JP4309220B2 - Image coding apparatus and image coding method - Google Patents

Description

  The present invention relates to an image encoding apparatus and an image encoding method that perform correction processing for monotonically decreasing the slope of code amount versus distortion with respect to the code amount, particularly in JPEG 2000 image encoding rate control processing.

  Currently, the still image coding algorithm JPEG is widespread mainly on the Internet, but on the other hand, the JPEG2000 project was newly established in 1997 as a next-generation coding method with the background of further performance improvement and addition of functions. Started by a joint organization of ITU. In December 2000, the main technical content of Part 1 that defines the basic method of the JPEG2000 algorithm was finalized (for example, see Non-Patent Document 1). The basic scheme of the JPEG2000 encoding algorithm will be described below according to Non-Patent Document 1. The functional configuration as an image encoding device is substantially the same as that of FIGS. 1 and 2 of the present invention.

  The input image signal is first subjected to a two-dimensional wavelet transform and is divided into a plurality of subbands. Here, the two-dimensional wavelet transform is realized as a combination of the one-dimensional wavelet transform. That is, a process of sequentially performing horizontal one-dimensional wavelet transform for each line and a process of sequentially performing vertical one-dimensional wavelet transform for each column. The one-dimensional wavelet transform is composed of a low-pass filter, a high-pass filter, and a downsampler having predetermined characteristics, as shown in FIG. The two-dimensional wavelet transform coefficient generated in this way is expressed by expressing the low-frequency component as L, the high-frequency component as H, the horizontal conversion by the first character, and the vertical sub-scanning conversion by the second character. As shown in FIG. 20B, they are expressed as LL, HL, LH, and HH. These band-divided components are called subbands. Here, the horizontal and vertical low-frequency components (LL components) are recursively subjected to wavelet transform. Each subband generated by each wavelet transform applied recursively is called a decomposition level, and the numbers described before LL, HL, LH, and HH in the figure correspond to this. In other words, when the number of wavelet transform decompositions is 2, the decomposition level of the lowest resolution component is 2, whereas the decomposition level of HL, LH, and HH of the highest resolution component is 1.

  Next, the wavelet transform coefficient of each subband is quantized by the quantization step size set for each subband. The quantized wavelet transform coefficient of each subband is divided into fixed-size areas called code blocks, and then multivalued data of each code block is converted into a binary bit plane representation. Divided into three encoding passes (Significant Propagation Decoding Pass, Magnitude Refinement Pass, and Cleanup Pass). The binary signal output from these three coding passes is subjected to context modeling for each coding pass and subjected to entropy coding. In parallel with the entropy encoding process, the code amount and distortion information for each encoding pass are calculated in each code block. Finally, rate control is performed to adjust the code amount to a target code size or less while minimizing image quality deterioration (distortion) using the Lagrange's undetermined multiplier method. An outline of the mechanism of the rate control unit according to Non-Patent Document 1 will be described below.

When the truncation point in each code block i is ni, the code amount up to each truncation point is R (i, ni), and the distortion is D (i, ni), Σ (R (i, ni) + λD ( The variable λ is adjusted until the code amount R generated by the truncation point that minimizes the value represented by i, ni)) is within the target code amount Rmax. When this is seen for each code block, it is necessary to find a truncation point ni that minimizes (R (i, ni) + λD (i, ni)) as follows. Here, k is a candidate for a cut-off point.
Set ni = 0
For k = 1,2,3, ...
Set ΔR (i, k) = R (i, k) -R (i, ni) and ΔD (i, k) = D (i, k) -D (i, ni)
If (ΔD (i, k) / ΔR (i, k))> λ ^-1 ,
then set ni = k

However, in this algorithm, the cut-off point ni cannot be obtained unless the above processing is executed for a large number of variables λ. Therefore, it is preferable to correct the slope S (i, k) = ΔD (i, k) / ΔR (i, k) so as to monotonously decrease with respect to the cut-off point candidate k. Specifically, the following program processing is performed. Here, p is one of the set elements of the cut-off point ni.
(1) set Ni = {n} (ie the set of all truncation point)
(2) Set p = 0
(3) For k = 1,2,3,4, ..., kmax
If k belongs to Ni
Set ΔR (i, k) = R (i, k) -R (i, p), and ΔD (i, k) = D (i, p) -D (i, k)
Set S (i, k) = ΔD (i, k) / ΔR (i, k)
If p ≠ 0 and S (i, k)> S (i, p),
then remove p from Ni, and go to step (2)
Otherwise, set p = k
With this process, the optimization of the truncation point for a given variable λ may be the maximum k in Ni (the set of code paths in the code block) that satisfies S (i, k)> λ ⁻¹ . A specific example of monotonic decrease will be described later with reference to the drawings in contrast to the present invention.

  FIG. 21 shows a flowchart regarding the monotonous decrease correction process. In the figure, i indicating a code block is omitted. In addition, ‘remove p from Ni’, that is, the operation of removing p from the candidate Ni for the cut point, realizes the consent process using a flag indicating validity / invalidity in the drawing.

  When the derivation of these pieces of information is completed for all the code blocks, code data having the target code amount Rmax is created. Specifically, the variable λ is found such that Rsum ≦ Rmax with respect to the total code amount Rsum for a certain variable λ, and the distortion is minimized. When the variable λ is obtained, a truncation point is uniquely obtained from the value of each code block, and code data up to the truncation point is collected from all the code blocks to form final code data. In this way, the code amount can be controlled to the target code amount Rmax while minimizing distortion.

Recommendation ISO / ITU 15444-1: 2000, J.I. 14.3 (p220 to p221)

  In the conventional technique described above, the process of correcting the slope S with respect to the cut-off point candidate k to monotonously decreases includes a division process (S (i, k) = ΔD (i, k) / ΔR (i, k)). However, when the maximum value of the cut-off point candidates k is kmax, the maximum value of the number of divisions in one code block is (kmax (kmax + 1) / 2) times. Since the normal division processing requires a large number of processing clocks when implemented by hardware, there is a problem that the processing speed becomes slow due to an increase in the number of divisions. Even if it is implemented by software, when it is implemented by a high-speed CPU such as a personal computer or workstation, it does not have much influence, but it has sufficient processing capability such as a digital still camera or a mobile phone. When executed by a CPU that does not have, an increase in division processing affects the overall processing time. As described above, regardless of whether the implementation is hardware or software, there is a problem that the processing speed becomes slow due to the large number of divisions.

  The present invention has been made in order to solve the above-described problems, and an image encoding device that enables high-speed encoding processing by improving the efficiency of division processing for calculating the slope of code amount versus distortion. And it aims at obtaining the image coding method.

The image coding apparatus according to the present invention generates a wavelet transform coefficient for each subband by two-dimensionally wavelet transforming an input image signal, and the wavelet transform coefficient is set with a quantization step size set for each subband. The quantized wavelet transform coefficient of each subband is divided into code blocks, the multi-value data of each code block is decomposed into binary bit planes, and then divided into coding passes. Entropy coding is performed for each coding pass, distortion and code amount are calculated for each divided coding pass, and the slope of code amount versus distortion is calculated based on the obtained code amount and distortion for each coding pass. Then, for each code block, a truncation point where the distortion is minimized within the target code amount is derived using the code amount, distortion, and the slope of the code amount versus distortion. An image encoding device for creating a code data having the minimum distortion from a higher encoding pass code data from the derived truncated point Te in the target code amount, the slope of the code amount versus strain for each coding pass If the slope calculation means used for the calculation calculates that the slope of a certain coding path is larger than the slope of the coding path that has just been valid, the slope is invalid, while if the slope is small, the slope is valid. Correction processing is performed so that the slope of the code amount versus distortion monotonously decreases from the bit toward the lower bit.

  According to the present invention, there is an effect of enabling high-speed encoding processing by effectively reducing the amount of division processing for calculating the gradient of code amount versus distortion.

Embodiment 1 FIG.
FIG. 1 is a block diagram showing a functional configuration of an image encoding device according to each embodiment.
In the figure, a wavelet transform unit 101 recursively performs two-dimensional wavelet transform on an input image signal horizontally and vertically using a low-pass filter and a high-pass filter, and obtains a wavelet transform coefficient for each subband. Means for generating. The quantization means 102 is a means for quantizing the wavelet transform coefficient generated by the wavelet transform means 101 with a quantization step size set for each subband. The coefficient modeling means 103 divides the quantized wavelet transform coefficients of each subband into code blocks, decomposes the multi-value data of each code block into binary bit planes, and performs binary arithmetic coding to perform binary arithmetic coding. This is means for dividing the bit plane into three types of encoding passes. The entropy encoding unit 104 is a unit that performs entropy encoding after performing context modeling for each divided encoding pass. The code memory 105 is means for temporarily storing the entropy-coded code data. The rate control information extraction unit 106 calculates distortion data and code amount data necessary for rate control from the divided coding paths, holds code amount versus distortion slope data calculated elsewhere, and converts them into rate It is a means to output as control information. The rate control means 107 is means for generating optimum code data that fits within the target code amount from the rate control information output from the rate control information extraction means 106.

FIG. 2 is a block diagram showing a detailed configuration of the rate control information extraction unit 106 and rate control unit 107 of FIG.
In the figure, the rate control information extraction unit 106 includes a distortion calculation unit 201, a code amount calculation unit 202, and a rate distortion memory 203. The distortion calculation unit 201 is a unit that calculates distortion from the wavelet transform coefficient for each coding pass divided by the coefficient modeling unit 103. The code amount calculation unit 202 is a unit that counts the code amount for each encoding pass. The rate distortion memory 203 temporarily stores distortion data and code amount data obtained for each coding pass, and a code amount versus distortion slope for each coding pass calculated by the slope calculation unit 301 of the rate control unit 107. It is means to do.

  The rate control unit 107 includes an inclination calculation unit 301, a cut-off point derivation unit 302, and a code extraction unit 303. The slope calculation means 301 calculates the slope S of the code amount versus distortion for each coding pass based on the code amount R and distortion D for each coding pass stored in the rate distortion memory 203, and at this time, from the upper bits This is a means for performing correction processing so that the slope S of the code amount versus distortion monotonously decreases toward the lower bits. In this example, the calculated slope S is stored in the rate distortion memory 203 of the rate control information extraction unit 106. The truncation point deriving unit 302 derives a truncation point at which the distortion is minimized within the target code amount using the code amount, distortion, and the gradient of the code amount versus distortion stored in the rate distortion memory for each code block. Means. The code extraction unit 303 collects code data of a coding pass higher than the truncation point derived in each code block from the code memory 105, creates code data having the minimum distortion within the target code amount, and an image compression apparatus It is a means to output as final data.

Next, the operation will be described.
For example, an image signal is input from an image scanner, a digital camera, or an image input device (not shown) such as a network or a storage medium. In the wavelet transform unit 101, one-dimensional wavelet transform is performed two-dimensionally on the input image signal in the horizontal direction and the vertical direction, and divided into four subbands. Here, the one-dimensional wavelet transform is constituted by a filter bank of a low-pass filter and a high-pass filter. FIG. 3 shows an example of a subband that has been recursively subjected to two-dimensional wavelet transform twice. In the figure, the first number represents the decomposition level, and the following two alphabetic characters L or H represent the types of filters in the horizontal and vertical directions. L represents a low-pass filter, and H represents the result of applying a high-pass filter. In addition, “recursively” performing wavelet transform twice means that when 1LL, 1HL, 1LH, and 1HH are generated by the first wavelet transform, the second wavelet transform is performed on the 1LL. To generate 2LL, 2HL, 2LH, and 2HH. The wavelet transform unit 101 obtains wavelet transform coefficients for such subbands.

  The quantizing unit 102 quantizes the wavelet transform coefficient obtained by the wavelet transform unit 101 with the quantization step size set for each subband. Next, the coefficient modeling means 103 divides the wavelet transform coefficients of each subband into fixed-size areas called code blocks, and then converts the multi-value data of each code block into a binary bit plane. Convert to (bit plane). Usually, the code block size of 64 × 64 or 32 × 32 is used.

Here, the decomposition of the bit plane will be described in detail with reference to FIG.
FIG. 4A shows an example of a 4 × 4 code block. When these data are converted into a 1-bit signal representing positive and negative and an absolute value representation, and the result of binary representation of the data in the vertical direction is arranged in units of rows, the result is as shown in FIG. . Next, FIG. 4C shows a collection of bits having the same bit numbers as in FIG. 4B. Here, when the LSB (Least Significant Bit) is the 0th bit and the MSB (Most Significant Bit) is the 3rd bit, the 0th bit plane is a collection of the 0th bit. A collection of 1 bits is a first bit plane, a collection of 2 bits is a second bit plane, and a collection of 3 bits is a third bit plane. In addition to this, a sign bit plane is created as a collection of bits representing positive and negative.

  In the coefficient modeling means 103, each bit plane is further divided into a significant propagation decoding decoding pass (significant propagation decoding pass), a magnitude refinement pass (significant refinement pass: significant). Are divided into three types of coding passes, that is, coding of a large coefficient) and a cleanup pass (cleanup pass: coding of remaining coefficient information).

  Next, the entropy encoding unit 104 first performs context modeling for arithmetic encoding for each encoding pass. However, bit planes that are all 0 from the MSB plane are not subjected to context modeling and encoding, and only the number of all 0 bit planes is written in the header. For the bit plane in which 1 appears first, context modeling is performed only with the cleanup pass. As for the other bit planes, as described above, context modeling is performed for three types of coding passes. FIG. 5 shows an example in which the number of bit planes in the code block is 6 and the number of effective bit planes is 4. When the contest modeling is completed, entropy coding is performed by arithmetic coding. The encoded data generated by the entropy encoding unit 104 is temporarily stored in the code memory 105.

  In parallel with the entropy encoding process, the rate control information extraction unit 106 performs the following process. In the distortion calculation means 201, the distortion D is calculated for each coding pass unit of each code block. Here, the distortion D indicates how much the mean square error with respect to the reproduced image is reduced when a certain bit plane is sent, and strictly speaking, it is a reduction amount of the distortion. Therefore, the amount of distortion reduction when no bit plane is sent is 0, and the amount of distortion reduction when the last bit plane is sent is equal to the mean square error. At the same time, the code amount calculation unit 202 counts the number of output bytes of the entropy encoding unit 104 for each encoding pass, and calculates the code amount up to the code pass. These distortion data and code amount data are stored in the rate distortion memory 203 after being assigned indexes such as tile numbers, decomposition levels, subbands, code blocks, and encoding passes.

  Next, the slope calculation means 301 in the rate control means 107 calculates the slope S, that is, the code quantity versus distortion, for each coding pass with respect to the code amount data R and distortion data D stored in the rate distortion memory 203. calculate. The calculated slope S is stored in the position of the rate distortion memory 203 corresponding to the distortion data and code amount data of the same coding pass. A data configuration example of the rate distortion memory 203 at this time is as shown in FIG. The flag will be described later. In FIG. 2, the slope calculation unit 301 is described as being included in the rate control unit 107, but may be included in the rate control information extraction unit 106.

  When entropy coding and rate control information extraction are completed for all code blocks, the rate control unit 107 performs code amount control so that a desired code amount is obtained. Since the encoded data for all the encoding passes has already been stored in the code memory 105, the rate control means 107 determines the final encoding up to which encoding pass in each code block so that the desired code amount is obtained. In other words, it is determined which encoding pass is included in the encoded data. In each code block, the boundary between the encoding pass employed for the final code data and the encoding pass to be cut off is referred to as a cut-off point.

  When the gradients can be calculated for all paths, the truncation point deriving unit 302 finds a truncation point for which all the code blocks have a minimum distortion (maximum reduction in distortion) below the target code amount Rmax. To start. This optimization problem can be realized by using Lagrange's undetermined multiplier method. Specifically, when the code amount up to the truncation point ni in a code block i is R (i, ni) and the distortion is D (i, ni), Σ (R (i, ni) + λD (i, ni) )) Finds the cut-off point that minimizes, and obtains a variable λ such that the total code amount Rsum at that time becomes the target code amount Rmax. As shown in FIG. 7, when a code amount R and distortion D are represented in a graph (hereinafter referred to as an RD curve), a truncation point where the variable λ is tangent is found, and the code amount of the truncation point is found. This is the same as adjusting the variable λ until the sum Rsum of R becomes Rmax. In FIG. 7, in two code blocks c1 and c2, truncation points at which the slope of the tangent line is λ are nc1 and nc2, and the code amounts up to the truncation points are R (c1, nc1) and R (c2, nc2). Represents that. Such a code amount R is added to all code blocks, and a process of comparing with Rmax is performed.

Next, a specific example of the λ adjustment method will be described with reference to the flowchart of FIG. First, a lower limit value λmin and an upper limit value λmax are defined, and a variable λ that satisfies λ = (λmin + λmax) / 2 is calculated. In the code block i, the maximum truncation point ni satisfying S (i, ni) ≧ λ is obtained from the relationship between the slope S (i, ni) and λ with respect to the truncation point ni, and the code amount is represented by R (i, ni ). ΣR (i, ni) = Rsum is calculated for all code blocks. here,
If Rsum ≧ Rmax, λmin = λ,
If Rsum <Rmax, λmax = λ.
When new λmin and λmax are obtained, λ = (λmin + λmax) / 2 is calculated in the same manner as before, and the same is repeated thereafter. The above processing is repeated until λ ≧ λmax. Thus, the cut-off point ni for the finally obtained λ becomes the cut-off point to be obtained.
The code extraction means 303 collects code data of a coding pass higher than the truncation point ni in each code block from the code memory 105 to create code data having a minimum distortion below the desired code amount Rmax.

  In FIG. 7, the slope S of the RD curve decreases monotonously with respect to the code amount R, but actually does not monotonously decrease as shown in FIG. There may be multiple points that are tangents. In such a case, R (i, ni) + λD (i, ni) must be calculated for each point to find the point that minimizes it. However, in this case, the number of calculations is increased, which is not preferable in terms of mounting. Therefore, the slope calculation means 301 not only calculates the slope, but also corrects the slope S so that the code amount R monotonously decreases in one code block, and adjusts the number of calculations. This has been described in the prior art method, but in the present invention, further improvement can be achieved by the processing described below.

In the inclination calculation means 301 in the first embodiment, the monotonous decrease correction process is performed in the following program steps.
(1) Set Ni = {n} (ie the set of all truncation point)
(2) Set p = 0, m = 1
(3) For k = m, m + 1, m + 2, m + 3, ..., kmax
If k belongs to Ni
Set ΔR (i, k) = R (i, k) -R (i, p), and ΔD (i, k) = D (i, p) -D (i, k)
Set S (i, k) = ΔD (i, k) / ΔR (i, k)
If p ≠ 0 and S (i, k)> S (i, p),
then remove p from Ni, and set p = p-1, m = k, go to step (3)
Otherwise, set p = k
This is shown in a flowchart in FIG. Here, the operation of removing p from the set Ni of candidates for cut-off points (remove p from Ni) implements the consent process using the flags indicating validity and invalidity in FIG. This flag is a signal stored in the “flag” column of the rate distortion memory 203 described above with reference to FIG.

Next, a specific example of the monotonic decrease correction process shown in FIG. 11 will be described.
Consider a case where the relationship (RD curve) between the code amount before correction and distortion in one code block is as shown in FIG. Here, the cut point candidate k is in the range of 1 to 4 (kmax = 4). FIG. 11B shows the case of the method described in Non-Patent Document 1, and the numbers described on the line graph in the figure indicate the order in which the slope S is calculated. As can be seen from this figure, a total of seven divisions are performed, and the fifth and sixth divisions are the same values as the first and second calculations and are redundant. The monotonic decrease correction in the first embodiment is as shown in FIG. 11C, and the number of divisions is five, which can be implemented less than in the prior art. Consider a case where the RD curve before correction in another code block is as shown in FIG. This represents an example in which the number of divisions becomes the maximum number. The number of divisions is 10 times as shown in FIG. 11 (e) in the case of the conventional method, and FIG. 11 (f) in case of the method of the first embodiment. It will be 7 times. As can be seen from the above, in the first embodiment, the slope calculation unit 301 previously performed the slope calculation once for the same coding pass is excluded from the same calculation thereafter, and is shifted from the upper bit to the lower bit. Thus, correction processing is performed so that the slope of the code amount versus distortion monotonously decreases.

Therefore, in the first embodiment, the number of divisions is the maximum (2 kmax-1). Therefore, considering the difference from the prior art example,
kmax (kmax + 1) / 2- (2kmax-1)
= (Kmax ² -3kmax + 2) / 2
= (Kmax-1) (kmax-2) / 2
Thus, the above difference becomes positive when kmax> 2, and it can be seen that the number of divisions decreases. For kmax, for example, when the number of bit planes is 7, the number of coding passes is 6 × 3 + 1 = 19, and thus kmax is 19. The number of time divisions is 190 in the conventional method and 37 in the first embodiment, and when compared with the maximum number of times, it is possible to reduce the division processing by approximately 80%.

  As described above, according to the first embodiment, when calculating the slope of the code amount versus distortion for each coding pass, the slope calculation performed once for the same coding pass is the same as the following. Except for the calculation, correction processing is performed so that the slope of the code amount versus distortion decreases monotonically from the upper bit toward the lower bit. Therefore, compared to the conventional method, the number of divisions for calculating the slope of the code amount versus distortion is reduced. Since it can be significantly reduced, the effect of enabling high-speed encoding processing can be obtained. Therefore, for example, even when executed by a CPU that does not have sufficient processing capability, such as a digital still camera or a mobile phone, there is an effect that a processing time that sufficiently satisfies the functions can be obtained. Also, when implemented with hardware, the number of processing clocks can be reduced, so that the processing speed can be increased.

Embodiment 2. FIG.
The configuration on the block of the image coding apparatus according to the second embodiment is the same as that of the first embodiment, as shown in FIGS. 1 and 2, but the processing function of the inclination calculating means 301 is different. Specifically, the second embodiment slope calculation means 301 calculates the slope while performing the monotonic decrease correction process in the following program steps.
(1) Set Ni = {n} (ie the set of all truncation point)
(2) Set p = 0
(3) For k = 1, 2, 3, ..., kmax
If k belongs to Ni
Set ΔR (i, k) = R (i, k) -R (i, k-1), and ΔD (i, k) = D (i, k-1) -D (i, k)
Set S (i, k) = ΔD (i, k) / ΔR (i, k)
If p ≠ 0 and S (i, k)> S (i, k-1),
then remove (k-1) from Ni
Set ΔR (i, k) = R (i, k) -R (i, p-1), and ΔD (i, k) = D (i, p-1) -D (i, k)
Set S (i, k) = ΔD (i, k) / ΔR (i, k)
Otherwise, set p = k
(4) Set p = kmax
(5) For k = kmax-1, kmax-2, ..., 2, 1
If k belongs to Ni, and S (i, k) <S (i, p),
then remove k from Ni
Otherwise, Set p = k

  This process is shown in a flowchart in FIG. Steps (1) to (3) are the monotone decrease correction A in FIG. 12 and steps (4) to (5) are the monotone decrease correction B in FIG. The feature of the second embodiment is that although two scans are performed on the candidate k of the cut-off point, the feedback loop (go to step (3)) is eliminated as compared with the first embodiment. Since there is no feedback loop in this way, control complexity can be reduced.

Further, the number of divisions will be compared with the case of Embodiment 1 using a specific example shown in FIG. When considering the case of the maximum number of divisions as shown in FIG. 13A in the RD curve before correction in a certain code block, the method of the first embodiment is as shown in FIG. The method of 2 is also as shown in FIG. Next, consider an uncorrected RD curve as shown in FIG. In the case of the first embodiment, the number of divisions is six as shown in FIG. 13E, whereas in the case of the second embodiment, five divisions are sufficient as shown in FIG. 13F. become. FIG. 13 (f) shows the processing result of the monotonic decrease correction A in FIG. 12 (a), and it can be seen that the monotonous decrease cannot be completely corrected. Therefore, when the re-correction process shown in the monotonic decrease correction B in FIG. 12B is performed, the monotonic decrease can be completely achieved as shown in FIG. In FIG. 13G, since the first slope is smaller than the fourth slope, the first slope is invalidated, and only the fourth and fifth slopes are valid. The division process is not performed even once. In FIG. 13G, two curves are shown, and the upper side represents the result corrected by the monotonic phenomenon correction B. From this, it seems that the distortion has increased, but in the monotonic decrease correction process, only the slope is significant, and it is not meaningful here that the curve goes up and down. Here, the first slope is removed, and the curve is redrawn starting from 0, which is expressed for convenience.
Thus, it can be seen that the number of divisions is six in the first embodiment, but only five in the second embodiment. In the second embodiment, when the slope calculation unit 301 calculates the slope of the code amount versus distortion for each coding pass, if the slope calculated for a certain coding pass is larger than the slope of the immediately preceding pass, these are calculated. Instead of the path, calculate the slope of the path that connects the candidates for the previous and next truncation points, and remove the inappropriate truncation points so that they increase monotonically in the descending order of the truncation points. Correction processing is performed so that the slope of the code amount versus distortion monotonously decreases.

  As described above, according to the second embodiment, when the slope calculated for a certain coding path is larger than the slope of the immediately preceding path, the path of the path connecting the preceding and following cut point candidates instead of these paths. By calculating the slope and removing the inappropriate truncation point so that it increases monotonically in descending order of the truncation point, the slope of the code amount versus distortion decreases monotonically from the upper bit to the lower bit. Since the processing is performed, the number of divisions can be further reduced as compared with the case of the first embodiment, and an effect of enabling high-speed encoding processing can be obtained.

Embodiment 3 FIG.
The block configuration of the image coding apparatus according to the third embodiment is the same as that of the first embodiment, as shown in FIGS. 1 and 2, but the processing function of the inclination calculating means 301 is different. Specifically, the inclination calculation means 301 of the third embodiment calculates the inclination while performing monotonic decrease correction processing in the following program steps.
(1) Set Ni = {n} (ie the set of all truncation point)
(2) Set p = 0
(3) For k = 1, 2, 3, ..., kmax
Set ΔR (i, k) = R (i, k) -R (i, p), and ΔD (i, k) = D (i, p) -D (i, k)
Set S (i, k) = ΔD (i, k) / ΔR (i, k)
If p ≠ 0 and S (i, k)> S (i, p),
then remove k from Ni
Otherwise, set p = k

  This process is shown in a flowchart in FIG. First, after calculating the slope S (k) in the coding pass, it is compared with the slope S (p) of the coding pass that has become valid immediately before. If the current slope is smaller, the slope is determined to be valid. On the other hand, if the current inclination is large, the inclination is invalid. In the case of the third embodiment, since there is no feedback loop and only one scan is required for the cut-off point candidate k, the complexity of control can be greatly reduced. Also, the number of divisions is always kmax from FIG. Compared to the maximum number of times (2kmax-1) in the first embodiment, the number of divisions can be reduced by nearly 50%.

Next, a specific example of the monotonic decrease correction process shown in FIG. 15 will be described. Consider the RD curve before correction in a certain code block shown in FIG. FIG. 15B shows the calculation order of the slope, and it can be seen that the number of calculations is four. However, since the third slope is larger than the second slope, the third slope must be invalidated. Therefore, as shown in FIG. 15C, the first, second, and fourth inclinations become effective as a result of performing the monotonic decrease correction in the processing of the third embodiment.
As described above, according to the third embodiment, if the calculated slope of a certain coding pass is larger than the slope of the coding pass that has just been validated, an invalid slope is obtained. As a result, the correction processing is performed so that the slope of the code amount versus distortion monotonously decreases from the upper bit toward the lower bit, so that the number of divisions is significantly reduced compared to the case of the first embodiment. Therefore, the effect of enabling high-speed encoding processing can be obtained.

Embodiment 4 FIG.
The block configuration of the image coding apparatus according to the fourth embodiment is the same as that of the first embodiment, as shown in FIGS. 1 and 2, but the processing function of the inclination calculating means 301 is different. Specifically, the inclination calculation means 301 of the fourth embodiment calculates the inclination while performing monotonic decrease correction processing in the following program steps.
(1) Set Ni = {n} (ie the set of all truncation point)
(2) Set p = kmax
(3) For k = kmax-1, kmax-2, ..., 1, 0
Set ΔR (i, k) = R (i, k) -R (i, p), and ΔD (i, k) = D (i, p) -D (i, k)
Set S (i, k) = ΔD (i, k) / ΔR (i, k)
If p ≠ kmax and S (i, k) <S (i, p),
then remove k from Ni
Otherwise, set p = k

  FIG. 16 is a flowchart showing this process. The monotonic decrease correction process of the fourth embodiment is similar to the processing method of the third embodiment. In the third embodiment, the monotonic decrease correction is performed in the ascending order of the candidate k of the cut point. The fourth embodiment differs in that correction processing is performed so that the slope of the code amount versus distortion monotonously decreases as a result of processing in descending order of the candidate k of the truncation points from the higher bits to the lower bits. . Accordingly, the number of divisions is always kmax.

Next, the reason why processing is performed in descending order in the fourth embodiment will be described.
Consider the RD curve before correction in a certain code block shown in FIG. FIG. 17B shows an example in which correction processing is performed by the method of the third embodiment. When processing in ascending order, there is no slope smaller than the first slope, so only the first slope is effective. In the rate control, as described in FIG. 17A, when a variable λ that is slightly larger than the first gradient is derived, all the coding pass data for this code block is discarded. As a result, image quality deterioration is likely to occur. On the other hand, when the correction is performed in the fourth embodiment, it is understood that only the first slope becomes invalid when processing is performed in descending order of the candidate k of the cut point as shown in FIG. . Here, as assumed in FIG. 17B, when λ having a slope slightly larger than the first slope is derived in the rate control, all coding passes are included in the final coded data. Therefore, it can be seen that the image quality is not deteriorated.

  As described above, according to the fourth embodiment, processing is performed according to the descending order of the cut-off point candidates k, and as a result, the slope of the code amount versus distortion decreases monotonously from the upper bit toward the lower bit. Thus, the control is simplified as in the third embodiment, and at the same time, the number of divisions can be greatly reduced, and the effect of enabling high-speed encoding processing can be obtained. In addition, an effect that image quality deterioration can be suppressed as compared with the third embodiment can be obtained.

Embodiment 5 FIG.
In the first to fourth embodiments described so far, the truncation point candidate k is for all coding passes. In this fifth embodiment, the truncation point candidate is a bit plane. Is. In the embodiments so far, the accumulated distortion and the code amount are calculated for each coding pass. In Embodiment 5, these calculations are performed for each bit plane. That is, the distortion calculation unit 201 calculates distortion data for each bit plane, and the code amount calculation unit 202 calculates code amount data for each bit plane. In addition, the inclination calculation unit 301 calculates the inclination of the code amount versus distortion based on the code amount and distortion of the bit plane, and the rate distortion memory 203 stores distortion data, code amount data, and the inclination of each bit plane. .
Here, when the number of bit planes is B and the number of coding passes is CP, CP = kmax = 1 + 3 (B−1). As a result, the number of divisions can be reduced to nearly 1/3 compared to the method of calculating for each path. Therefore, the capacity of the rate distortion memory 203 can also be reduced to about 1/3 because information such as distortion and code amount is stored for each bit plane.

  As described above, according to the fifth embodiment, in the first to fourth embodiments, the truncation point candidate is a bit plane instead of the coding pass. The number of divisions can be greatly reduced, and the effect of enabling high-speed encoding processing can be obtained. In addition, an effect of reducing the memory scale of the rate distortion memory can be obtained.

Embodiment 6 FIG.
According to the method of the fifth embodiment, the speed can be significantly increased and the memory scale can be reduced. However, since the data of the three encoding passes are combined into one, the controllable increment is increased, and the code amount is increased. There is concern that the control accuracy will drop. Along with this, there may be a case where the image quality is slightly deteriorated. Therefore, the sixth embodiment is characterized in that a cut-off point candidate k is selected so as not to deteriorate the control accuracy.

  FIG. 18 is a block diagram showing a functional configuration of the rate control information extracting means 106 according to the sixth embodiment. In this configuration, a cut-off point setting unit 1901 is newly added to the configuration of the rate control information extracting unit 106 in FIG. This truncation point setting means 1901 prepares a plurality of types of truncation point candidate sets, and selects a truncation point candidate set that is most suitable for the current encoding. Therefore, the distortion calculation unit 201 calculates the distortion according to the selected cut-off point candidate. Similarly, the code amount calculation unit 202 also calculates the code amount according to the candidate for the cut point selected from the cut point setting unit 1901.

Next, with reference to FIG. 19, candidates for cut-off points will be described in comparison with the other embodiments.
In the figure, the numeral indicates a bit plane number, the subsequent S indicates a Significant Propagation Decoding Pass, M indicates a Magnitude Refinement Pass, and C indicates a Cleanup Pass. In addition, the horizontal arrow is a candidate for a cut-off point. FIG. 19A shows a case where truncation point candidates are set in the coding pass, and the first to fourth embodiments correspond to this. FIG. 19B shows a case where a cut-off point candidate is a bit plane, and the fifth embodiment corresponds to this. FIG. 19C shows an example of the sixth embodiment, in which candidates for the cut-off point are set finely in the lower bits and coarsely in the upper bits. FIG. 19 (d) shows another example in the sixth embodiment, in which candidates for truncation points are set coarsely in the lower bits and finely in the upper bits.

  For example, when the user sets the encoding rate high, in other words, when the compression rate is set low, the normal truncation point is on the lower bit side. In such a case, the step of the lower bit is fine. A candidate for a cut-off point as shown in (c) is set. On the other hand, when the encoding rate is set low, in other words, when the compression rate is set high, the normal truncation point is on the upper bit side. A candidate for a cut-off point as shown in (d) is set. In this way, by preparing a plurality of candidates for switching points and switching to an optimum one according to the encoding rate, it is possible to finely cut off the truncation points around the original truncation points. It becomes like this. Therefore, it is possible to prevent the control accuracy from being deteriorated and to suppress the image quality deterioration. In the above description, the set of truncation point candidates is switched according to the encoding rate. In addition to this, switching according to the separation level, subband, color component, and the like is performed. Furthermore, it is possible to suppress deterioration of control accuracy and to suppress deterioration of image quality.

In the sixth embodiment, as in the fifth embodiment, the number of cut-off point candidates can be reduced as compared with the first to fourth embodiments, so that the number of divisions can be reduced. Needless to say, there is an effect of speeding up the processing. In addition, since the calculation of distortion data and code amount data stored in the rate distortion memory 203 can be reduced as in the fifth embodiment, an effect of reducing the memory scale can be obtained.
Needless to say, by combining with the methods shown in the first to fourth embodiments, the number of divisions can be further reduced and the speed can be increased.

It is a block diagram which shows the structure of the image coding apparatus in each embodiment of this invention. It is a block diagram which shows the detailed structure of the rate control information extraction means and rate control means which concern on the first embodiment. It is explanatory drawing which shows a subband when the wavelet transformation is performed to the decomposition level 2 by the wavelet transformation means according to the first embodiment. It is explanatory drawing regarding the bit plane converted by the coefficient modeling means based on the first embodiment. It is explanatory drawing which shows a mode that a bit plane is decomposed | disassembled into an encoding pass by the entropy encoding means which concerns on the same Embodiment 1. FIG. 3 is an explanatory diagram showing a data configuration of a rate distortion memory according to the first embodiment. FIG. It is explanatory drawing which shows the relationship (RD curve) of the code amount versus distortion for demonstrating the method to derive | lead-out the truncation point which concerns on the same Embodiment 1. FIG. 6 is a flowchart for calculating an optimal code amount versus distortion gradient (λ) for deriving a truncation point according to the first embodiment. It is explanatory drawing which shows the RD curve from which the relationship of code amount versus distortion is not monotonously decreasing. 7 is a flowchart showing a procedure for correcting the slope of the code amount versus distortion to monotonically decrease according to the first embodiment. It is explanatory drawing which shows the specific example of the monotone decrease correction process which concerns on the same Embodiment 1. FIG. 10 is a flowchart showing a procedure for correcting the slope of the code amount versus distortion according to the second embodiment to a monotone decrease. It is explanatory drawing which shows the specific example of the monotone decrease correction process which concerns on Embodiment 2 of this invention. It is a flowchart which shows the procedure which correct | amends the inclination of the code amount versus distortion which concerns on Embodiment 3 of this invention to monotone decrease. It is explanatory drawing which shows the specific example of the monotone decrease correction process which concerns on Embodiment 3 of this invention. It is a flowchart which shows the procedure which correct | amends the inclination of the code amount versus distortion which concerns on Embodiment 4 of this invention to monotone decrease. It is explanatory drawing which shows the specific example of the monotone decrease correction process which concerns on Embodiment 4 of this invention. It is a block diagram which shows the structure of the rate control information extraction means which concerns on Embodiment 6 of this invention. It is explanatory drawing shown about the candidate of the truncation point which concerns on Embodiment 6 of this invention. It is explanatory drawing shown about a wavelet transformation. It is a flowchart which shows the procedure of the conventional monotone decrease correction process.

Explanation of symbols

  DESCRIPTION OF SYMBOLS 101 Wavelet transformation means, 102 Quantization means, 103 Coefficient modeling means, 104 Entropy encoding means, 105 Code memory, 106 Rate control information extraction means, 107 Rate control means, 201 Distortion calculation means, 202 Code amount calculation means, 203 Rate Distortion memory, 301 slope calculation means, 302 truncation point derivation means, 303 code extraction means, 1901 truncation point setting means.

Claims (30)

Two-dimensional wavelet transform is performed on the input image signal to generate a wavelet transform coefficient for each subband. The wavelet transform coefficient is quantized with a quantization step size set for each subband, and each quantized subband is quantized. The wavelet transform coefficients are divided into code blocks, the multi-value data of each code block is decomposed into binary bit planes, then divided into coding passes, and entropy coding is performed for each divided coding pass. The distortion and code amount are calculated for each divided coding pass, the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion for each coding pass, and the code amount for each code block Using the distortion and the slope of the code amount versus distortion, the truncation point that minimizes the distortion within the target code amount is derived, and from the truncation point derived in each code block An image encoding device for creating a code data having the minimum distortion from the code data within the target code amount of position encoding pass,
If the slope calculation means used to calculate the slope of the code amount versus distortion for each coding pass is greater than the slope of the coding path that was just valid, the slope is invalid. On the other hand, an image coding apparatus characterized in that correction processing is performed so that the slope of the code amount versus distortion monotonously decreases from the upper bit toward the lower bit by setting the effective slope as small as possible. Two-dimensional wavelet transform is performed on the input image signal to generate a wavelet transform coefficient for each subband. The wavelet transform coefficient is quantized with a quantization step size set for each subband, and each quantized subband is quantized. The wavelet transform coefficients are divided into code blocks, the multi-value data of each code block is decomposed into binary bit planes, then divided into coding passes, and entropy coding is performed for each divided coding pass. The distortion and code amount are calculated for each divided coding pass, the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion for each coding pass, and the code amount for each code block Using the distortion and the slope of the code amount versus distortion, the truncation point that minimizes the distortion within the target code amount is derived, and from the truncation point derived in each code block An image encoding device for creating a code data having the minimum distortion from the code data within the target code amount of position encoding pass,
The slope calculation means used for calculating the slope of the code amount versus distortion for each coding pass processes the descending order of the candidates for the cutoff point with respect to the slope of the code amount versus distortion, and as a result, from the upper bits. An image encoding apparatus that performs correction processing so that the slope of the code amount versus distortion monotonously decreases toward lower bits. Two-dimensional wavelet transform is performed on the input image signal to generate a wavelet transform coefficient for each subband. The wavelet transform coefficient is quantized with a quantization step size set for each subband, and each quantized subband is quantized. The wavelet transform coefficients are divided into code blocks, the multi-value data of each code block is decomposed into binary bit planes, then divided into coding passes, and entropy coding is performed for each divided coding pass. The distortion and code amount are calculated for each divided coding pass, the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion for each coding pass, and the code amount for each code block Using the distortion and the slope of the code amount versus distortion, the truncation point that minimizes the distortion within the target code amount is derived, and from the truncation point derived in each code block An image encoding device for creating a code data having the minimum distortion from the code data within the target code amount of position encoding pass,
The slope calculation means used to calculate the slope of the code amount versus distortion for each coding pass excludes the slope calculation performed once for the same coding pass from the same calculation thereafter, from the upper bit to the lower order. Perform correction processing so that the slope of the code amount versus distortion monotonously decreases toward the bit,
Wherein the candidate truncation point, picture coding apparatus you characterized in that the bit planes instead of coding passes. Two-dimensional wavelet transform is performed on the input image signal to generate a wavelet transform coefficient for each subband. The wavelet transform coefficient is quantized with a quantization step size set for each subband, and each quantized subband is quantized. The wavelet transform coefficients are divided into code blocks, the multi-value data of each code block is decomposed into binary bit planes, then divided into coding passes, and entropy coding is performed for each divided coding pass. The distortion and code amount are calculated for each divided coding pass, the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion for each coding pass, and the code amount for each code block Using the distortion and the slope of the code amount versus distortion, the truncation point that minimizes the distortion within the target code amount is derived, and from the truncation point derived in each code block An image encoding device for creating a code data having the minimum distortion from the code data within the target code amount of position encoding pass,
When the slope calculation means used to calculate the slope of the code amount versus distortion for each coding pass is larger than the slope of the previous pass when the slope calculated for a certain coding pass is larger than the slope of the previous pass, Calculate the slope of the path connecting the candidates for the truncation point, remove the inappropriate truncation point so that it increases monotonically in descending order of the truncation point, and monotonically the slope of the code amount versus distortion from the upper bit to the lower bit We perform correction processing to become decrease,
Wherein the candidate truncation point, picture coding apparatus you characterized in that the bit planes instead of coding passes. The image coding apparatus according to claim 1 or 2 , wherein a candidate for a cut-off point is a bit plane instead of a coding pass. Two-dimensional wavelet transform is performed on the input image signal to generate a wavelet transform coefficient for each subband. The wavelet transform coefficient is quantized with a quantization step size set for each subband, and each quantized subband is quantized. The wavelet transform coefficients are divided into code blocks, the multi-value data of each code block is decomposed into binary bit planes, then divided into coding passes, and entropy coding is performed for each divided coding pass. The distortion and code amount are calculated for each divided coding pass, the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion for each coding pass, and the code amount for each code block Using the distortion and the slope of the code amount versus distortion, the truncation point that minimizes the distortion within the target code amount is derived, and from the truncation point derived in each code block An image encoding device for creating a code data having the minimum distortion from the code data within the target code amount of position encoding pass,
The slope calculation means used to calculate the slope of the code amount versus distortion for each coding pass excludes the slope calculation performed once for the same coding pass from the same calculation thereafter, from the upper bit to the lower order. Perform correction processing so that the slope of the code amount versus distortion monotonously decreases toward the bit,
A plurality of sets of candidates for the truncation point are set in advance, and includes a truncation point setting means for switching the set according to the encoding rate,
Instead of the divided coding pass, the distortion and code amount are calculated for the cut point candidates set by the cut point setting means, and the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion. the calculated, picture coding apparatus you characterized in that the time toward the upper bits to the lower bits corrected so that the slope is monotonically decreasing. Two-dimensional wavelet transform is performed on the input image signal to generate a wavelet transform coefficient for each subband. The wavelet transform coefficient is quantized with a quantization step size set for each subband, and each quantized subband is quantized. The wavelet transform coefficients are divided into code blocks, the multi-value data of each code block is decomposed into binary bit planes, then divided into coding passes, and entropy coding is performed for each divided coding pass. The distortion and code amount are calculated for each divided coding pass, the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion for each coding pass, and the code amount for each code block Using the distortion and the slope of the code amount versus distortion, the truncation point that minimizes the distortion within the target code amount is derived, and from the truncation point derived in each code block An image encoding device for creating a code data having the minimum distortion from the code data within the target code amount of position encoding pass,
When the slope calculation means used to calculate the slope of the code amount versus distortion for each coding pass is larger than the slope of the previous pass when the slope calculated for a certain coding pass is larger than the slope of the previous pass, Calculate the slope of the path connecting the candidates for the truncation point, remove the inappropriate truncation point so that it increases monotonically in descending order of the truncation point, and monotonically the slope of the code amount versus distortion from the upper bit to the lower bit We perform correction processing to become decrease,
A plurality of sets of candidates for the truncation point are set in advance, and includes a truncation point setting means for switching the set according to the encoding rate,
Instead of the divided coding pass, the distortion and code amount are calculated for the cut point candidates set by the cut point setting means, and the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion. the calculated, picture coding apparatus you characterized in that the time toward the upper bits to the lower bits corrected so that the slope is monotonically decreasing. A plurality of candidate sets of truncation points are set in advance, and includes a truncation point setting means for switching the set according to the encoding rate,
Instead of the divided coding pass, the distortion and code amount are calculated for the cut point candidates set by the cut point setting means, and the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion. The image encoding apparatus according to claim 1 or 2 , wherein the correction is performed so that the slope monotonously decreases from the upper bit toward the lower bit. Two-dimensional wavelet transform is performed on the input image signal to generate a wavelet transform coefficient for each subband. The wavelet transform coefficient is quantized with a quantization step size set for each subband, and each quantized subband is quantized. The wavelet transform coefficients are divided into code blocks, the multi-value data of each code block is decomposed into binary bit planes, then divided into coding passes, and entropy coding is performed for each divided coding pass. The distortion and code amount are calculated for each divided coding pass, and the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion for each coding pass. The correction processing is performed so that the slope of the code amount versus distortion decreases monotonically toward the target, and the code amount, distortion, and the slope of the code amount versus distortion are used for each code block. Deriving a truncation point where distortion becomes minimum in the issue amount, to create a code data having the minimum distortion in the target code amount from the code data of each code upper than truncated point derived in block coding pass In an image encoding device,
The slope calculation means used to calculate the slope of the code amount versus distortion for each coding pass excludes the slope calculation performed once for the same coding pass from the same calculation thereafter, from the upper bit to the lower order. Perform correction processing so that the slope of the code amount versus distortion monotonously decreases toward the bit,
A plurality of sets of candidates for the truncation point are set in advance, and includes a truncation point setting means for switching the set according to the encoding rate,
Instead of the divided coding pass, the distortion and code amount are calculated for the cut point candidates set by the cut point setting means, and the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion. the calculated, picture coding apparatus you characterized in that the time toward the upper bits to the lower bits corrected so that the slope is monotonically decreasing. Two-dimensional wavelet transform is performed on the input image signal to generate a wavelet transform coefficient for each subband. The wavelet transform coefficient is quantized with a quantization step size set for each subband, and each quantized subband is quantized. The wavelet transform coefficients are divided into code blocks, the multi-value data of each code block is decomposed into binary bit planes, then divided into coding passes, and entropy coding is performed for each divided coding pass. The distortion and code amount are calculated for each divided coding pass, and the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion for each coding pass. The correction processing is performed so that the slope of the code amount versus distortion decreases monotonically toward the target, and the code amount, distortion, and the slope of the code amount versus distortion are used for each code block. Deriving a truncation point where distortion becomes minimum in the issue amount, to create a code data having the minimum distortion in the target code amount from the code data of each code upper than truncated point derived in block coding pass In an image encoding device,
When the slope calculation means used to calculate the slope of the code amount versus distortion for each coding pass is larger than the slope of the previous pass when the slope calculated for a certain coding pass is larger than the slope of the previous pass, Calculate the slope of the path connecting the candidates for the truncation point, remove the inappropriate truncation point so that it increases monotonically in descending order of the truncation point, and monotonically the slope of the code amount versus distortion from the upper bit to the lower bit We perform correction processing to become decrease,
A plurality of sets of candidates for the truncation point are set in advance, and includes a truncation point setting means for switching the set according to the encoding rate,
Instead of the divided coding pass, the distortion and code amount are calculated for the cut point candidates set by the cut point setting means, and the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion. the calculated, picture coding apparatus you characterized in that the time toward the upper bits to the lower bits corrected so that the slope is monotonically decreasing. Two-dimensional wavelet transform is performed on the input image signal to generate a wavelet transform coefficient for each subband. The wavelet transform coefficient is quantized with a quantization step size set for each subband, and each quantized subband is quantized. The wavelet transform coefficients are divided into code blocks, the multi-value data of each code block is decomposed into binary bit planes, then divided into coding passes, and entropy coding is performed for each divided coding pass. The distortion and code amount are calculated for each divided coding pass, and the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion for each coding pass. The correction processing is performed so that the slope of the code amount versus distortion decreases monotonically toward the target, and the code amount, distortion, and the slope of the code amount versus distortion are used for each code block. Deriving a truncation point where distortion becomes minimum in the issue amount, to create a code data having the minimum distortion in the target code amount from the code data of each code upper than truncated point derived in block coding pass In an image encoding device,
A plurality of candidate sets of truncation points are set in advance, and includes a truncation point setting means for switching the set according to the encoding rate,
Instead of the divided coding pass, the distortion and code amount are calculated for the cut point candidates set by the cut point setting means, and the slope of the code amount versus distortion is calculated based on the obtained code amount and distortion. The image encoding apparatus according to claim 1 or 2 , wherein the correction is performed so that the slope monotonously decreases from the upper bit toward the lower bit. The image coding apparatus according to any one of claims 6 to 11, wherein the truncation point setting means switches a set of truncation point candidates according to a coding rate and a decomposition level. 12. The image encoding device according to claim 6 , wherein the truncation point setting means switches the set of truncation point candidates according to the encoding rate, the decomposition level, and the subband information. . The image code according to any one of claims 6 to 11, wherein the truncation point setting means switches a set of truncation point candidates according to a coding rate, a decomposition level, subband information, and a color component. Device. Two-dimensional wavelet transform of the input image signal to generate wavelet transform coefficients for each subband,
This wavelet transform coefficient is quantized with the quantization step size set for each subband,
Divide the quantized wavelet transform coefficients of each subband into code blocks,
After the multi-value data of each code block is decomposed into binary bit planes, it is divided into coding passes,
Entropy coding is performed for each divided coding pass, distortion and code amount are calculated for each divided coding pass,
Calculate the slope of the code amount versus distortion based on the obtained code amount and distortion for each coding pass,
In this slope calculation, the slope calculation performed once for the same coding pass is excluded from the same calculation thereafter, and the slope of the code amount versus distortion decreases monotonously from the upper bit to the lower bit. The correction process is performed so that
Deriving a truncation point that minimizes distortion within the target code amount using the code amount, distortion, and the gradient of code amount versus distortion for each code block,
Creating code data having a minimum distortion within the target code amount from code data of a coding pass higher than a truncation point derived in each code block;
Wherein the correction process at the time of slope calculations, images coding method you characterized by being performed by the following procedure.
(1) Set Ni = {n} (ie the set of all truncation point)
(2) Set p = 0, m = 1
(3) For k = m, m + 1, m + 2, m + 3, ..., kmax
If k belongs to Ni
Set ΔR (i, k) = R (i, k)-R (i, p), and ΔD (i, k) = D (i, p)-D (i, k)
Set S (i, k) = ΔD (i, k) / ΔR (i, k)
If p ≠ 0 and S (i, k)> S (i, p),
then remove p from Ni, and set p = p-1, m = k, go to step (3)
Otherwise, set p = k
Here, Ni is a set of coding passes of the code block i, k is a candidate for a truncation point, p is a pass immediately before k, R is a code amount, D is distortion, and S is a slope. Two-dimensional wavelet transform of the input image signal to generate wavelet transform coefficients for each subband,
This wavelet transform coefficient is quantized with the quantization step size set for each subband,
Divide the quantized wavelet transform coefficients of each subband into code blocks,
After the multi-value data of each code block is decomposed into binary bit planes, it is divided into coding passes,
Entropy coding is performed for each divided coding pass,
Calculate distortion and code amount for each divided coding pass,
Calculate the slope of the code amount versus distortion based on the obtained code amount and distortion for each coding pass,
If the slope calculated for a certain coding path is larger than the slope of the previous path during this slope calculation, the slope of the path connecting the candidates for the previous and next cut points is calculated instead of these paths, and the truncation is performed. Remove the inappropriate truncation point so that it increases monotonically in descending order of points, and perform correction processing so that the slope of the code amount versus distortion monotonously decreases from the upper bit to the lower bit,
Deriving a truncation point that minimizes distortion within the target code amount using the code amount, distortion, and the gradient of code amount versus distortion for each code block,
Creating code data having a minimum distortion within the target code amount from code data of a coding pass higher than a truncation point derived in each code block;
Wherein the correction process at the time of slope calculations, images coding method you characterized by being performed by the following procedure.
(1) Set Ni = {n} (ie the set of all truncation point)
(2) Set p = 0
(3) For k = 1, 2, 3, ..., kmax
If k belongs to Ni
Set ΔR (i, k) = R (i, k)-R (i, k-1), and ΔD (i, k) = D (i, k-1)-D (i, k)
Set S (i, k) = ΔD (i, k) / ΔR (i, k)
If p ≠ 0 and S (i, k)> S (i, k-1),
then remove (k-1) from Ni
Set ΔR (i, k) = R (i, k)-R (i, p-1), and ΔD (i, k) = D (i, p-1)-D (i, k)
Set S (i, k) = ΔD (i, k) / ΔR (i, k)
Otherwise, set p = k
(4) Set p = kmax
(5) For k = kmax-1, kmax-2, ..., 2, 1
If k belongs to Ni, and S (i, k) <S (i, p),
then remove k from Ni
Otherwise, Set p = k
Here, Ni is a set of coding passes of the code block i, k is a candidate for a truncation point, p is a pass immediately before k, R is a code amount, D is distortion, and S is a slope. Two-dimensional wavelet transform of the input image signal to generate wavelet transform coefficients for each subband,
This wavelet transform coefficient is quantized with the quantization step size set for each subband,
Divide the quantized wavelet transform coefficients of each subband into code blocks,
After the multi-value data of each code block is decomposed into binary bit planes, it is divided into coding passes,
Entropy coding is performed for each divided coding pass, distortion and code amount are calculated for each divided coding pass,
Calculate the slope of the code amount versus distortion based on the obtained code amount and distortion for each coding pass,
When calculating the slope, if the slope of a certain coding pass is larger than the slope of the coding pass that was valid immediately before, the slope is invalid. The correction processing is performed so that the slope of the code amount versus distortion decreases monotonously,
Deriving a truncation point that minimizes distortion within the target code amount using the code amount, distortion, and the gradient of code amount versus distortion for each code block,
An image coding method for creating code data having a minimum distortion within the target code amount from code data of a coding pass higher than a cut-off point derived in each code block. The image encoding method according to claim 17 , wherein the correction process in calculating the inclination is performed according to the following processing procedure.
(1) Set Ni = {n} (ie the set of all truncation point)
(2) Set p = 0
(3) For k = 1, 2, 3, ..., kmax
Set ΔR (i, k) = R (i, k)-R (i, p), and ΔD (i, k) = D (i, p)-D (i, k)
Set S (i, k) = ΔD (i, k) / ΔR (i, k)
If p ≠ 0 and S (i, k)> S (i, p),
then remove k from Ni
Otherwise, set p = k
Here, Ni is a set of coding passes of the code block i, k is a candidate for a truncation point, p is a pass immediately before k, R is a code amount, D is distortion, and S is a slope. Two-dimensional wavelet transform of the input image signal to generate wavelet transform coefficients for each subband,
This wavelet transform coefficient is quantized with the quantization step size set for each subband,
Divide the quantized wavelet transform coefficients of each subband into code blocks,
After the multi-value data of each code block is decomposed into binary bit planes, it is divided into coding passes,
Entropy coding is performed for each divided coding pass,
Calculate distortion and code amount for each divided coding pass,
Calculate the slope of the code amount versus distortion based on the obtained code amount and distortion for each coding pass,
When calculating the slope, processing is performed in descending order with respect to the slope of the code amount versus distortion, and as a result, the slope of the code quantity versus distortion is monotonously decreased from the upper bit toward the lower bit. Process,
Deriving a truncation point that minimizes distortion within the target code amount using the code amount, distortion, and the gradient of code amount versus distortion for each code block,
An image coding method for creating code data having a minimum distortion within the target code amount from code data of a coding pass higher than a cut-off point derived in each code block. The image encoding method according to claim 17 , wherein the correction process in calculating the inclination is performed according to the following processing procedure.
(1) Set Ni = {n} (ie the set of all truncation point)
(2) Set p = kmax
(3) For k = kmax-1, kmax-2, ..., 1, 0
Set ΔR (i, k) = R (i, k)-R (i, p), and ΔD (i, k) = D (i, p)-D (i, k)
Set S (i, k) = ΔD (i, k) / ΔR (i, k)
If p ≠ kmax and S (i, k) <S (i, p),
then remove k from Ni
Otherwise, set p = k
Here, Ni is a set of coding passes of the code block i, k is a candidate for a truncation point, p is a pass immediately before k, R is a code amount, D is distortion, and S is a slope. Two-dimensional wavelet transform of the input image signal to generate wavelet transform coefficients for each subband,
This wavelet transform coefficient is quantized with the quantization step size set for each subband,
Divide the quantized wavelet transform coefficients of each subband into code blocks,
After the multi-value data of each code block is decomposed into binary bit planes, it is divided into coding passes,
Entropy coding is performed for each divided coding pass, distortion and code amount are calculated for each divided coding pass,
Calculate the slope of the code amount versus distortion based on the obtained code amount and distortion for each coding pass,
In this slope calculation, the slope calculation performed once for the same coding pass is excluded from the same calculation thereafter, and the slope of the code amount versus distortion decreases monotonously from the upper bit to the lower bit. The correction process is performed so that
Deriving a truncation point that minimizes distortion within the target code amount using the code amount, distortion, and the gradient of code amount versus distortion for each code block,
Creating code data having a minimum distortion within the target code amount from code data of a coding pass higher than a truncation point derived in each code block;
Wherein the candidate truncation point, picture coding apparatus you characterized in that the bit planes instead of coding passes. Two-dimensional wavelet transform of the input image signal to generate wavelet transform coefficients for each subband,
This wavelet transform coefficient is quantized with the quantization step size set for each subband,
Divide the quantized wavelet transform coefficients of each subband into code blocks,
After the multi-value data of each code block is decomposed into binary bit planes, it is divided into coding passes,
Entropy coding is performed for each divided coding pass,
Calculate distortion and code amount for each divided coding pass,
Calculate the slope of the code amount versus distortion based on the obtained code amount and distortion for each coding pass,
If the slope calculated for a certain coding path is larger than the slope of the previous path during this slope calculation, the slope of the path connecting the candidates for the previous and next cut points is calculated instead of these paths, and the truncation is performed. Remove the inappropriate truncation point so that it increases monotonically in descending order of points, and perform correction processing so that the slope of the code amount versus distortion monotonously decreases from the upper bit to the lower bit,
Deriving a truncation point that minimizes distortion within the target code amount using the code amount, distortion, and the gradient of code amount versus distortion for each code block,
Creating code data having a minimum distortion within the target code amount from code data of a coding pass higher than a truncation point derived in each code block;
Wherein the candidate truncation point, picture coding apparatus you characterized in that the bit planes instead of coding passes. Candidate truncation point, the image encoding apparatus of any one of claims 20 to claim 15, characterized in that a bit plane in place of the coding pass. Two-dimensional wavelet transform of the input image signal to generate wavelet transform coefficients for each subband,
This wavelet transform coefficient is quantized with the quantization step size set for each subband,
Divide the quantized wavelet transform coefficients of each subband into code blocks,
After the multi-value data of each code block is decomposed into binary bit planes, it is divided into coding passes,
Entropy coding is performed for each divided coding pass, distortion and code amount are calculated for each divided coding pass,
Calculate the slope of the code amount versus distortion based on the obtained code amount and distortion for each coding pass,
In this slope calculation, the slope calculation performed once for the same coding pass is excluded from the same calculation thereafter, and the slope of the code amount versus distortion decreases monotonously from the upper bit to the lower bit. The correction process is performed so that
Deriving a truncation point that minimizes distortion within the target code amount using the code amount, distortion, and the gradient of code amount versus distortion for each code block,
Creating code data having a minimum distortion within the target code amount from code data of a coding pass higher than a truncation point derived in each code block;
A plurality of sets of candidates for the truncation point are set in advance, and the sets are switched according to the coding rate. Instead of the divided coding pass, distortion and code are set for the set candidates for the truncation point. The amount of code is calculated, the slope of code amount versus distortion is calculated based on the obtained code amount and distortion, and at that time, the slope is corrected so as to monotonously decrease from the upper bit toward the lower bit. images coding method you. Two-dimensional wavelet transform of the input image signal to generate wavelet transform coefficients for each subband,
This wavelet transform coefficient is quantized with the quantization step size set for each subband,
Divide the quantized wavelet transform coefficients of each subband into code blocks,
After the multi-value data of each code block is decomposed into binary bit planes, it is divided into coding passes,
Entropy coding is performed for each divided coding pass,
Calculate distortion and code amount for each divided coding pass,
Calculate the slope of the code amount versus distortion based on the obtained code amount and distortion for each coding pass,
If the slope calculated for a certain coding path is larger than the slope of the previous path during this slope calculation, the slope of the path connecting the candidates for the previous and next cut points is calculated instead of these paths, and the truncation is performed. Remove the inappropriate truncation point so that it increases monotonically in descending order of points, and perform correction processing so that the slope of the code amount versus distortion monotonously decreases from the upper bit to the lower bit,
Deriving a truncation point that minimizes distortion within the target code amount using the code amount, distortion, and the gradient of code amount versus distortion for each code block,
Creating code data having a minimum distortion within the target code amount from code data of a coding pass higher than a truncation point derived in each code block;
A plurality of sets of candidates for the truncation point are set in advance, and the sets are switched according to the coding rate. Instead of the divided coding pass, distortion and code are set for the set candidates for the truncation point. The amount of code is calculated, the slope of code amount versus distortion is calculated based on the obtained code amount and distortion, and at that time, the slope is corrected so as to monotonously decrease from the upper bit toward the lower bit. images coding method you. A plurality of truncation point candidate sets are set in advance, and the set is switched according to the coding rate. Instead of the divided coding pass, distortion and code amount are set for the set truncation point candidates. And calculating the slope of the code amount versus distortion based on the obtained code amount and distortion, and correcting the slope so that the slope decreases monotonically from the upper bit to the lower bit. The image encoding method according to any one of claims 15 to 20 . Two-dimensional wavelet transform of the input image signal to generate wavelet transform coefficients for each subband,
This wavelet transform coefficient is quantized with the quantization step size set for each subband,
Divide the quantized wavelet transform coefficients of each subband into code blocks,
After the multi-value data of each code block is decomposed into binary bit planes, it is divided into coding passes,
Entropy coding is performed for each divided coding pass, distortion and code amount are calculated for each divided coding pass,
A set of a plurality of truncation point candidates set in advance is switched according to the encoding rate, and distortion and code amount are calculated for the set truncation point candidates,
Calculate the slope of the code amount versus distortion based on the obtained code amount and distortion for each coding pass,
In this slope calculation, the slope calculation performed once for the same coding pass is excluded from the same calculation thereafter, and the slope of the code amount versus distortion decreases monotonously from the upper bit to the lower bit. The correction process is performed so that
Deriving a truncation point that minimizes distortion within the target code amount using the code amount, distortion, and the gradient of code amount versus distortion for each code block,
An image coding method for creating code data having a minimum distortion within the target code amount from code data of a coding pass higher than a cut-off point derived in each code block. The image encoding method according to any one of claims 24 to 27, wherein a set of candidates for truncation points is switched according to an encoding rate and a decomposition level. The image encoding method according to any one of claims 24 to 27, wherein a set of candidates for cut-off points is switched according to an encoding rate, a decomposition level, and subband information. The image encoding method according to any one of claims 24 to 27, wherein a set of candidates for cut-off points is switched according to an encoding rate, a decomposition level, subband information, and a color component.

Images