JP2005150845A - Image data coding apparatus and method thereof, computer program and computer-readable storage medium - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To efficiently obtain a rate/distortion gradient in image coding with a simple operation, and to obtain image data in which a target coding quantity is compression-coded. <P>SOLUTION: Image data inputted from an image data input section 200 is subjected to wavelet transform and quantization by well-known procedures, code data strings in all paths of each code block are generated and stored in a code string storing section 206, and information indicating the relation between an increasing quantity of code data at each path position of each code block and distortion is stored in a code block information storing section 207. A bit searching processor 209 continuously decides target rate/distortion gradient information per bit from its upper bit to lower bit on the basis of the information stored in the section 207. A code string forming section 205 reads code data strings from the section 206 until the corresponding path position on the basis of the rate/distortion gradient information obtained by the section 209, and generates final coded data. <P>COPYRIGHT: (C)2005,JPO&NCIPI

Description

  The present invention relates to an image encoding technique for encoding frame data of a still image or a moving image.

  One technique for improving the efficiency of image coding is a rate / distortion optimization technique. The rate / distortion optimization method obtains an index value related to the generated code amount and image distortion for each of the multiple sections constituting the encoded data, and minimizes the total distortion index value based on the condition that the total code amount is equal to or less than the target value. Is intended.

  In JPEG2000 (ISO / IEC 15444), an international standard for still image coding established by standardization work in ISO / IEC JTC1 / SC29 / WG1, the coefficient of each subband obtained by wavelet transform is called a code block. It is configured to be divided into rectangular areas and encoded independently. Each code block is encoded several times (divided into several passes), and the rate / distortion optimization method is calculated by obtaining the generated code amount and the image distortion index value in units of passes. Application is considered. There is one application method of the above-described rate / distortion optimization method for implementing the JPEG2000 standard (Non-Patent Document 1).

Distortion in the entire image, where ni is the code truncation possible point of the code block Bi, the code amount of the code block Bi when the code is terminated at ni is Ri (ni), and the distortion index value is Di (ni) The index value D and the total code amount R are:
D = ΣDi (ni)
R = ΣRi (ni)
(Where Σ is a summing function with “i = 1, 2,..., Total number of code blocks” as a variable). As the distortion index value, various values representing the degree of image quality degradation such as a mean square error and a visually weighted mean square error can be used.

The goal of the rate / distortion optimization is to obtain a set of censoring points ni that minimize D under the condition that the total code amount Rmax is less than the target, that is, R ≦ Rmax. This optimization problem can be solved by the generalized Lagrange multiplier method (Non-Patent Document 2).
Σ (Di (ni) + λ × Ri (ni))
To the problem of minimizing The value of λ is adjusted so that the total code amount R is equal to or less than Rmax.

  Minimizing this equation becomes an individual minimization problem for each code block. Hereinafter, a simple algorithm for obtaining the code truncation point ni that minimizes Di (ni) + λ · Ri (ni) for the code block Bi will be described.

  FIG. 4 shows the flow of processing for determining the code truncation point ni for the code block Bi whose number of effective coding passes is ki_max. As shown in FIG. 4, first, the code cut-off point ni is initialized to 0 (step S401). Next, the variable k indicating the code censorable point of interest is set to 1 (step S402). Subsequently, the code amount increase ΔRi (k) and the distortion index value decrease ΔDi (k) when the code stop point of the code block Bi is moved from ni to k are obtained for the target code stoppable point k. (Step S403). ΔDi (k) / ΔRi (k) is calculated as the code amount versus distortion index value of the code between the code cutoff point ni and the target code cutoff possible point k, and compared with λ (step S404). Is updated to k (step S405). Next, 1 is added to k to lower the target code cutoff point by one (step S406), and the updated value of k is compared with the number of coding passes ki_max of this code block (step S407), k ≦ ki_max. If so, the updated k is repeated from step S403. In this way, if k> ki_max, the processing is terminated, and ni at the end point is determined as the code cutoff point of the target code block Bi at a given λ.

  Considering that this algorithm is implemented for various values of λ, it is more efficient to predetermine code truncation point candidates for code blocks.

  Since code truncation of code blocks is performed in units of coding passes, code truncation is basically possible at all coding pass boundaries. However, when the above-described code truncation point ni determination algorithm is applied, code truncation candidate points A value indicating the slope of the rate / distortion between them Si (k) = ΔDi (k) / ΔRi (k) is determined to be monotonically decreasing with k, and the candidate points for censoring are determined, and the coding path boundary does not satisfy the condition Is excluded from candidates for code censoring points.

  For example, as shown in FIG. 10, a code block encoded by four encoding passes is considered. The sign can be cut off at the boundary of the four paths (censorable points 0 to 4 in the figure). However, at the censorable point 2, the rate / distortion gradient does not decrease monotonously, and the code block Since truncating the code is not efficient, it is not selected as a code truncation point in the above algorithm.

  Here, the rate / distortion gradient decreasing means that the gradient becomes nearly horizontal. That is, it should be noted that the sign of the arithmetic slope of “rate / distortion slope” is excluded.

  In the following, an algorithm for determining a code cutoff point candidate from the boundary of all coding passes will be described.

  FIG. 5 is a diagram showing a flow of processing for selecting a candidate for a code interruption point. Here, a set of candidate code cut points is Ni. First, a set of boundaries of all coding passes of a code block of interest is set as an initial state of a set of candidate code break points Ni (step S501). That is, assuming that the number of coding passes of the code block Bi is ki_max, Ni = {1, 2, 3,..., Ki_max}.

  Next, the code truncation point p to be subjected to candidate determination is set to 0 (step S502). K is set to “1” as the next cut-off point of the code cut-off point to be subjected to candidate determination (step S503). It is determined whether or not k belongs to the set Ni (step S503). If it belongs to Ni, the code amount increase ΔRi (k) and distortion index value when the code cutoff point is moved from p to k. The decrease ΔDi (k) is obtained, and the rate / distortion gradient Si (k) in this section is further obtained (step S505). If it does not belong to Ni, the process proceeds to step S508 described later. If p ≠ 0, Si (k) and Si (p) are compared (step S506), and if Si (k)> Si (p), p is removed from the set Ni (step S510), and the process proceeds to step S502. Return. Otherwise, k is set to p (step S507), 1 is added to the value of k and updated (step S508), and k and ki_max are compared. The process is performed from S504, and if k> ki_max, the process is terminated, and the numbers remaining in the set Ni at this point become a candidate set of code censoring points (passes).

  When the above processing is performed, in the case of the code block as shown in FIG. 10 described above as an example, the candidate set Ni of the truncation points of the focused code block Bi that remains collected is {1, 3, 4}, With respect to these censoring candidate points, it is promised that the rate / distortion gradient monotonously decreases with k as shown in FIG.

  The values of Si (k) and Ri (k) are held for k belonging to the code termination point candidate set Ni obtained as described above, and the maximum k satisfying Si (k)> λ is selected. When the value of λ is small, the code truncation point goes down and fewer codes are discarded. Conversely, when the value of λ is large, the code truncation point rises and more codes are discarded. It can be regarded as a parameter. While changing from a large λ value to a small value, search for λ where the total code amount R = Rmax or R ≒ Rmax, and determine the code cutoff point of each code block based on that λ, rate / distortion Can be optimized.

  Hereinafter, an embodiment in which the rate / distortion optimization technique is applied to JPEG2000 will be described. However, since the specific method of encoding by JPEG2000 is described in detail in the recommendation, only a rough processing flow will be described here for a simple example.

  For simplification of explanation, the encoding target image is a 512 × 512 pixel monochrome image in which each pixel is expressed by 8 bits (256 gradations of 0 to 255), and the horizontal direction of each pixel of the encoding target image is set. The pixel position (coordinates) is x, the pixel position in the vertical direction is y, and the pixel value at the pixel position (x, y) is represented by P (x, y). Also, JPEG2000 encoding conditions include no tile division (512 × 512 pixels = 1 tile), two discrete wavelet transforms, use of a 9 × 7 irreversible filter (9-7 Irreversible Filter), and a code block size of 64 × 64, it will be described as code string formation in one layer. In addition, various condition settings such as an entropy encoding option are necessary, but are not particularly mentioned here.

  FIG. 2 is a diagram showing the flow of JPEG2000 encoding processing. In the figure, 200 is an image input unit, 201 is a discrete wavelet transform unit, 202 is a coefficient quantization unit, 203 is a code block division unit, 204 is a code block encoding unit, 205 is a code string forming unit, and 206 is a code string storage. 207 is a code block information storage unit, and 208 is a code output unit.

  First, each pixel value P (x, y) constituting the image data to be encoded is sequentially input from the image data input unit 200. The image data input unit 200 performs DC level shift of input data from 0 to 255 by subtracting the intermediate value 128 from each input pixel value P (x, y), and data P ′ (x from −128 to 127) , Y) and output to the discrete wavelet transform unit 201. The discrete wavelet transform unit 201 appropriately stores the input data P ′ (x, y) after the DC level shift in an internal buffer (not shown) and performs a two-dimensional discrete wavelet transform. The two-dimensional discrete wavelet transform is performed by applying the one-dimensional discrete wavelet transform in the horizontal and vertical directions. The discrete wavelet transform unit 201 uses a 9 × 7 tap irreversible filter as described above as the one-dimensional discrete wavelet transform. FIG. 3 is a diagram for explaining subbands of an encoding target image processed by the two-dimensional discrete wavelet transform.

  That is, a one-dimensional discrete wavelet transform is first applied to the encoding target image as shown in FIG. 3A in the vertical direction, and the low frequency subband L and the high frequency are applied as shown in FIG. Decomposes into subband H. Next, by applying a horizontal one-dimensional discrete wavelet transform to each subband, it is decomposed into four subbands LL, HL, LH, and HH as shown in FIG. Complete the first wavelet transform.

  The discrete wavelet transform unit 201 further applies the two-dimensional discrete wavelet transform (second wavelet transform) to the subband LL obtained by the above-described two-dimensional discrete wavelet transform. As a result, the encoding target image can be decomposed into seven subbands LL, HL1, LH1, HH1, HL2, LH2, and HH2.

  FIG. 6 shows seven subbands obtained by the second two-dimensional discrete wavelet transform. On the decoding side, by decoding the coefficients of the LL subband, it is possible to reproduce an image with 1/4 the size of the original image in the horizontal and vertical directions (the number of pixels in the horizontal and vertical directions is 1/4). , LH1 and HH1 can be decoded to reproduce an image having a size of 1/2 both horizontally and vertically, and by decoding up to HL2, LH2 and HH2, an image having the same size as the original image can be reproduced. . In other words, this means that the resolution of the image becomes higher in this order, so that an image generated based on the LL subband is an image generated with a resolution of 0, a resolution of 0, and an image generated with subbands of LH1, HL1, and HH1. In other words, an image having a resolution level of 1, a resolution of 1, and an image generated in subbands of LH2, HL2, and HH2 have an image level of 2.

  Now, the coefficient in each subband is represented as C (Sb, x, y). Here, Sb represents the type of subband, and is one of LL, LH1, HL1, HH1, LH2, HL2, and HH2. Further, (x, y) represents the coefficient position (coordinates) in the horizontal direction and the vertical direction when the coefficient position of the upper left corner in each subband is (0, 0).

The coefficient quantization unit 202 quantizes the coefficient C (S, x, y) of each subband generated by the discrete wavelet transform unit 201 using a quantization step delta (S) determined for each subband. To do. Here, if the quantized coefficient value is expressed as Q (S, x, y), the quantization processing performed by the coefficient quantization unit 203 can be expressed by the following equation.
Q (S, x, y) = sign {C (S, x, y)}
× floor {| C (S, x, y) | / delta (S)}
Here, sign {I} is a function representing the sign of the integer I, and returns 1 when I is positive and -1 when negative. Further, floor {R} represents the maximum integer value not exceeding the real number R.

  The code block dividing unit 203 stores the subband coefficient C (S, x, y) quantized by the coefficient quantizing unit 202 in an internal buffer (not shown) as appropriate, and is a rectangle of a predetermined size called a code block. Divide and cut out. The code block division is performed by dividing the code block into, for example, 64 × 64 blocks with the upper left corner of the subband as a reference.

  Therefore, in the case of the twice wavelet transform, each subband of LL, HL1, LH1, and HH1 has a size of 128 × 128. Therefore, 2 × 2 = 4 code blocks are included, and HL2, LH2, and HH2 are included. Each of the subbands includes 4 × 4 = 16 code blocks. Each code block is assigned an identification number i (= 0 to 63) that does not overlap in order, and a code block is specified in the form of Bi such as B0, B1, B2,. As described above, the identification numbers i are assigned in the order of resolution levels, and are numbered in the order of HL, LH, and HH subbands within the same resolution level, and in the order of raster scans within the same subband.

  FIG. 7 shows a state of code block division and number allocation in the code block division unit 203. In the figure, solid lines represent subband boundaries, dotted lines represent code block boundaries, and rectangles delimited by dotted lines or solid lines are code blocks.

  The code block encoding unit 204 calculates the absolute value of the quantized coefficient value Q (S, x, y) (hereinafter simply referred to as “coefficient value”) in the code block Bi cut out by the code block dividing unit 203. Expressed in a natural binary number, binary arithmetic coding is performed with priority given to the bit plane direction from the upper digit to the lower digit, and the coded data of the code block is stored in the code string storage unit 206.

  At this time, each bit plane is divided into three passes and encoded except for the highest bit plane. The reason why the most significant bit plane is not divided into a plurality of paths is that the bits of the coefficient values existing in the most significant bit plane are dominant with respect to the coefficient values and are highly important.

  FIG. 12 shows the above in an easy-to-understand manner. In the above, one code block has a size of 64 × 64, but in FIG. As described above, since the most significant bit plane is the most significant bit plane of the corresponding code block, all bits are treated as encoding targets. Then, the bit planes lower than the most significant bit plane are divided into three groups as shown in the figure, and the encoding target bit positions are set to exclusive positions as shown in the figure. Meaning that there will be one bit plane). If the pass that terminates the encoding described so far is the n-th pass, it means that the encoded data from the higher order to the n-th pass is used, and the encoded data of the subsequent passes are discarded.

  The division of each bit plane into paths, excluding the most significant bit plane, and the specific encoding method in each path follow the recommendations. A path amount increase ΔRi (k) and distortion index value decrease ΔDi (k) of the path of interest for each pass encoding are obtained, and the table shown in FIG. 8 is constructed and stored in an internal buffer (not shown). As the distortion index value, a mean square error, a weighted mean square error derived by weighting each subband, or the like is used. One method for deriving a decrease in distortion index value of each coefficient for each of the three types of paths is described in Appendix J of the recommendation. When the encoding of all passes is completed for the code block Bi of interest and the table as shown in FIG. 8 is completed, the above-described algorithm (FIG. 5) for selecting the candidate for the code truncation point is executed. A code truncation point candidate set Ni in which Si (k) monotonously decreases is obtained from the set of generalized path boundaries. As shown in FIG. 9, the number NP of elements of the candidate set Ni, the pass number k, the rate / distortion gradient Si (k), and the code amount Ri (k) at each abort candidate point are stored in the code block information storage unit 207.

  When the code block encoding unit 204 finishes encoding all the code blocks, the code string forming unit 205 refers to the information on the abort candidate points of each code block stored in the code block information storage unit 205 while referring to the total code Searching for λ where the quantity R = Rmax or R≈Rmax, the codes of the parts where Si (k)> λ are collected to form and output the final code string.

  The flow of λ determination processing in the code string forming unit 205 is shown in FIG. 13 and will be described below. In the following description, it should be noted that S is introduced as a variable representing a threshold value, and the value of the variable S at the end of processing is λ.

  First, the minimum value Smin and the maximum value Smax of Si (k) are obtained with reference to the information on the abort candidate points of all code blocks stored in the code block information storage unit 207 (step S1401). Next, Smax obtained in step S1401 is set to the variable S representing the threshold (step S1402). Subsequently, the predetermined threshold change width ΔS is subtracted from the variable S, and the value of the variable S is slightly lowered (step S1403). The variable i representing the code block number is set to 0, and the accumulated code amount R is initialized to 0 (step S1404). Again, with reference to the abort candidate point information of the code block Bi stored in the code block information storage unit 205, the maximum k satisfying Si (k)> S is obtained and set as the abort point ni of the code block Bi (step S1405). ). Since the value of Si (k) is monotonously decreased in the order of the candidate points for termination of the code block Bi, ni can be obtained by comparing Si (k) in the order of the candidate points. The code amount Ri (ni) of the code block Bi at the censoring point ni obtained in step S1405 is added to the accumulated code amount R (step S1406). Next, the variable i is updated by adding 1 (step S1407). Then, the variable i is compared with 64. If i = 64, the process proceeds to step S1409; otherwise, the process proceeds to step S1405, and the code amount is added for the next code block (step S1408).

  When the value of the variable i reaches 64, that is, when calculation of the accumulated code amount from all code blocks for the threshold S is completed, the accumulated code amount R is compared with the target code amount Rmax, and if R <Rmax, step If not, the process proceeds to step S1411 (step S1409). When the process is shifted to step S1411, the current threshold value S exceeds the target code amount, so ΔS is added to return to the previous threshold value S and the process is terminated (step S1411). In step S1410, the threshold value S is compared with the minimum value Smin obtained in step S1401, and if S> Smin, the process returns to step S1403, the value of S is slightly reduced, and the processing up to step S1409 is performed again. If S ≦ Smin, the process ends. S at the end of processing is selected as the threshold λ.

  For the threshold value λ obtained by the above-described processing, the code of the portion where Si (k)> λ is read from each code block from the code string storage unit 206, and information (main header, tile header) is read according to the JPEG2000 code string format. , Packet header, etc.) are added to form a JPEG2000 code string and output to the code output unit 208.

The code output unit 208 outputs the JPEG2000 encoded data generated by the code string forming unit 205 to the outside of the apparatus. The code output unit 208 is, for example, a storage medium such as a hard disk, a magneto-optical disk, or a memory, or an interface to a network.
Annex J of the standard recommendation (ISO / IEC 15444-1) “Generalized Lagrange Multiplier Method for Solving Problems of Optimum Allocation of Resources”, Operation Research, vol.11, pp.399-417, 1963

  However, in order to find the maximum λ where R≈Rmax by the above method, it is necessary to repeat the process of obtaining the total code amount R at various λ and comparing it with the target code amount Rmax. Many processes due to the calculation cost required to obtain the total code amount R, such as many accesses to the memory storing the information Si (k), Ri (k), and comparison between Si (k) and λ There are problems such as requiring time.

  Further, as the threshold change width ΔS is reduced, R can be made closer to Rmax, but the calculation amount increases in inverse proportion to the change width. Moreover, just because the amount of calculation is increased, there is a problem that it is not guaranteed that R closest to Rmax is obtained.

  The present invention has been made in view of such circumstances, and provides a technique for determining the optimum rate / distortion gradient as a target code amount and generating compression-encoded image data with simpler processing. It is something to try.

In order to solve this problem, for example, an image data encoding device of the present invention has the following configuration. That is,
The image data is frequency converted, the converted data is divided into a plurality of blocks of a predetermined size, and the bit information of each coefficient value in the divided block is generated in the order of the pass from the higher order to the lower order, and each block is generated. The rate that is expressed by the target code amount Rmax, the increase amount of the code data amount at each pass position, and the distortion of the image, from which the encoded data up to the lower pass position is used as the compression encoded data. An image data encoding device that determines based on a distortion gradient λ,
First storage means for storing code data at each pass stage for each block of frequency-converted data;
Second storage means for storing information on increasing code amount and distortion amount in each pass stage in each block, and distortion amount with respect to the increasing code amount;
Setting means for setting temporary rate / distortion gradient information having a significant bit length n and the most significant bit i of “1” being “1”;
Calculating means for calculating a total code amount R of code data that becomes rate / distortion gradient information equal to or higher than the provisional rate / distortion gradient information set by the setting means, by referring to the second storage means;
Comparing means for comparing the total code amount R calculated by the calculating means with the target code amount Rmax;
According to the comparison result of the comparison means, when R> Rmax, i is decreased by “1”, bit i of the provisional rate / distortion gradient information is updated to “1”, and R <Rmax Updating means for correcting the bit i to “0”, subtracting the i by “1”, and updating the bit i of the provisional rate / distortion gradient information to “1”, if any;
R = Rmax in the comparison means, or until all the bit information of the provisional rate / distortion gradient information is determined, or a code amount obtained by collecting codes up to the provisional rate / distortion gradient information and the Rmax A bit search unit that repeats the processing of the calculation unit and the determination unit in order to perform a search process of the bit information of the temporary rate / distortion gradient information until the difference is within a predetermined value;
The final provisional rate / distortion gradient information obtained by the bit search means is set as the target rate / distortion gradient information λ, and the corresponding pass stage stored in the first storage means based on the rate / distortion gradient λ. The encoded data is read out and output as compressed encoded data.

  According to the present invention, when the target rate / distortion gradient information is represented by n bits, the information of all bits can be determined by calculating and comparing the code amount at most n times, and the rate / distortion gradient is made efficient. Thus, it is possible to obtain the encoded data of the target code amount efficiently.

  DESCRIPTION OF EMBODIMENTS Hereinafter, embodiments according to the present invention will be described in detail with reference to the accompanying drawings.

  FIG. 1 is a block diagram of an image encoding processing apparatus according to the embodiment.

  The difference from FIG. 2 is that a bit search processing unit 209 for accessing the code block information storage unit 207 is provided, and this bit search processing unit 209 determines each bit of the target rate / distortion gradient information in order. The code string storage unit 205 reads the code string at the pass position of each corresponding code block from the code string storage unit 205 based on the processing result (rate / distortion gradient information) obtained by the bit search processing unit 209, and outputs the output code. Data will be generated. Other than that, it is assumed that the contents are the same as those described above, and the characteristic part of the present embodiment will be described below.

  Although details will be described later, the code block encoding unit 204 according to the embodiment encodes one image data, generates a code data string at each pass stage of each code block, and rate / distortion gradient. And a means (register or the like) for storing and holding the larger one, comparing the obtained rate / distortion gradient with the previous rate distortion gradient. This is because, when the generation of the code data string for one entire image is completed, the maximum rate / distortion gradient information for the entire image is used in the initial processing.

  The processing of the entire apparatus shown in FIG. 1 is processed according to the procedure shown in FIG.

  In the figure, first, image data is input in step S100. The image data is input by, for example, an image scanner. However, the image data may be read from a storage medium storing the image data, or the image data may be input via a network interface.

  Next, a discrete wavelet transform process is performed on the image data input in step S101, and a quantization process is performed on each coefficient of the transform result in step S102. Then, the quantization result is divided into code blocks (step S103), and each code block encoding process is performed. At this time, each code block is divided into bit planes, and bit planes other than the most significant bit plane are divided into three paths, and code data sequences are generated in order from the most significant path (step S104). Next, in step S105, the encoded data for each pass is stored in a predetermined storage means, and in step S106, the code block information ( As shown in FIG. 8).

  Next, a bit search process, which is a characteristic part in the present embodiment, is performed to obtain a target rate / distortion gradient λ (step S107). As a result, the pass stop position for each code block is obtained. In step S109, the encoded data string up to the corresponding pass position in each previously stored code block is extracted, and the final code of the entire image is extracted. An encoded data string is generated, and the encoded code string generated in step S110 is output.

  During the above processing, the characteristic processing in the present embodiment is in step S107, that is, the bit search processing unit 209 in FIG. Therefore, in the following, this point will be mainly described.

  First, in order to simplify the description, the encoded image data to be encoded by the image encoding apparatus of the present embodiment is monochrome image data of 8 bits of each pixel of 512 × 512, as in the above-described conventional example. . Further, there is no tile division (1 tile = 512 × 512), the code block size is 64 × 64, and wavelet transform is performed twice using a 9 × 7 tap filter. That is, it is the same as the conventional example described above.

  In addition, the code string storage unit 206 stores code data sequences of each code block at each pass stage, and the code block information storage unit 207 already creates the table shown in FIG. 8 for each code block. It shall be. The rate / distortion gradient information is represented by 8 bits.

  The bit search processing unit 209 in the embodiment lowers the target rate / distortion gradient information λth from the upper bits to obtain the target code amount or a code amount approximate thereto based on the information stored in the code block information storage unit 207. One bit is determined in the direction toward the bit. The principle will be described below.

  First, in the initial stage, the bit search processing unit 209 inputs the maximum rate / distortion gradient information λmax stored and held in the code block encoding unit 204. The maximum rate / distortion gradient information λmax can also be obtained by searching the code block information storage unit 207. However, the above processing is convenient because the processing required for the search is omitted.

  It is promised that the target rate / distortion gradient information λth is 0 or more and not more than the maximum rate / distortion gradient information λmax. In other words, when m bits going from the most significant bit of the maximum rate / distortion gradient information λmax to the lower order are “0”, at least m going from the most significant bit of the target rate / distortion gradient information λth to the lower order. The bit is also promised to be “0”. Therefore, obtaining the maximum rate / distortion gradient information λmax is equivalent to obtaining the number of bits that can be set to 0 from the most significant bit of the target rate / distortion gradient information λth.

  Therefore, the target rate / distortion gradient information λth should be obtained in what state the lower n (= 8−m) bits (note that the rate / distortion gradient information is 8 bits in the embodiment). become.

  Here, it is assumed that the maximum rate / distortion gradient information λmax in all code blocks input from the code block encoding unit 204 is “11” (“00001011” in binary 8 bits). In this case, at least the upper 4 bits of the target rate / distortion gradient information λth are 0, and the lower 4 bits are undefined. That is, the target rate / distortion gradient information λth is “0000xxxx” (where x is 0 or 1) in binary notation.

  That is, it can be said that the target rate / distortion gradient information λth exists between “000011111” (hereinafter referred to as λvmax) and “00000000” (hereinafter referred to as λvmin) in binary numbers. In the above case, λvmax> λmax. Strictly speaking, λvmax should be λmax. However, in this embodiment, λh is determined for each bit, and thus λvmax has λmax. A significant bit group is determined by the most significant bit position.

  Now, when λvmax and λvmin are determined, a provisional rate distortion gradient threshold λm = 00001000 is obtained as an approximately intermediate value between them. In other words, λm assumes that the most significant bit in the undetermined bit group of λth (the lower 4 bits in the above) is “1”, and the lower bits are set to 0.

  Reference numerals 1501 to 1504 in FIG. 15 indicate the code amount and distortion. Note that the horizontal axis is the code amount, but can also be viewed as a pass stage.

A code 1501 in FIG. 15 indicates the code amount and distortion D in each of λvmax, λm, and λvmin. The accumulated code amount R that is equal to or greater than the threshold λm (the code data amount when the rate / distortion gradient is λm) When the information storage unit 207 is checked and cumulatively added) and the target code amount Rmax are compared, the relationship that can be taken is
R> Rmax or
R <Rmax or
R = Rmax
One of them.

  Here, if there is a relationship of “R> Rmax”, the λth to be obtained exists on the left side of the current λm as indicated by reference numeral 1502 in the figure. Therefore, in this case, λvmin is updated with λm at that time, and an intermediate position between λvmax and the updated λvmin, that is, the value “00001100” is set as a new λm. In other words, it is assumed that the most significant bit among the 4 bits in which λth is indefinite is determined as “1” and the lower-order bit is “1”.

  On the other hand, when the relationship of “R <Rmax” is established, as shown by the reference numeral 1504 in the figure, it indicates that the λth to be found exists on the right side of λm at the present time. Therefore, in this case, λvmax is updated with the current λm, and a value “00000100”, which is an intermediate position between λmin and the updated λvmax, is set as a new λm. In other words, it is assumed that the most significant bit among the 4 bits in which λth is indefinite is determined as “0”, and the lower bit thereof is “1”.

  If there is a relationship “R = Rmax”, it is determined that λm at that time is the λth to be obtained, and this processing is terminated.

  Here, in the case of either “R> Rmax” or “R <Rmax”, the above processing is repeated to determine the 4 bits of λth that are initially undetermined in order from the upper bit position. To go.

  FIG. 17 shows a case where the maximum rate / distortion gradient λmax of all code blocks is “11” in decimal notation (“00001011” in binary notation). Since λmax is checked from the upper part and starts at bit 3 and becomes “1”, the temporary λm sets the upper 4 bits to “0”, and the undetermined bits are the lower 4 bits. Accordingly, bit 3 at the maximum position in the undetermined bit group is set as a start bit position, and the bit positions are determined in the order of bit 2, bit 1, and bit 0 hereinafter.

  In short, when the undetermined bit of the initial λth is n bits, the final λth satisfying R ≦ Rmax can be obtained by performing the process of obtaining the total code amount R at most n times and the comparison process. Means. In the embodiment, since the case where the rate / distortion gradient information is expressed by 8 bits is described, it can be said that it is necessary to perform 8 operations at the maximum.

  The specific configuration of the bit search processing unit 209 in the embodiment is shown in FIG. 16, and the configuration will be described below according to the contents of the operation processing.

  The maximum significant bit position detection unit 100 includes a maximum rate / distortion in all code blocks stored and held in the code block encoding unit 204 after the code block encoding unit 204 completes encoding of one image data. The gradient information λmax is input, the bit going from the most significant bit MSB to the lower order is examined, and the information S indicating the bit position that is first “1” is output to the threshold value generation unit 101 and the determination unit 106. For example, when the value of the maximum rate / distortion gradient information λmax is “11” in decimal as described above, it becomes “00001011” in binary, and therefore “0” from bit 7 to bit 4. Since it becomes “1” only after becoming bit 3, “3” is output.

  The threshold generation unit 101 receives the information S from the maximum significant bit position detection unit 100 at the first time, generates a temporary threshold λm (initial λm) with the significant bit position set to “1”, and uses this as the code amount. The data is output to the R determination unit 103. For example, if the maximum significant bit position is “3”, λm = “00001000”.

  The code amount R determining unit 103 searches the code block information storage unit 207 for the total code data amount R that becomes rate / gradient information greater than the input rate / distortion gradient information λm, and cumulatively adds the corresponding code amount. And the result is output to the comparison unit 104.

  The comparison unit 104 compares the target code amount Rmax set in advance with the code amount R from the code amount R determination unit 103, and any of the comparison results, that is, R <Rmax, R> Rmax, R = Rmax. Is output to the determination unit 106.

  The code amount R determination unit 103 performs any one of the following three determination processes. However, the value S output from the maximum significant bit position detection unit 100 is used as an initial value, and a counter that decreases by 1 when any of the three processes is performed is provided. When the value held in the counter is CT, CT = S at the first time.

<Determination process 1>
When the comparison result information from the comparison unit 104 is information indicating “R = Rmax”, the current λm is output to the code string forming unit 205 as the target rate / distortion gradient information λth. At this time, even if the value CT of the counter is other than 0, this process ends.

<Determination process 2>
When the comparison result information from the comparison unit 104 is information indicating “R> Rmax”, the bit information of the bit CT indicated by the value of the counter CT is determined to be correct by “1” (CT = S = 3 In this case, it is assumed that bit 3 of λm is “1” and correct). Then, CT is decreased by 1. Further, in order to temporarily set the bit position (bit CT) indicated by the reduced CT to “1”, the threshold generation unit 101 is requested to update λm.

  For example, when S = CT = 3, λm was “00001000”, but by setting CT = CT−1 = 2, a new λm = “00001001” in which bit 2 of the current λm is “1”. Request to generate.

  When CT after the update becomes <0, the state of all bits determined to be significant bits is confirmed, so that λm obtained for the last code amount R is encoded as the target rate / distortion gradient information λth. The data is output to the column forming unit 205, and this process ends.

<Determination process 3>
When the comparison result information from the comparison unit 104 indicates “R <Rmax”, the bit information of the bit CT indicated by the value of the counter CT is erroneous and is determined to be “0” (CT = S = 3 In this case, bit 3 of λm is set to “0”). Then, CT is decreased by 1. Furthermore, in order to temporarily set the bit position indicated by the reduced CT to “1”, the threshold generation unit 101 is requested to update λm.

  For example, when S = CT = 3, λm was “00001000”, but since the value of bit 3 was found to be correct, it is possible to set CT = CT−1 = 2. , Request to generate a new λm = “00000100” with bit 2 of the current λm set to “1”. When CT after updating <0, the state of all bits determined to be significant bits is confirmed, but the least significant bit of λm for which the code amount R is obtained is set to 0. The result is output to the code string forming unit 205 as the target rate / distortion gradient information λth of λm, and this processing is finished.

  As described above, the target rate / distortion gradient information λth is obtained. The code string forming unit 205 sets the given target rate / distortion gradient information λth as the maximum rate / distortion gradient information, and the code block information storage unit 207. Is determined for each code block, the code data string up to the determined path position is read from the code string storage unit 206, the data is configured in a predetermined format, and is output to the code output unit 208. become.

  As described above, according to the present embodiment, when the rate / distortion gradient information is expressed by n bits, a process for obtaining the code amount R at most n times and a process for comparing the target code amount Rmax are performed. This makes it possible to obtain target rate / distortion gradient information. Therefore, compared with the processing in which Δλ is sequentially added to the initial value λ as in the past, the calculation processing amount and the number of accesses to the code block information storage unit are significantly reduced, and the processing speed is greatly improved. Is possible.

  Further, the significant bit number indicating the maximum value of the rate / distortion gradient information of all code blocks obtained when performing the encoding process is obtained, and the target rate / Since each bit state of the distortion gradient information λth is determined, it is possible to simplify the processing because the determination processing of the bit states other than the significant bits is unnecessary. Further, the maximum value of the rate / distortion gradient information of all code blocks can be reduced by one processing for searching the code block information storage unit 207 by causing the code block encoding unit 204 to hold the maximum value, thereby further increasing the processing amount. It becomes possible to reduce.

  FIG. 19 shows a configuration when the processing for obtaining the target rate / distortion gradient information λth in the embodiment is realized by a general-purpose information processing apparatus such as a personal computer.

  In the figure, 1 is a CPU that controls the entire apparatus, 2 is a ROM that stores a boot program and BIOS, and 3 is a RAM that is used as a work area of the CPU 1. Reference numeral 4 denotes an external storage device such as a hard disk device, in which an OS and an application program (JPEG2000 application) relating to image compression processing are stored as shown in FIG. In addition, a communication program or the like for finally distributing a code amount approximate to Rmax with a target code amount Rmax or less than Rmax on the network is stored. Reference numeral 5 denotes an image input unit, which is an image scanner, for example. When image data is stored in a storage medium, means for accessing the storage medium, and when the image data exists in a server on the network Can also be viewed as a network interface. Reference numeral 6 denotes a keyboard (KB) and a pointing device (PD). Reference numeral 7 denotes a display control unit which is equipped with a video RAM, draws data in the video RAM, reads data from the video RAM, and outputs it as a video signal. Reference numeral 8 denotes a display device that displays based on a video signal from the display control unit 7.

  In the above configuration, when the power is turned on, the CPU 1 loads the OS from the external storage device to the RAM 3 in accordance with the boot program stored in the ROM 2, and then loads the image compression program, whereby the image input from the image input unit 5 is loaded. Compressed encoded data having a code amount that approximates the target code amount Rmax set in advance or not exceeding the target code amount Rmax is generated and stored in the external storage device. Note that the output destination of the finally generated compression-encoded image data may be output via a network instead of the external storage device, and the output destination is not limited.

  Here, as described above, the compression encoding processing program for the image data is performed according to the processing procedure of FIG. 14, and the code block storage unit and the code string storage unit generated up to step S106 are used. The data is temporarily stored in the external storage device 4 as shown in FIG. The RAM may be used as long as it can be secured on the RAM.

  And the bit search process in step S107 in this embodiment should just perform the process shown in FIG. Hereinafter, an example realized by software will be described with reference to FIG.

  First, in step S1801, with reference to the code block information storage unit stored in the external storage device 4, the maximum rate / distortion gradient information λmax of all code blocks of the input image is obtained. In the code block information storage unit, data in the format as shown in FIG. 8 is generated for one code block. Therefore, the ratio between the increasing code amount and the distortion index value is calculated, and all the code blocks are included. What is necessary is just to calculate what becomes the maximum value of. Note that when generating the encoded data string of the code block, the maximum value is stored and held, and if it is used, λmax can be easily obtained.

  In step S1802, the maximum bit position in the significant bit group in λmax is obtained and stored in the variable n. In step S1803, a variable λm (however, n bits or more) for obtaining the target rate / distortion gradient is cleared to zero, and in step S1804, the bit n is set to “1”.

  In step S1805, the code block information storage unit is referred to, and the code amount of each code block having a rate / distortion gradient of λm or more is cumulatively added to obtain a total code amount R of λm or more. In step S1805, the calculated total code amount R is compared with the target code amount Rmax (which may be specified by the user using a keyboard or the like, or determined based on the size of the document image). .

  If it is determined that R <Rmax, the bit n of λm is not “1” but “0”, so the bit n of λm is corrected to “0” in step S1807.

  If it is determined in step S1806 that R> Rmax, or if the processing in step S1807 is performed, the process proceeds to step S1808, whether or not the variable n is “0”, that is, whether or not the LSB determination is completed. Determine whether. If NO (n> 0), the variable n is reduced by “1” in step S1809, the bit n of λm is set to “1” in step S1810, the process returns to step S1805, and the above processing is repeated.

  If it is determined in step S1806 that R = Rmax or n = 0 in step S1808, λm at that time is determined as the target rate / distortion gradient λth (step S1811), and this process is performed. Return to the calling process. Thereafter, processing steps S109 and S110 shown in FIG. 14 may be performed.

  Although the embodiment according to the present invention has been described above, the present invention is suitable for implementation in JPEG 2000, but in other encoding schemes in which encoded data is divided into small sections to obtain a code amount and a distortion index value. Can also be applied.

  In the above-described embodiments, 512 × 512 pixel 8-bit monochrome image data has been described as an encoding target image. However, images of other sizes and bit depths, and each pixel is represented by a plurality of color components. The present invention may be applied to other image data such as a color image. Furthermore, the present invention may be applied to each frame, field, etc. of a moving image.

  Furthermore, the processing in the embodiment can be applied to a computer program executed by a general-purpose information processing apparatus such as a personal computer as described above. In addition, since the computer program is normally executable by setting a computer-readable storage medium such as a CDROM in the computer and copying or installing it in the system, the present invention also includes such a computer-readable storage medium. It is clear to include.

It is a block block diagram of the image coding apparatus in embodiment. It is a block block diagram which shows the structure of the prior art example of an image coding apparatus. It is a figure for demonstrating the subband of the encoding object image processed by two-dimensional discrete wavelet transform. It is a flowchart which shows the flow of the process which determines the code | symbol truncation point ni of code block Bi. It is a flowchart which shows the flow of the process of monotonic reduction by integration of the path | pass of code block Bi. It is a figure for demonstrating seven subbands obtained by two times of two-dimensional discrete wavelet transforms. It is a figure which shows the mode of the code block division | segmentation in the code block division part 203. FIG. 6 is a diagram illustrating an example of information of a code block Bi constructed inside a code block encoding unit 204. FIG. FIG. 6 is a diagram illustrating information on code abort candidate points stored in a code block information storage unit 207. It is a figure which shows an example of the relationship between the rate of each path | pass of a code block, and distortion. It is a figure which shows a mode that a path | pass is integrated by monotonic reduction process. It is a figure which shows the relationship between each bit plane of a code block, and the path | pass of an encoding. 10 is a flowchart showing a flow of λ determination processing in a conventional code string forming unit 205. It is a flowchart which shows the process sequence of the image data compression process in embodiment. It is a figure for demonstrating the principle of the bit search process part in FIG. It is a figure which shows an example of a structure of the bit search process part in FIG. It is a figure for demonstrating the processing content of the bit search process part in FIG. It is a flowchart which shows the process sequence for implement | achieving a bit search process with software. It is a figure which shows the structural example of the apparatus at the time of implement | achieving the image compression process in embodiment by a computer program. It is a figure which shows the program and work area which are stored and ensured in an external storage device.

Claims (7)

The image data is frequency converted, the converted data is divided into a plurality of blocks of a predetermined size, and the bit information of each coefficient value in the divided block is generated in the order of the pass from the higher order to the lower order, and each block is generated. The rate that is expressed by the target code amount Rmax, the increase amount of the code data amount at each pass position, and the distortion of the image, from which the encoded data up to the lower pass position is used as the compression encoded data. An image data encoding device that determines based on a distortion gradient λ,
First storage means for storing code data at each pass stage for each block of frequency-converted data;
Second storage means for storing information on increasing code amount and distortion amount in each pass stage in each block, and distortion amount with respect to the increasing code amount;
Setting means for setting temporary rate / distortion gradient information having a significant bit length n and the most significant bit i of “1” being “1”;
Calculating means for calculating a total code amount R of code data that becomes rate / distortion gradient information equal to or higher than the provisional rate / distortion gradient information set by the setting means, by referring to the second storage means;
Comparing means for comparing the total code amount R calculated by the calculating means with the target code amount Rmax;
According to the comparison result of the comparison means, when R> Rmax, i is decreased by “1”, bit i of the provisional rate / distortion gradient information is updated to “1”, and R <Rmax Update means for correcting the bit i to “0”, subtracting the i by “1”, and updating the bit i of the provisional rate / distortion gradient information to “1”, if any;
A code amount and a target code amount Rmax obtained by collecting the codes up to the provisional rate / distortion gradient information until R = Rmax in the comparison unit, or all bit information of the provisional rate / distortion gradient information is determined. A bit search unit that repeats the processing of the calculation unit and the determination unit in order to perform a search process of the bit information of the provisional rate / distortion gradient information until the difference is within a preset error,
The final provisional rate / distortion gradient information obtained by the bit search means is set as the target rate / distortion gradient information λ, and the corresponding pass stage stored in the first storage means based on the rate / distortion gradient λ. An image data encoding apparatus, wherein the encoded data is read out and output as compressed encoded data.   The image data encoding apparatus according to claim 1, wherein the encoding is encoding according to JPEG2000. Furthermore, storage unit for storing and holding the maximum rate / distortion gradient information among the rate / distortion gradient information at each pass stage of each code block when generating encoded data,
Significant bit number calculating means for obtaining the maximum number of significant bits of rate / distortion gradient information stored and held by the storage holding means,
2. The image data encoding apparatus according to claim 1, wherein the number of bits n of the provisional rate / distortion gradient information is the number of bits calculated by the significant bit number calculation means. Further, search means for obtaining the maximum rate / distortion gradient information by searching the second storage means,
Significant bit number calculating means for obtaining a significant bit number of the maximum rate / distortion gradient information searched by the search means,
2. The image data encoding apparatus according to claim 1, wherein the number of bits n of the provisional rate / distortion gradient information is the number of bits calculated by the significant bit number calculation means. The image data is frequency converted, the converted data is divided into a plurality of blocks of a predetermined size, and the bit information of each coefficient value in the divided block is generated in the order of the pass from the higher order to the lower order, and each block is generated. The rate that is expressed by the target code amount Rmax, the increase amount of the code data amount at each pass position, and the distortion of the image, from which the encoded data up to the lower pass position is used as the compression encoded data. An image data encoding method for determining based on a distortion gradient λ,
A first storage step of storing code data at each pass stage for each block of frequency converted data;
A second storage step for storing information on an increasing code amount and distortion amount at each pass stage in each block and a distortion amount with respect to the increasing code amount;
A setting step of setting provisional rate / distortion gradient information having a significant bit length n and the highest bit i of i = n−1 being “1”;
A calculation step of calculating a total code amount R of code data that becomes rate / distortion gradient information equal to or higher than the provisional rate / distortion gradient information set in the setting step by referring to the second storage unit;
A comparison step of comparing the total code amount R calculated in the calculation step with the target code amount Rmax;
When the comparison result of the comparison step is R> Rmax, i is reduced by “1”, bit i of the provisional rate / distortion gradient information is updated to “1”, and when R <Rmax. Updating the bit i to “0”, subtracting the i by “1”, and updating the bit i of the temporary rate / distortion gradient information to “1”;
In the comparison step, R = Rmax, or until all bit information of the provisional rate / distortion gradient information is determined, or a code amount obtained by collecting codes up to the provisional rate / distortion gradient information and the Rmax A bit search step that repeats the processing of the calculation step and the determination step in order to perform the search processing of the bit information of the temporary rate / distortion gradient information until the difference is within a predetermined value,
The final provisional rate / distortion gradient information obtained by the bit search step is set as the target rate / distortion gradient information λ, and the corresponding pass step stored in the first storage step based on the rate / distortion gradient λ. An image data encoding method characterized in that the encoded data is read out and output as compressed encoded data. The image data is frequency converted, the converted data is divided into a plurality of blocks of a predetermined size, and the bit information of each coefficient value in the divided block is generated in the order of the pass from the higher order to the lower order, and each block is generated. The rate that is expressed by the target code amount Rmax, the increase amount of the code data amount at each pass position, and the distortion of the image, from which the encoded data up to the lower pass position is used as the compression encoded data. A computer program that functions as an image data encoding device that is determined based on a distortion gradient λ,
First storage means for storing code data at each pass stage for each block of frequency-converted data;
Second storage means for storing information on increasing code amount and distortion amount in each pass stage in each block, and distortion amount with respect to the increasing code amount;
Setting means for setting temporary rate / distortion gradient information having a significant bit length n and the most significant bit i of “1” being “1”;
Calculating means for calculating a total code amount R of code data that becomes rate / distortion gradient information equal to or higher than the provisional rate / distortion gradient information set by the setting means, by referring to the second storage means;
Comparing means for comparing the total code amount R calculated by the calculating means with the target code amount Rmax;
According to the comparison result of the comparison means, when R> Rmax, i is decreased by “1”, bit i of the provisional rate / distortion gradient information is updated to “1”, and R <Rmax Update means for correcting the bit i to “0”, subtracting the i by “1”, and updating the bit i of the provisional rate / distortion gradient information to “1”, if any;
R = Rmax in the comparison means, or until all the bit information of the provisional rate / distortion gradient information is determined, or the difference between Rmax and the amount of code obtained by collecting the codes up to the provisional rate / distortion gradient information A bit search unit that repeats the processing of the calculation unit and the determination unit in order to perform a search process of bit information of the provisional rate / distortion gradient information until is within a predetermined value,
The final provisional rate / distortion gradient information obtained by the bit search means is set as the target rate / distortion gradient information λ, and the corresponding pass stage stored in the first storage means based on the rate / distortion gradient λ. A computer program that functions to read out the encoded data in and output as compressed encoded data.   A computer-readable storage medium storing the computer program according to claim 6.

Images