case, the value of α. The length of the level value of the fixed length code is changed.

  That is, the variable length encoder 36 stores in advance an internal memory suitable for the DCT coefficient bit precision as a variable length coding table, and converts the quantization coefficient into a quantization coefficient based on the bit precision α of the DCT coefficient. The variable length coding table to be applied and the length of the fixed length code following the escape code can be changed.

  Thereby, in the variable length encoder 36, it is possible to compression-encode the quantized coefficient corresponding to the establishment of the appearance of the quantized coefficient that changes with the change of the DCT coefficient bit precision and the statistical property appropriately, Thus, it is possible to prevent the image quality from being deteriorated by the encoding process.

  Further, in the variable length encoder 36, the range that the level value of the quantization coefficient can take is −28 + α to 28 + α−1. Therefore, it is necessary and sufficient that the fixed length code length for representing the level value is 9 + α bits. In other words, by changing the length of the fixed-length code according to the value of α, it is possible to avoid wasting unnecessary bits on the fixed-length code, and thus it is possible to prevent deterioration of the image quality.

  Incidentally, FIG. 7 shows an example in the case of N = 8, but it is necessary and sufficient that the fixed length code length for representing the level is n + 1 + α bits even for N = n.

  That is, in the encoder 30, when N = 10 and, for example, it is possible to handle an input image with bit accuracy of three types of DCT coefficients of α = 3, 4, and 5, the variable length encoder 36 has three types ( Each of them has a variable length coding table (for α = 3, 4, 5), and the fixed length code length for representing the level value may be 14, 15, 16.

  Here, in the variable length encoder 36 described above with reference to FIGS. 6 and 7, a variable length coding table according to the bit accuracy N of the pixel of the individual input image and the bit accuracy α of the DCT coefficient, Although the case where the fixed-length code length representing the level value of the quantization coefficient is switched as appropriate has been described, it is also possible to combine these two variable-length coding schemes.

  That is, FIG. 8 shows a plurality of types of variable length coding tables drawn outside the variable length encoder 36 for the sake of simplicity of explanation. The variable length coding table and the fixed length code length are appropriately switched in consideration of both the bit accuracy N of the pixel of the input image and the bit accuracy α of the DCT coefficient. Incidentally, the fixed length code length in the variable length encoder 36 is N + 1 + α.

  By the way, in the variable length coder 36, the bit precision of the pixel of the input image is set to N = n, and the bit precision of the DCT coefficient at α = a is set to n + 1 + a. A case will be described in which the variable length coding table for variable length coding of the quantization coefficient obtained by the quantization and the fixed length code length of the level value are changed according to the value of the quantization scale. This deals with the fact that the statistical nature of the probability establishment of the quantization coefficient changes according to the value of the quantization scale, and the image quality deteriorates by changing the range of values handled by the escape code from the appearance establishment. This is to prevent this.

  Here, FIG. 9 shows a configuration in which the variable length encoding table can be changed in accordance with the value of the quantization scale in the variable length encoder 36. In order to simplify the description, a plurality of types of variable-length coding tables stored in the memory are depicted outside the variable-length encoder 36.

  In the variable-length encoder 36, for example, when the bit accuracy of the pixel of the input image is N = 8 and α = 3 is fixed, the bit accuracy of the DCT coefficient is 12, and the 12-bit accuracy DCT A quantized coefficient obtained by quantizing the coefficient in the quantizer 35 is given.

  In the variable length encoder 36, the range of DCT coefficient values that are input to the quantizer 35 is -2048 to +2047. At this time, when the quantization scale is 1, When the quantization coefficient value can take a range of −2048 to +2047, and the quantization scale is 1, the bit precision of the quantization coefficient is 12 bits.

  In the variable-length encoder 36, when the quantization scale is 2, the range that the quantization coefficient can take is -1024 to +1023, and the bit precision of the quantization coefficient is 11 bits. Further, in the variable length encoder 36, when the quantization scale is 4, the range that the quantization coefficient can take is −512 to +511, and the bit precision of the quantization coefficient is 10 bits.

  FIG. 10 shows the bit accuracy of the quantization coefficient based on the value of the quantization scale. In this case, in the quantization scale, different variable-length coding tables and fixed-length code lengths of level values are assigned to the respective groups, and the variable-length encoder 36 gives the value of the given quantization scale. In response to this, the corresponding variable length coding table and the fixed length code length of the level value in the case of an escape code are switched. Incidentally, the fixed length code length of the level value is the same as the quantization coefficient bit precision value of FIG.

  FIG. 9 shows the variable length encoder 36 that switches the variable length encoding table based on the quantization scale. As shown in FIG. 11, in addition to this, FIG. 9 and FIG. The variable-length coding table and the fixed-length code length representing the level value of the quantization coefficient can be switched as appropriate using the bit precision N of the pixel of the input image and the bit precision α of the DCT coefficient.

  In FIG. 11, various variable length coding tables stored inside the variable length encoder 36 are drawn outside the variable length encoder 36 to simplify the description.

  Further, the variable length encoder 36 uses at least one of the bit accuracy N of the pixel of the input image, the bit accuracy α of the DCT coefficient, and the quantization scale, and the quantization coefficient. The fixed length code length representing the level value can be switched as appropriate.

  Further, the variable length encoder 36 determines whether or not to make a macroblock data of I picture, P picture, and B picture (I-VOP, P-VOP, B-VOP) to be a skip macroblock, A flag indicating the determination result is set, and the flag is also variable-length encoded and transmitted.

Incidentally, in the encoder 30 performs the encoding of the DCT coefficients for each block, when it is determined that there is the block outside the object by the shape information outputted from the shape information decoding unit 52, DCT coefficients Is not encoded.

(3) Configuration of Decoder According to this Embodiment On the other hand, in FIG. 12, reference numeral 100 denotes a decoder to which the present invention is applied, which is supplied from the encoder 30 (FIG. 1) described above with reference to FIG. The encoded bit stream is decoded.

In the decoder 100 is fed heat from the encoder 30, a predetermined receiving apparatus is received (not shown) coded bit stream, or recorded on a recording medium by the encoder 30, a predetermined reproduction device (shown The encoded bit stream reproduced from the recording medium is taken into the buffer 101 and temporarily stored.

  The variable length decoder 102 appropriately reads the encoded bit stream from the buffer 101 corresponding to the processing state of the block in the subsequent stage, and performs variable length decoding on the encoded bit stream, thereby obtaining a quantization coefficient, a motion vector, The encoding mode information, quantization scale code, size data FSZ_B, offset data FPOS_B, and the like are separated.

  Then, the variable length decoder 102 sends the quantization coefficient, the quantization scale code, the motion vector, and the encoding mode information to the inverse quantizer 103, and the motion vector, the encoding mode information, the size data FSZ_B, and the offset Data FPOS_B is sent to the motion compensator 107, and shape information is sent to the shape information decoder 110.

  The inverse quantizer 103, the IDCT unit 104, the computing unit 105, the frame memory 106, the shape decoder 110, the padding processor 108, and the motion compensator 107 are the inverse quantizer 38 (see FIG. 1), IDCT unit 39 (FIG. 1), computing unit 40 (FIG. 1), frame memory 41 (FIG. 1), shape information decoder 52 (FIG. 1), padding processor 51 (FIG. 1), motion compensator The same processing as that in FIG.

  The shape information decoder 110 decodes the encoded shape information given from the variable length decoder 102 and sends the decoded shape information to the padding processor 108.

Inverse quantizer 103, based on the quantization scale type is set for each quantization scale code given and the picture from the variable length decoder 102 (linear quantization and non-linear quantization), as well as variable-length The quantized coefficients given from the decoder 102 are inversely quantized, and the obtained DCT coefficients are sent to the IDCT unit 104.

  The IDCT unit 104 performs IDCT processing on the DCT coefficient given from the inverse quantizer 103 and supplies the obtained data to the arithmetic unit 105.

The arithmetic unit 105 adds the output from the motion compensator 107 and the output of the IDCT unit 104 in units of one pixel when inter-frame prediction is performed, and the obtained addition result is sent to the padding processor 108 as texture information. Sends out and outputs to the outside. Further, the arithmetic unit 105, intra-frame prediction (intra-coding) is the taking place Rutoki not operate particularly.

  Based on the shape information given from the shape information decoder 110, the padding processor 108 performs padding processing on the texture information given from the computing unit 105, and sends the padded texture information to the frame memory 106. And remember.

  The texture information stored in the frame memory 106 is appropriately read out by the motion compensator 107 and sent to the computing unit 105, and thereafter operates in the same manner as the frame memory 41 and the motion compensator 42 described above with reference to FIG. .

  Here, a case where the MPEG4 standard is applied to the encoder 30 and the decoder 100 will be described below.

  First, the quantizer 35 and inverse quantizer 38 of the encoder 30 and the inverse quantizer 103 of the decoder 100 will be described.

  FIG. 13 shows the VOL (Video Object Layer) syntax in the MPEG4 format syntax. The VOL describes information related to all the encoded video information stored in the encoded bitstream. Layer. In this syntax, bits_per_pixel is prepared as a flag for specifying the bit precision of the pixel of the input image.

  In the present invention, dct_precision is prepared as a flag for designating the bit precision of the DCT coefficient, and the value of α is described therein. That is, the bit precision of the DCT coefficient is bits_per_pixel + 1 + dct_precision bits.

  Further, according to the present invention, according to the bit accuracy, the VOP layer (describes information for each VOP) shown in FIG. 14 and the macroblock layer (Macroblock layer) describing information for each macroblock shown in FIG. In the syntax of Layer)), quantizer_scale_code is provided as a flag indicating the quantization scale code, and the bit length of the quantization scale is changed using this flag.

  That is, the bit length of quantizer_scale_code is bits_per_pixel + dct_precision-6 according to the values of bits_per_pixel and dct_percision of the VOL syntax. Incidentally, in the MPEG2 system, bits_per_pixel and dct_precision are 8 and 3, respectively, and the bit length of quantizer_scale_code is 5, but in the present invention, the bit length of quantizer_scale_code is 5, Have compatibility.

  In the encoder 30 and the decoder 100, bits_per_pixel, dct_precision, and quantizer_scale_code are designated according to these syntaxes, and then the macroblock data is quantized and dequantized.

  First, in the quantizer 35, when linear quantization is used, the equation (7) is used.

The quantization scale quantizer_scale is calculated from the quantization scale code quantizer_scale_code represented by

  Further, the quantizer 35 uses the expression (8) when nonlinear quantization is used.

The quantization scale quantizer_scale is calculated from the quantization scale code quantizer_scale_code represented by

  Incidentally, in the encoder 30 and the decoder 100, a flag called quant_scale_type in the VOP syntax shown in FIG. 14 determines whether linear quantization or nonlinear quantization is applied to a picture to be encoded. The value specifies which quantization type to use. Thus, the quantizer 35 can perform quantization using the quantization scale quantizer_scale calculated using the designated quantization type.

  However, when the macroblock coding mode is the intra coding mode, a different quantization method is used for the DC component of the DCT coefficient.

  Actually, the quantization scale dc_scaler of the DC component of the DCT coefficient can be derived by the following procedure.

  That is, when the bit precision of the DC component of the DCT coefficient is bits_per_pixel + dct_precision, the bit precision of the quantization coefficient can be specified in picture units from the bit precision shown in FIG. For example, the user or the like designates it with a flag called intra_dc_precision in the VOP syntax shown in FIG.

  Then, when the bit precision of the quantization coefficient is designated by intra_dc_precision, the quantizer 35 calculates a corresponding quantization scale dc_scaler from the relationship between the bit precision shown in FIG. 16 and the quantization scale, Is used to quantize the DC component of the DCT coefficient in the intra coding mode.

  Next, the inverse quantizers 38 and 103 calculate the quantization scale quantizer_scale from the quantization scale code quantizer_scale_code in the same manner as the quantizer 35 described above, and perform the inverse using the quantization scale quantizer_scale. Perform quantization.

  However, when the coding mode of the macroblock data is the intra coding mode, the dequantizers 38 and 103 use different dequantization methods for the DC component of the DCT coefficient. A quantization scale dc_scaler is obtained based on intra_dc_precision shown in FIG. 16, and inverse quantization is performed using the quantization scale.

  In this way, the inverse quantizers 38 and 103 can perform the inverse quantization of the DC component of the DCT coefficient by calculating the quantization scale dc_scale and using it for the inverse quantization.

  Next, the variable length encoder 36 and the variable length decoder 102 according to the present invention will be described.

  FIG. 17 shows the block layer (Block Layer) in the syntax of the MPEG4 system. In the current MPEG4 system, a symbol representing the DCT Coefficient of this block syntax regardless of the bit precision bits_per_pixel of the input image. Is encoded using a single variable length coding table.

  Here, the variable-length encoding method includes the final code (LAST) in a predetermined block to be encoded, the number of consecutive zeros in the block (RUN), and the value of non-zero coefficient (LEVEL) in the block. 3D variable length encoding is performed from the three types of data. Incidentally, dct_precision is fixed at 3 in the current MPEG4 system.

  That is, the variable length encoder 36 uses the variable length encoding table shown in FIGS. 18 to 20 when performing 3D variable length encoding.

  Further, in the current MPEG4 system, regardless of the bit precision bits_per_pixel of the input image, when an escape code is selected at the time of variable length coding of a block, a level value and a run length are set as shown in FIG. Encoding is performed using a fixed-length code. The bit length for encoding the level value is also fixed regardless of bits_per_pixel, and the length is 12 bits.

  This is the bit precision of the DCT coefficient in the case of bits_pre_pixel = 8, dct_precision = 3 of the present invention, so it cannot be said to be a sufficient value for an input image with a bit precision higher than an 8-bit precision input. .

  Therefore, in the variable length encoder 36, as shown in FIG. 22, the variable length encoding table is switched according to the values of bits_per_pixel, dct_precision, and quantizer_scale.

  For example, FIG. 22 shows a case where bits_per_pixel = 8, 10, 12, dct_precision = 5 can be selected, and VLCt_0 to 20 are variable length coding tables as shown in FIGS. , All can be different tables.

  In the variable length encoder 36, the bit length of the fixed length code indicating the level value can be switched according to the values of bits_per_pixel, dct_precision, and quantizer_scale. 23. This example also shows a case where bits_per_pixel = 8, 10, 12, and dct_precision = 5 can be selected.

  On the other hand, in the variable-length decoder 102, similarly to the variable-length coding in the variable-length coder 36, the variable-length coding table and the escape according to the values of bits_per_pixel, dct_precision, and quantizer_scale. Variable length decoding is performed by switching the bit length of the level value at the time of code generation.

(4) Operation and effect of the present embodiment In the configuration described above, the encoder 30 can capture an input image having an arbitrarily set bit accuracy N and can arbitrarily set the bit accuracy α of the DCT coefficient. When the difference value data is generated from the frame image data of the input image using a motion vector or the like in units of macro blocks, the generated difference value data is DCT processed to generate a DCT coefficient having a set bit accuracy α. To do.

  Then, the encoder 30 quantizes the DCT coefficient using the quantization scale expanded based on the bit precision N of the input image and the bit precision α of the DCT coefficient which are arbitrarily set, and the obtained quantization coefficient Is a variable length coding table selected based on at least one of the bit precision N of the input image, the bit precision α of the DCT coefficient, and the value of the quantization scale used for quantization. Variable length coding.

  At this time, the encoder 30 uses the escape code defined in the variable length coding table when the value of the quantization coefficient is outside the range of values defined in the selected variable length coding table. The quantized coefficient is processed as a fixed length code.

  On the other hand, the decoder 100 quantizes the encoded bit stream given from the encoder 30 by expanding the bit precision N of the input image and the bit precision α of the DCT coefficient, which are arbitrarily set, and the bit precision N and α. Decoding is performed using a variable length coding table selected based on at least one of the quantization scales, and quantization coefficients are separated together with various data.

  In the decoder 100, the quantization coefficient is inversely quantized using the quantization scale expanded based on the bit precision N of the input image set arbitrarily and the bit precision α of the DCT coefficient, and the set bit is set. A DCT coefficient with accuracy α is generated, and IDCT processing is performed on the DCT coefficient. Thereafter, decoded frame image data is obtained.

  Therefore, the encoder 30 and the decoder 100 are suitable for the bit precision α of the DCT coefficient arbitrarily set when performing quantization and inverse quantization, and are suitable for the MPEG2 system. Compatible linear and nonlinear quantization can be performed.

  Further, in the encoder 30 and the decoder 100, when performing variable-length coding and variable-length decoding, an appropriate variable-length coding table and escape code are set with respect to the set bit precision α and quantization scale of the DCT coefficient. The bit length of the level value following can be selected.

  As a result, the encoder 30 and the decoder 100 can prevent the image quality of the input image from deteriorating when the input image is compressed and encoded, and can also decode the encoded input image. Also, it is possible to prevent the image quality from deteriorating.

  According to the above configuration, an input image with arbitrarily set bit accuracy N can be captured, and the bit accuracy α of the DCT coefficient can be arbitrarily set, and the DCT coefficient is set to the bit accuracy N of the input image, Quantization is performed using the quantization scale expanded based on the bit accuracy α of the DCT coefficient, and the obtained quantization coefficient is converted into the bit accuracy N of the input image, the bit accuracy α of the DCT coefficient, and the quantum used for the quantization. Since the variable length encoding is performed using the variable length encoding table selected based on at least one of the values of the conversion scale, when the input image is compressed and encoded, An encoder that can prevent deterioration in image quality can be realized.

  In addition, an input image with arbitrarily set bit precision N can be taken in, and the bit precision α of the DCT coefficient can be arbitrarily set, and the bit precision N of the input image and the bit precision of the DCT coefficient of the encoded bit stream can be set. Quantization coefficient obtained by variable length decoding using a variable length coding table selected based on at least one of α and a quantization scale expanded based on bit precision N and α Is decoded using a quantization scale expanded based on the bit accuracy N of the input image and the bit accuracy α of the DCT coefficient, so that when the compression-encoded input image is decoded, It is possible to realize a decoder that can prevent the image quality of the input image from deteriorating.

  Therefore, according to the present embodiment, it is possible to prevent the image quality from being deteriorated during compression encoding and decoding of the input image. Thus, after compressing and encoding the input image, the image obtained by decoding is decoded. It is possible to realize an encoder and a decoder that can make the image quality substantially equal to the image quality of the input image.

(5) Other Embodiments In the above-described embodiment, the case where bits_per_pixel = 8, 10, 12, and dct_precision = 5 can be selected has been described. However, the present invention is not limited to this. Other combinations are possible, and one or both can be fixed at a certain value. This should be selected according to the purpose of use of the system, but the present invention is applicable to any such system.

  In the above-described embodiment, the encoder 30 performs the quantization and variable length encoding according to the bit accuracy. However, the present invention is not limited to this, and the encoder is configured to perform bit encoding. Either one of quantization or variable length coding according to accuracy may be executed, and even in this case, deterioration of image quality can be prevented in advance.

  Furthermore, in the above-described embodiment, the case where the decoder 100 performs quantization and variable length coding according to the bit precision has been described. However, the present invention is not limited to this, and the decoder uses a bit. Either one of quantization or variable length coding according to accuracy may be executed, and even in this case, deterioration of image quality can be prevented in advance.

  30 ... Encoder, 31 ... Frame memory, 34 ... DCT, 35 ... Quantizer, 36 ... Variable length encoder, 38,103 ... Inverse quantizer, 39,104 ... IDCT , 100... Decoder, 102... Variable length decoder.

Claims (28)

Differential value data generation means for generating differential value data subject to discrete cosine transform processing in predetermined data units from input image data having arbitrary input bit precision;
Discrete cosine transform coefficient generation means for performing discrete cosine transform processing on the difference value data according to arbitrarily set output bit precision and generating discrete cosine transform coefficients;
A predetermined quantization scale is expanded based on at least one of the output bit precision and the input bit precision of the discrete cosine transform coefficient, and the discrete cosine transform coefficient is quantized using the expanded quantization scale. Quantization coefficient generating means for generating quantization coefficients
Comprising Ru marks Goka device and an encoding means for compression encoding using a predetermined variable length coding table the quantized coefficients. The encoding means is
A plurality of types of the variable-length coding tables are stored in advance, and arbitrary variable-length codes from each of the variable-length coding tables based on at least one of the input bit precision, the output bit precision, and the quantization scale selects the quantization table, you compression coding said quantized coefficients using variable length coding table said selected
Encoding apparatus according to 請 Motomeko 1. The encoding means is
When out of range of the quantized coefficients are defined by the variable-length coding table above selected, based on the variable length coding table were those 該選 boss, you fixed length encoding the quantized coefficients
Encoding apparatus according to 請 Motomeko 2. Differential value data generation means for generating differential value data subject to discrete cosine transform processing in predetermined data units from input image data having arbitrary input bit precision;
Discrete cosine transform coefficient generating means for generating a discrete cosine transform coefficient by performing discrete cosine transform processing on the difference value data according to arbitrarily set output bit precision;
Quantization coefficient generation means for generating a quantization coefficient by quantizing the discrete cosine transform coefficient using a predetermined quantization scale;
A plurality of types of the variable length coding tables are held in advance, and arbitrary variable lengths are obtained from each of the variable length coding tables based on at least one of the input bit precision and the output bit precision of the discrete cosine transform coefficients. select an encoding table, comprising Ru marks Goka device and an encoding means for compressing and encoding the quantized coefficients using variable length coding tables corresponding selected. The quantization coefficient generation means includes
A predetermined quantization scale is expanded based on at least one of the output bit precision and the input bit precision of the discrete cosine transform coefficient, and the discrete cosine transform coefficient is quantized using the expanded quantization scale. that generates the quantization coefficient Te
Encoding apparatus according to 請 Motomeko 4. The encoding means is
An arbitrary variable length coding table is selected from each of the variable length coding tables based on at least one of the input bit precision, the output bit precision, and the extended quantization scale, and the selected variable length code compression coding said quantized coefficients using quantization table
Encoding apparatus according to 請 Motomeko 5. The encoding means is
When out of range of the quantized coefficients are defined by the variable-length coding table above selected, based on the variable length coding table were those 該選 boss, you fixed length encoding the quantized coefficients
Encoding apparatus according to 請 Motomeko 6. A differential value data generation step for generating differential value data to be subjected to discrete cosine transform processing in a predetermined data unit from input image data having arbitrary input bit precision;
A discrete cosine transform coefficient generation step for generating a discrete cosine transform coefficient by performing a discrete cosine transform process on the difference value data according to an arbitrarily set output bit precision;
A predetermined quantization scale is expanded based on at least one of the output bit precision and the input bit precision of the discrete cosine transform coefficient, and the discrete cosine transform coefficient is quantized using the expanded quantization scale. A quantization coefficient generation step for generating a quantization coefficient by
Marks Goka method Ru comprising an encoding step of compression encoding using the quantization coefficients predetermined variable length coding table. In the encoding step,
A plurality of types of the variable-length coding tables are stored in advance, and arbitrary variable-length codes from each of the variable-length coding tables based on at least one of the input bit precision, the output bit precision, and the quantization scale selects the quantization table, you compression coding said quantized coefficients using variable length coding table said selected
Encoding method according to 請 Motomeko 8. In the encoding step,
When out of range of the quantized coefficients are defined by the variable-length coding table above selected, based on the variable length coding table were those 該選 boss, you fixed length encoding the quantized coefficients
Encoding method according to 請 Motomeko 9. A differential value data generation step for generating differential value data to be subjected to discrete cosine transform processing in a predetermined data unit from input image data having arbitrary input bit precision;
A discrete cosine transform coefficient generation step for generating a discrete cosine transform coefficient by performing a discrete cosine transform process on the difference value data according to an arbitrarily set output bit precision;
A quantization coefficient generating step for generating a quantization coefficient by quantizing the discrete cosine transform coefficient using a predetermined quantization scale;
A plurality of types of the variable length coding tables are held in advance, and arbitrary variable lengths are obtained from each of the variable length coding tables based on at least one of the input bit precision and the output bit precision of the discrete cosine transform coefficients. select an encoding table, marks Goka method Ru comprising an encoding step of compressing and encoding the quantized coefficients using variable length coding tables corresponding selected. In the quantization coefficient generation step,
A predetermined quantization scale is expanded based on at least one of the output bit precision and the input bit precision of the discrete cosine transform coefficient, and the discrete cosine transform coefficient is quantized using the expanded quantization scale. that generates the quantization coefficient Te
Encoding method according to 請 Motomeko 11. In the encoding step,
An arbitrary variable length coding table is selected from each of the variable length coding tables based on at least one of the input bit precision, the output bit precision, and the extended quantization scale, and the selected variable length code compression coding said quantized coefficients using quantization table
Encoding method according to 請 Motomeko 12. In the encoding step,
When out of range of the quantized coefficients are defined by the variable-length coding table above selected, based on the variable length coding table were those 該選 boss, you fixed length encoding the quantized coefficients
Encoding method according to 請 Motomeko 13. Decoding means for generating a quantized coefficient by decoding the compression-coded quantized coefficient using a predetermined variable length coding table;
The predetermined quantization scale is expanded based on at least one of an arbitrary input bit precision of the predetermined input image data that is the basis of the quantization coefficient and an output bit precision of a preset discrete cosine transform coefficient. and, Ru decrypt apparatus comprising an inverse quantization means for generating the discrete cosine transform coefficients by inverse quantization processing to the quantized coefficient using a quantization scale and the extension. The decoding means includes
A plurality of types of the variable-length coding tables are stored in advance, and arbitrary variable-length codes from each of the variable-length coding tables based on at least one of the input bit precision, the output bit precision, and the quantization scale selects the quantization table, that turn into decodes the quantized coefficients the compressed coded using a variable length coding table said selected
Decoding apparatus according to 請 Motomeko 15. The decoding means includes
The compression encoded the quantized coefficients, when outside the range defined by the variable-length coding table above selected, based on the variable length coding table were those 該選 boss, is the compression coding that turn into fixed-length decodes the quantized coefficients
Decoding apparatus according to 請 Motomeko 16. A plurality of types of the variable length coding tables are held in advance, and the supplied compression-encoded quantization coefficient is set to the input bit accuracy of the input image data supplied to the compression encoding side, and a preset discrete Decoding means for generating a quantized coefficient by decoding using any variable length coding table selected from each of the variable length coding tables based on at least one of output bit precisions of cosine transform coefficients; ,
Comprising Ru decrypted device and an inverse quantization unit to inverse quantization processing to generate the discrete cosine transform coefficients using a prescribed quantization scale to the quantization coefficients. The inverse quantization means is
A predetermined quantization scale is expanded based on at least one of the input bit precision and the output bit precision, and the quantization coefficient is inversely quantized using the expanded quantization scale. that generates the discrete cosine transform coefficients
Decoding apparatus according to 請 Motomeko 18. The decoding means includes
An arbitrary variable length coding table is selected from each of the variable length coding tables based on at least one of the input bit precision, the output bit precision, and the extended quantization scale, and the selected variable length code that turn into decodes the quantized coefficients are the compression encoded using a quantization table
Decoding apparatus according to 請 Motomeko 19. The decoding means includes
The compression encoded the quantized coefficients, when outside the range defined by the variable-length coding table above selected, based on the variable length coding table were those 該選 boss, is the compression coding that turn into fixed-length decodes the quantized coefficients
Decoding apparatus according to 請 Motomeko 20. A decoding step of decoding the compression-coded quantized coefficients using a predetermined variable length coding table to generate quantized coefficients;
The predetermined quantization scale is expanded based on at least one of an arbitrary input bit precision of the predetermined input image data that is the basis of the quantization coefficient and an output bit precision of a preset discrete cosine transform coefficient. and, decrypt method of Ru comprising an inverse quantization step and the inverse quantization processing to the quantized coefficient using a quantization scale and the extension to generate the discrete cosine transform coefficients. In the decryption step,
A plurality of types of the variable-length coding tables are stored in advance, and arbitrary variable-length codes from each of the variable-length coding tables based on at least one of the input bit precision, the output bit precision, and the quantization scale selects the quantization table, that turn into decodes the quantized coefficients the compressed coded using a variable length coding table said selected
Decoding method according to 請 Motomeko 22. In the decryption step,
The compression encoded the quantized coefficients, when outside the range defined by the variable-length coding table above selected, based on the variable length coding table were those 該選 boss, is the compression coding that turn into fixed-length decodes the quantized coefficients
Decoding method according to 請 Motomeko 23. A plurality of types of the variable length coding tables are held in advance, and the supplied compression-encoded quantization coefficient is set to the input bit accuracy of the input image data supplied to the compression encoding side, and a preset discrete A decoding step of generating a quantized coefficient by decoding using any variable length coding table selected from each of the variable length coding tables based on at least one of the output bit precision of the cosine transform coefficients; ,
Decrypted How to the quantized coefficients to inverse quantization processing using a predetermined quantization scale Ru comprising an inverse quantization step of generating the discrete cosine transform coefficients. In the inverse quantization step,
A predetermined quantization scale is expanded based on at least one of the input bit precision and the output bit precision, and the quantization coefficient is inversely quantized using the expanded quantization scale. that generates the discrete cosine transform coefficients
Decoding method according to 請 Motomeko 25. In the decryption step,
An arbitrary variable length coding table is selected from each of the variable length coding tables based on at least one of the input bit precision, the output bit precision, and the extended quantization scale, and the selected variable length code that turn into decodes the quantized coefficients are the compression encoded using a quantization table
Decoding method according to 請 Motomeko 26. In the decryption step,
The compression encoded the quantized coefficients, when outside the range defined by the variable-length coding table above selected, based on the variable length coding table were those 該選 boss, is the compression coding that turn into fixed-length decodes the quantized coefficients
Decoding method according to 請 Motomeko 27.

Images