JPS5832824B2 - Chiyotsukouhenkanfugoukahoushiki - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION The present invention relates to an encoding method using orthogonal transformation, and the present invention relates to an encoding method using orthogonal transformation. The purpose is to reproduce images with few images.

If we look at the power distribution of television signals statistically, we can see that there are few high-frequency components, and therefore it is possible to compress the overall band by reducing the number of transmission bits used for the high-frequency components. I have a way of thinking.

An example of this is a method using Agemar transformation as a means of converting the time domain to the frequency domain.

For example, if an 8th-order Hadamard transform is performed, it can be expressed as the following equation.

Here, (xi, x2...X8] is the value obtained by sampling 8 points on the time series, CH) is the Hadamard transformation matrix, and hl, h2... ... h8 is an 8th order sequence transformed into the frequency domain by Hadamard transform.

In a sequence converted in this way, when looking at the properties of the image statistically, there is a certain tendency in the energy distribution.

Therefore, by allocating a number of bits close to the original number of bits (generally 2 bits) to sequences with a large energy distribution, and by allocating a small number of bits to sequences with a small energy distribution, the average number of bits per pixel is reduced. can do.

This reduces the transmission bit rate and can be considered equivalent to compressing the band.

Specifically, when a sample per -pixel is quantized to 8 bits by A/D conversion, the total number of bits at 8 points from point X1 to point X8 in equation (1) is 64 bits.

Next, the total number of bits of the eight sequences from hl to h8 subjected to Hadamard transformation is 64 bits, assuming 8 pits per sequence, and no information is increased or decreased in this transformation.

However, if we utilize the characteristics of the image energy distribution mentioned earlier and allocate the bits as shown in Figure 1, the total number of bits will be 36 bits, with an average of 4 bits per sequence.
．． It becomes 5 bits.

This is about half the original 8-bit data, which means that it has been compressed to about half.

Now, if we linearly compress sequence h4 to 3 bits, conventionally the quantization step is 1 and the lower 3 bits b
Lb2. b3 was being transmitted [see Figure 2 a].

In some cases, b2. b3. There are cases where 3 bits of b4 are transmitted [
See Figure 2].

In this method, the range that can be represented by the three transmission bits changes depending on the quantization step, but the number of bits required for transmission is fixed at the initial setting and does not correspond to changes in the input signal.

However, although it is true that the power distribution of high-order sequences is small from a statistical perspective, in special images, the energy may be stronger than expected.

For such an image, the value of the sequence h4 becomes larger than the original 3 bits (for example, the lower 3 bits), and the flow from the lower 3 bits ends up being changed to the lower 3 bits. Since only small values can be transmitted even though the values are originally large, large errors occur, which has the disadvantage of causing distortion in the reproduced signal.

If the probability of overflow is small in this way, it will not be noticeable on the screen, but if the probability of overflow increases, the deterioration of the screen will become noticeable.

The present invention was made to deal with such a problem, and detects the number of overflows of each sequence, and for sequences with a large number of overflows, the quantization step is performed without increasing the number of transmission bits. By switching and coarsening the value, the limit of the maximum value to be transmitted is increased.

This means that the accuracy will be rough, but the value will be transmitted as a value close to the original value, and the reproduced value of the inverse transform value in the inverse Hadamard transform on the receiving side will be more accurate, resulting in distortion. This means that images with fewer images can be transmitted.

Here, even if the number of values to be transmitted increases in the event of an overflow, if the number of bits allocated to a certain sequence itself is increased, the overall bit rate will increase and the transmission rate will increase.

This not only increases the transmission rate, but also increases or decreases the number of bits in a certain sequence in the hardware configuration, which becomes very complicated, which causes many problems.

From this perspective, the number of bits to be allocated is fixed, and when the sequence value overflows without changing the transmission rate, the quantization step is coarsened to send information close to the correct value.

For example, as shown in FIG. 3a, the lower three bits are normally transmitted, and only when there is an overflow due to a special image, the three bits of the quantization step are transmitted as shown in FIG.

Figure 3 is b4. b5. Not limited to b6 but b3. b4. b5
Even b5. It may be b5, b7 or any other 3 bits, but which 3 bits are used depends on the energy at the time.

An embodiment of the present invention will be described below based on the drawings.

First, the configuration of the encoding side of the present invention will be explained using FIGS. 4 and 5.

The analog video signal supplied to the input terminal A is converted into a parallel 8-bit digital signal by the A/D conversion circuit 1.

This signal is supplied to the Hadamard transform circuit 2, and each sequence h1......hi......h
We get an output of 8.

Now considering a block of sequence hi, the output of hi will be an 8-bit signal if compression is not considered.

This signal is stored in the hi register 3.

For example, considering h4 as hi, the lower 3 of normal register 3
Suppose we are transmitting bits.

However, consider a case where the value of h4 overflows from the lower three bits due to a special image (input signal).

The following explanation will be made assuming that the current switching criterion is that the number of overflows up to the 5th bit is 4 or more times.

The 5th bit signal of the hi register 3 [see Figure 5] is gated with the reference clock supplied to the terminal B by the AND circuit 4, and is supplied to the clock terminal of the probability counter 5 [see Figure 5C]. ].

This probability counting counter 5 is for counting the number of times overflow occurs during a certain period.

Therefore, the reset terminal of this counter 5 is supplied with a reset pulse having a period of a certain period.

That is, if a certain period is one vertical scanning period, a pulse with a vertical synchronization period is supplied to terminal C, and if a certain period is one horizontal scanning period, a pulse with a horizontal synchronization period is supplied to terminal C. [See Figure 5a].

If this counter 5 reaches 4 times or more, an output is obtained and the r 5th
(see FIG. 5), and is held in the fixed period register 6 for the next fixed period (see FIG. 5, f).

Since the output of the counter 5 appears in the middle of a fixed period, it is often inconvenient to switch it in the middle, so the register 6 is used to hold it for the next fixed period.

In other words, switching is performed from the next cycle after the overflow, but there is no practical problem because the images have a lot of correlation.

The output of the register 6 obtained in this manner is supplied to a quantization step switching circuit 7 and a mode instruction signal gate circuit 8.

The quantization step switching circuit 7 normally transmits in the mode of the lower 3 bits, but when an overflow is detected and there is an output in the register 6, the quantization step switching circuit 7 transmits in the mode of the other 3 bits [for example, b4. b5°b6) and transmitted.

Furthermore, the mode instruction signal gate circuit 8 gates and transmits which mode is being transmitted during the certain period using a mode instruction signal gate pulse g (see FIG. 5g).

This signal is inserted into the blanking period.

The waveform is shown in FIG. In FIG. 6, a is a horizontal or vertical synchronizing signal, and b is a transmission signal.

In b, ``a'' is a mode instruction signal, and in order to simplify the explanation, only one pulse is shown in the figure, but in reality, the sequence number hi (i = 1.2
, . . . song... 8) is transmitted.

The mouth is a digitized, orthogonally transformed and compressed video signal.

In this way, the encoding side receives mode instruction signals m1 to m8.
and digital signals Bil and Bi whose modes have been switched.
2°Bi3 etc. can be obtained.

Next, the decoding side will be explained based on FIG.

In FIG. 7, gate circuit 110 input terminals M1 to M8
are supplied with mode instruction signals m1 to m8 issued on the encoding side.

A mode instruction signal gate pulse is supplied to the input terminal CF).

The mode instruction signal obtained here is held in the mode register 12.

Now describing the sequence hi, it is a compressed digital signal Bi 1 .

Bi 2. Bi3 is switched to the mode instructed by the mode changeover switch 13.

That is, normally the lower 3 bits are being transmitted, and in overflow mode, B4. Since the data of BS and B6 has been transmitted, it is stored in the hi register 14 as such.

This signal is supplied to the Hadamard inverse transform circuit 15 and is inversely transformed to become a digital 8-bit signal in the time domain.

This digital 8-bit signal is converted into an analog video signal by the D/A conversion circuit 16 and is obtained at the output terminal G.

Although two switching modes have been described above, it is possible to use even more modes.

In this case, the mode instruction signal is a multi-bit signal indicating which area to send.

By the way, when a fixed period is defined as one horizontal period, switching is performed every horizontal period, so for video that has a quick response and requires switching the entire screen, the first few lines can be used without having the next field. Switching is possible.

Another advantage is that the number of stages of the probability counting counter is small.

However, when a part of the screen changes rapidly, such as in a wind pattern, when the screen is almost stationary, it may appear that there is a border line between the switched lines because the quantization steps are different. be.

Furthermore, when the constant period is one vertical period, contrary to the above case, switching is performed in field units, so the border line due to switching is not visible on the screen.

Furthermore, due to the characteristics of the eye, switching on a field-by-field basis has the advantage of being unnoticeable.

However, conversely, the number of stages of the probability counting counter increases.

As described above, according to the orthogonal transform encoding method of the present invention, the following effects can be obtained.

Conventionally, in orthogonal transform coding, the quantization step and number of transmission bits for each sequence are statistically set to certain values, which overflows for special images and causes image deterioration. By switching the number of quantization steps without increasing the number of transmission bits, overflow can be prevented and images with less deterioration can be transmitted.

[Brief explanation of drawings]

Figure 1 is an explanatory diagram showing an example of compressing the bits of each sequence subjected to orthogonal transformation. 3 is an explanatory diagram of a transmission area different from that in FIG. 2, FIG. 4 is a circuit diagram on the encoding side in an embodiment of the present invention, FIG. 5 is a waveform diagram at each point in FIG. 4, and FIG. 6 7 is a waveform diagram showing the insertion state of the mode instruction signal, and FIG. 7 is a circuit diagram on the decoding side in an embodiment of the present invention. 1...A/D conversion circuit, 2...Hadamard conversion circuit, 3...hi register, 4...
...AND circuit, 5, ...probability counting counter,
6... Constant cycle register, 7... Quantization step cutting circuit, 8... Instruction signal gate circuit, 11... Gate circuit, 1286010.
Mode register, 13...mode selection switch, 14...hi register, 15...Hadamard inverse conversion circuit, 16...D/Ai conversion circuit.

Claims (1)

[Claims] 1 In a television signal encoding method using orthogonal transformation, the statistics of each sequence during a certain period are detected, and the number of quantization steps is determined according to the statistics while keeping the number of bits of each sequence fixed. An orthogonal transform encoding method characterized by switching the amplitude region and transmitting it by switching.