JPH03216075A - Transform encoding method - Google Patents

Abstract

(57) [Summary] This bulletin contains application data before electronic filing, so abstract data is not recorded.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to band compression of image data using a linear transformation coding method.

[Prior art] Figure 3 shows, for example, W, H, CHEN, W, K,
PRATT, 'Scene Adaptive Co
der', (IEEE Transactions
on communications, vol. C
OM-32. No. 3. 1984) is a block diagram showing a conventional transform encoding method, in which (1) indicates a blocking unit that blocks an input signal;
(2) is a linear conversion unit that linearly converts a blocked signal; (3) is a scan conversion unit that rearranges a signal sequence within a block; (4) is a quantization unit; (5) is a valid/invalid identification unit;
(6) is an encoding unit, (7) is a transmission buffer, and (8) is an encoding control unit.

Next, the operation will be explained. For one frame of input image signal (1 0 1) that has been digitized, the blocking unit (1) collects N pixels in the horizontal and vertical directions (N is a natural number, for example, N = 4.8.16). Divide into two-dimensional pixel blocks. Blocked image signal (1
0 2), the linear transformation unit (2) performs two-dimensional linear transformation (for example, orthogonal transformation such as discrete cosine transformation) to generate a transform coefficient block (1 0 3) in the spatial frequency domain. Here, for example, an 8×8 pixel block f(x, y)
The two-dimensional discrete cosine transformation for (x. y=0. 1, . . . , 7) is given by the following equation.

Here, u, v=0.1, 7, x, y are coordinates in the pixel domain, and u, v are coordinates in the transformation domain.

Transform coefficient block F(u, v) (u, v=0.
The properties of 1, ..., 7) will be explained based on Figure 4. The value of F(u, v) is the blocked image signal (1
0 to 2) respectively. The frequency in the horizontal direction increases as the value of U increases, and the frequency in the vertical direction increases as the value of V increases. That is, F (0.
The value of 0) corresponds to the strength of the DC component of the blocked image signal (102), and the value of F (7.7) corresponds to the strength of the AC component that has high frequencies in both the horizontal and vertical directions. Become. Therefore, for a flat image block such as a background with few changes in pixel values, non-zero significant coefficients appear only in low frequency components, and almost zero coefficients appear in high frequency components. On the other hand, for image blocks such as edge portions where pixels change rapidly, non-zero significant coefficients appear not only in low frequency components but also in high frequency components.

Next, in the scan conversion unit (3), the conversion coefficient block (1
0 3), the transform coefficients are rearranged in the order shown by the arrows in Fig. 4, for example, to form a one-dimensional transform coefficient sequence F (n
) (104) is output. In the case of the previous 8x8 pixel block, 64 coefficients continue for one block (n =
1 to 64) are output. The rearrangement is done by scanning in a zigzag pattern from the conversion coefficients of low frequency components where non-zero significant coefficients tend to appear to the conversion coefficients of high frequency components where significant coefficients are unlikely to appear, so that the significant coefficients are placed in the first half as much as possible and the zero coefficients are kept in the second half for a long time. I do it to make it work.

Next, the quantization unit (4) quantizes the transform coefficient sequence (1 0 4) with the given quantization step size (1 1 0) and outputs a quantized coefficient sequence Q(n) (105). The validity/invalidity identification unit (5) determines whether the quantized coefficient sequence (1 0 5) is all zero. Valid/invalid information (1 0 6) is output to the encoding unit (6) as an invalid block if all coefficients are zero, and as a valid block if there is even one non-zero significant coefficient. The encoding unit (6) assigns a code to the quantized coefficient sequence (1 0 5) only when the block is determined to be a valid block based on the validity/invalidity information (106), and sends it to the transmission buffer as encoded data (1 0 7). Output to (7). On the other hand, valid and invalid information (1 0
6), when the block is determined to be invalid, the code representing the invalid block is output as encoded data (1 0 7) to the transmission buffer (7).

Here, two-dimensional variable length coding will be described as an example of a code assignment method. This is a quantized coefficient sequence (1 0
For 5), combine the number of consecutive zero coefficients (hereinafter referred to as zero run) and the quantization level of the following non-zero coefficients, and assign one Huffman code to the combined event (zero run, quantization level). It is done by

For example, the quantized coefficient sequence (1 0 5) shown in FIG.
For , the event (zero run, quantization level) becomes:

(0.20), (2.15), (4.5), (3.2)
．． (7.1). EOB Here, EOB indicates that there are no non-zero significant coefficients thereafter, and zero coefficients continue until the end of the block. Therefore, in the case of this quantized coefficient sequence, a predetermined Huffman code is assigned to each of six events including EOB.

Next, the transmission buffer (7) smoothes the fluctuating amount of generated information and sends it to the transmission path (108) at a constant rate.

The encoding control unit (8) uses the buffer remaining amount, which is the remaining amount of data in the transmission buffer (7)! (109), the quantization step size (1 1 0) is adaptively feedback-controlled and output to the quantization unit (4). In other words, the buffer remaining f
When i(109) is large, the quantization step size (1 1 0
) to coarsely quantize the transform coefficient sequence (1 0 4).

On the other hand, when the remaining buffer ffi (1 0 9) is small, the quantization step size (1 1 0) is reduced to finely quantize the transform coefficient sequence (1 0 4) in order to increase the amount of information that will be generated. do.

[Problem to be solved by the invention] Since the conventional transform encoding method is configured as described above, all transform coefficients must be quantized in order to perform valid/invalid identification and two-dimensional variable encoding processing. had to be processed.

This invention was made to solve the above-mentioned problems, and it determines the transmission range of transform coefficients according to the quantized coefficient sequence in the transform coefficient block, quantizes the necessary coefficients, and The purpose of this invention is to obtain a transform coefficient encoding method that simultaneously generates events for performing valid/invalid identification and two-dimensional variable length encoding, and reduces the amount of calculations and processing time required for processing.

[Means for Solving the Problems] The transform coefficient encoding method according to the present invention is a transform code that linearly transforms an input signal sequence and sequentially quantizes and encodes transform coefficients from a low frequency band to a high frequency band in a transform domain. In the quantized conversion method, there is a means for counting the number of consecutive zero coefficients among the values of the quantized transform coefficients, and a means for counting the number of consecutive zero coefficients among the quantized coefficients and the number of non-zero coefficients counted by the counting means until the non-zero coefficients appear. means for storing a set of the number of consecutive zero coefficients and the number of consecutive zero coefficients, and a threshold value for the number of consecutive zero coefficients to be encoded and transmitted in order to bring the generated amount of encoded information close to a predetermined amount of transmitted information from the remaining amount of data in the transmission buffer. and means for discontinuing quantization of linearly transformed coefficients after the counted value of continuous zero coefficients exceeds a threshold value and assigning a code to the contents stored in the storage means.

1 Effect] The transform coding method according to the present invention linearly transforms an input signal block, quantizes it in a given order, counts the number of consecutive zero coefficients, and calculates non-zero coefficient values and their non-zero coefficients. The pair with the number of consecutive zero coefficients counted until the coefficient appears is temporarily stored as an event, and when the number of consecutive zero coefficients exceeds a threshold set from the remaining data amount of the transmission buffer, the conversion coefficient is Quantization is discontinued and a code is assigned to the stored event.

[Embodiments of the invention] An embodiment of the present invention will be described below with reference to FIG.

In the figure, (9) is a zero counter that counts consecutive zero quantization coefficients, (1o) is a threshold setting unit that sets a threshold,
(11) is a determination unit that compares the count value with a threshold value and determines whether it is large or small; (12) is an event storage unit that temporarily stores an event that is a set of a non-zero quantization coefficient value and the count value at that time; (13
) is a code assignment unit that assigns a code to an event, and the rest is the same as in FIG. 3.

Moreover, FIG. 2 is a flowchart diagram for explaining the operation.

Next, the operation will be explained with reference to FIG. Similar to Fig. 3, the input image signal for one frame is digitized (1
0 1) is divided into blocks of N×N pixels and linearly transformed, the scan conversion unit (3) rearranges the transform coefficients and outputs a transform coefficient sequence F (n) (104). The encoding control unit (8) calculates the quantization step size (1 1
0) and outputs it to the quantization section (4). Similarly, the threshold value setting section (10) determines the threshold value (1 1 2) from the remaining buffer amount (1 0 9) and outputs it to the determination section (11). Also, as an initial setting, the count value of the zero counter (9) (
1 1 1) to zero, clear the memory contents of the event storage unit (12), and create a conversion coefficient sequence F (
n) (104) is set to 1. Quantization section (4
), F is one of the transformation coefficient sequences F (n) (104)
(i) is quantized with a quantization step size (1 1 0), and Q(i), which is one of the quantized coefficient sequences Q(n) (1 0 5), is output. If the value of Q(1) is not zero, the event storage unit (12) stores the set of the count value (111) of the zero counter (9) and Q (i), which is a non-zero coefficient, as an event. Thereafter, the zero counter (9) is reset to zero, and if Q (i) is not the last quantized coefficient, quantization of the next transform coefficient is continued.

When the value of Q (i) is zero, 1 is added to the count value (1 1 1) of the zero counter (9). Next, the determination section (11
) is the input count value (1 1 1) and threshold value (1 1
2) and outputs the determination result (113) to the event storage unit (12). In the event storage unit (12), the judgment result (1 1 3) is the count value (111)≧threshold (1 1
2) or Q (i) or the quantized coefficient sequence Q (n
) (105), output the currently stored event (114) and end the processing of that pixel block. The code assignment unit (13) assigns a Huffman code to the output event (114) and
OB is added and output to the transmission buffer (7) as encoded data (1 0 7). On the other hand, if there is no output event (114), it is an invalid block, so the code representing the invalid block is changed to the encoded data (1 0 7
) to the transmission buffer (7).

On the other hand, the quantization in the quantization unit (14) is terminated by the determination unit (
Controlled by the determination results (1 1 3) from 11),
When the count value (1 1 1)≧threshold value (1 1 2), the quantization process is terminated for the subsequent transform coefficients.

Furthermore, in the example of FIG. 5, for example, when the threshold value is set to 4 or 5, the events stored in the event storage section (12) and the number of coefficients to be quantized by the quantization section (4) are as follows. become.

When the threshold is 4, there are four consecutive zero coefficients from Q(5) to Q(8), which satisfies the condition for quantization truncation, and the event stored as a set of zero run and non-zero coefficient value is (0.20). (
2.15), and the number of transform coefficients to be quantized is Q(
There are eight items from 1) to Q(8).

When the threshold value is 5, there are five consecutive zero coefficients from Q(14) to Q(18), so the number of events stored is (0.20). (2.
15), (4. 5), (3. 2), and the number of coefficients to be quantized is 18 from Q(1) to Q(18).

As mentioned earlier, the intensity of transform coefficients generally decreases from low frequency to high frequency components, so the quantized coefficient sequence Q (n) (105) as a result of quantization also becomes zero continuously as n increases. There is a high probability that it will happen. Therefore, the threshold (1
12), the transmission range of coefficients is restricted and the number of coefficients that require quantization is reduced, and at the same time, the amount of generated information is also reduced. Therefore, the value of the threshold (1 1 2) is changed to the remaining buffer amount (1 0
9), the amount of information generated can be smoothed more precisely by adaptive feedback control.

In addition, according to this embodiment, not all transform coefficients are necessarily quantized, so if only the necessary transform coefficients are determined, the amount of arithmetic processing for determining the transform coefficients can be reduced and further effects can be obtained. can get.

In the above embodiment, a combination of two-dimensional linear transformation and quantization has been described, but similar effects can be obtained with a combination of one-dimensional, three-dimensional, etc. linear transformation and quantization.

[Effects of the Invention] As described above, according to the present invention, it is determined whether or not to quantize and encode subsequent transform coefficients based on the number of consecutive zero quantization coefficients. This has the effect of reducing the amount of calculations and processing time required to generate events for invalidation identification and two-dimensional variable length encoding.

[Brief explanation of drawings]

Figure 1 is a professional ivy diagram explaining one embodiment of the present invention, Figure 2 is a flow diagram explaining the operation of the present invention, Figure 3 is a block diagram of a conventional example, and Figure 4 is a transform coefficient block diagram. FIG. 5 is a diagram for explaining the properties, and FIG. 5 is a diagram for explaining the assignment of codes. (1) is a blocking section, (2) is a linear conversion section, (3) is a scan conversion section, (4), (14) is a quantization section, (5)
is a valid/invalid identification unit, (6) is an encoding unit, (7) is a transmission buffer, (8) is an encoding control unit, (9) is a zero counter, (10) is a threshold setting unit, and (11) is a judgment unit. Part, (12)
is an event storage section, (13) is a code assignment section, (105) is a quantized coefficient sequence, (1 1 1) is a count value, (1 1 2
) is the threshold, (1 1 3) is the judgment result, (1 1 4)
is an event. In the drawings, the same reference numerals indicate the same or corresponding parts.

Claims (1)

[Claims] A transform coding method in which an input signal sequence is blocked, then linearly transformed, and transform coefficients are sequentially quantized and encoded from low to high frequencies in a transform domain, comprising: means for counting the number of consecutive zero coefficients among the quantized coefficients quantized with the quantization characteristic; and counting by the counting means until a non-zero coefficient and the non-zero coefficient appear from the quantized coefficient sequence. a means for storing the number of continuous zero coefficients as a set in units of blocks; and means for quantizing subsequent transform coefficients and assigning codes to each of the stored sets when the counted value of the continuous zero coefficients exceeds the threshold value. A transform encoding method characterized by: