JPH0270127A - Conversion encoding system - Google Patents

Abstract

PURPOSE:To improve transmission efficiency by variable length-encoding a linear quantizing result based on a quantizing table and transmitting the minimum conversion result of the encoding quantity. CONSTITUTION:A still picture for a transmission part 11 is divided in a block divider 100, and they are transmitted to a distributed computing circuit 102, which obtains a square sum as distributed information. A block for the converter 101 is conversion-processed by a coefficient from a coefficient memory 103, and that with least errors is selected. The conversion result is quantized in a quantizer 104, and the pattern of a bit pattern table 105 is decided in accordance with the dispersion of the circuit 102. A quantizing value is stored in a memory 106, and a value by an inverse quantizer 107 and the conversion result before quantization are compared in an error evaluation circuit 108. A minimum value detector 109 obtains data with minimum error and encoding data is stored in the memory 106. The data, two bit pattern information from the circuit 102 and four bit conversion type information from the detector 109 are multiplexed by a multiplexer 110 and transmitted to a transmission line 12. Thus, transmission efficiency can be improved.

Description

[Detailed Description of the Invention] [Object of the Invention] The present invention relates to a conversion encoding method for converting and encoding various types of information such as still images, moving images, audio, etc. and transmitting the converted information. This invention relates to a transform encoding method that adaptively switches between

(Prior art) In recent years, with the development of video conference systems and video telephone systems, communication systems and various information processing systems have
There is an increasing need to capture information such as audio and images on a platter. For this reason, data compression techniques have been developed that eliminate the redundancy of data such as audio and images that consume a huge amount of information and enable efficient data transmission. One of the known methods is the conversion 7coding method. This transform encoding method divides data to be transmitted into a plurality of blocks, performs a predetermined transform such as discrete cosine transform on each block, quantizes the transform result, and transmits the result. According to this method, effective data compression can be achieved by not transmitting transform coefficients corresponding to high-order basis functions with low power or by reducing the number of quantization bits.

By the way, as one of this type of transform encoding method, a method is known in which the transform encoding method is adaptively switched according to the situation of the image within the block (Japanese Patent Laid-Open No. 61-28586).
No. 9). According to this method, the image quality is improved because a transform encoding method suitable for each block is selected, but it has not always been properly evaluated from the perspective of transmission efficiency, and the transmission efficiency is not sufficiently improved. I couldn't let it happen.

(Issue to be Solved by the Invention 'XU) As described above, in the conventional transform encoding method in which the transform method is adaptively switched according to the information situation, it has not been possible to sufficiently improve transmission efficiency.

SUMMARY OF THE INVENTION An object of the present invention is to provide a transform encoding method that can solve the above problems and sufficiently improve transmission efficiency.

[Structure of the Invention] (Means for Solving the Problems) The present invention provides a block dividing means for dividing information to be transmitted into a plurality of blocks, and a plurality of types of conversion for each block divided by the block dividing means. a conversion means for performing
computing means for obtaining block dispersion information indicating the degree of dispersion of information in each block; and selecting one quantization method from a plurality of quantization methods based on the block dispersion information obtained by the computing means; quantization means for quantizing each transformation result; detection means for detecting a transformation result that minimizes the encoding error of the quantization result of this quantization means; and the transformation result detected by this detection means and determining it. and information indicating the type of conversion used.

The present invention also provides a block dividing means for dividing information to be transmitted into a plurality of blocks, a converting means for performing multiple types of transformation on each block divided by the block dividing means, and a block dividing means for dividing information to be transmitted into a plurality of blocks. quantization means for quantizing each of the conversion results of the quantization means, means for variable-length encoding each quantization result of this quantization means, and a code amount of each variable-length code obtained by this means. The present invention includes a detection means for detecting the minimum conversion result, and a means for multiplexing and transmitting the conversion result detected by the detection means and information indicating the type of conversion used to obtain it. It is a feature.

(Function) According to the present invention, one quantization method is selected from a plurality of quantization methods based on the block dispersion information indicating the degree of dispersion of information within the block. Fewer bits are allocated, and transmission efficiency can be improved without deteriorating image quality.

Furthermore, according to the present invention, after converting results based on a plurality of conversion methods are quantized, variable length coding is performed, and as a result, the conversion result with the least amount of code is selected and transmitted, which improves transmission efficiency. can be improved.

(Example) Hereinafter, an example of the present invention will be described based on the drawings.

FIG. 1 is a diagram showing the configuration of an image transmission system to which a transform encoding method according to an embodiment of the present invention is applied.

First, a still image inputted to the transmitter 11 is divided into blocks by the block divider 100 and provided to the converter 101 and the distributed calculation circuit 102 .

The variance calculation circuit 102 calculates the sum of squares as variance information for each block. Note that this dispersion information does not need to be dispersion in a strict sense, and may be something like the sum of absolute values, which indicates the degree of variation in density of the block. On the other hand, a plurality of conversion processes are sequentially performed on a block provided to the converter 101 based on a plurality of types of coefficients sequentially read out from the coefficient memory 103 . As the conversion process, for example, 16 types of conversion methods expressed in 4 bits are available, such as, for example, discrete cosine conversion, KL transformation, Barr transformation, Fourier transformation, and Hadamard transformation. Among these conversion methods, the one that causes the least error is selected through subsequent processing. converter 1
The conversion results based on each conversion method obtained in step 01 are quantized in a quantizer 104. At this time, the quantization bit pattern indicating how to allocate the number of quantization bits to each transform coefficient is, for example, 4 which can be expressed by 2 bits.
Various types of bit patterns are provided, and these are stored in the bit pattern table 105. These 4
Which pattern to use among the types of quantization bit patterns is determined according to the variance calculated by the variance calculation circuit 102. Taking discrete cosine transform coefficients as an example,
As shown in Figure 2 (a), (b), , (c) and (d), the culm with large dispersion, “1” in the upper left region on the conversion surface.
” is selected such that the area to which it is allocated is large. When the conversion result based on each conversion method is quantized using the bit pattern determined based on the variance, the quantized value is stored in the memory 106. While being stored, it is input to the dequantizer 107 and dequantized.Then, the conversion result before quantization and the dequantized value are compared in the error evaluation circuit, and the minimum value detector 109 sequentially The one with the smallest error is found. During this time, the encoded data obtained by the conversion method with the smallest error is always stored in the memory 106. When the errors are evaluated for all the conversion methods, the final result is The encoded data stored in the memory 106 becomes the data with the minimum error.

The encoded data is multiplexed with 2-bit bit pattern information from the dispersion calculation circuit 102 and 4-bit conversion type information from the minimum value detector 109, and then sent to the transmission line 1.
2.

In the receiving unit 13, a divider 111 divides the received signal into encoded data, bit pattern information, and conversion method information. The encoded data is given to the dequantizer 112 and dequantized based on the bit pattern specified by the bit pattern information. The inverse quantized data is stored in the coefficient memory 1 in the inverse transform circuit based on the transform method information.
Inverse transformation is performed based on the coefficient information given from 15. The inverse transform result is sent to the block synthesizer 116 as a locally reproduced image.
The blocks are combined here to obtain a reconstructed image.

According to this system, in addition to obtaining good image quality by performing conversion using the best conversion method with few errors, the best bit pattern is selected based on the calculation result in the distributed calculation circuit 102. Therefore, the amount of coding for blocks with small variance is reduced, and transmission efficiency can be improved.

In addition, in the above system, the variance is calculated for one block, one quantization bit pattern is selected corresponding to it, and the most suitable transformation is selected. It is also possible to select a number of stages of quantization bit pattern. Furthermore, it is also possible to select, for one block, a quantization pattern with a number of stages corresponding to a plurality of approximate variances, although not equal variances. In that case, it is sufficient to perform error evaluation using all the transformations for the plurality of selected quantization patterns, and select the combination of transformation and quantization with the smallest error. Note that as described above, the degree of variation in information within each block may be used as block dispersion information, and the sum of absolute values may be used.

FIG. 3 shows another embodiment of the invention. In this embodiment, without obtaining block distribution information, each transform change is quantized using a fixed quantization method, variable length coding is performed, and the transform method that minimizes the amount of code is selected. .

The input image inputted to the transmitter 21 is sent to the block divider 2
00 is divided into blocks, for example, 8×8, and the converter 20
1 is given. A plurality of conversion processes are sequentially performed on the block supplied to the converter 201 based on the conversion coefficients sequentially read out from the coefficient memory 202. As the conversion process, for example, 256 types of KL conversion expressed in 8 bits set in advance are used. The conversion result is sent to the quantizer 20
Linear quantization is performed at 3. This quantization is performed based on the quantization table 204. This quantization table 204
is composed of divisors set for each of the 8×8 blocks, as shown in FIG. 4, for example. The quantization result is the quotient of the conversion result of the converter 201 divided by this divisor. At this time, the numbers below the decimal point may be rounded down or rounded off. FIG. 4(a) is an example in which the upper right divisor where low-order terms are concentrated on the conversion surface is set small, and the lower left divisor where high-order terms are concentrated is set large. In this way,
It becomes possible to allocate a large number of bits to low-order terms that are effective as information. In the same figure (b), all the divisors are the same value “1”.
In this example, it is set to 6”.

In this case, the conversion table 204 becomes substantially unnecessary. Since the quantization result of the quantizer 203 has an indefinite length, it is variable-length encoded by the variable-length encoder 205. The encoding result is given to a code amount counter 206 and a memory 207. The code amount counter 206 is the variable length encoder 20
The amount of encoding of each of the 256 types of encoded data in 5 is counted. The minimum value detector 208 detects the one having the minimum encoded amount among the counted encoded amounts, and stores 8-bit data indicating the number (conversion number) in the memory 2.
Output on 07. Based on this data, memory 207 selects one encoded data from among 256 types of encoded data and outputs it to multiplexer 209 . Multiplexer 2
09 is also given 8-bit data (matrix index) indicating the type of conversion. The multiplexer 209 multiplexes the encoded data and the matrix index and transmits the multiplexed data to the receiving unit 23 via the transmission path 22.

In the receiving section 23, a divider 210 divides the received signal into encoded data and matrix index.

The encoded data is decoded by a decoder 211 and then dequantized by a dequantizer 212. This inverse quantization is obtained by multiplying the decoding result by a multiplier stored in the quantization table 213.

The quantization table 213 has the same contents as the quantization table 204 provided in the transmitter 21.

The inverse quantization result is inversely transformed in an inverse transformer 214. This inverse transformation is performed using coefficients read from coefficient memory 215 based on the matrix index. The inverse transformation result is given to the block synthesizer 116 as a locally reconstructed image, where the blocks are synthesized to obtain a reconstructed image.

According to this system, the linear quantization result based on the quantization table 204 is variable-length coded, and the conversion result with the minimum amount of code is transmitted, so that transmission efficiency can be improved.

Note that in this embodiment, the quantization table is set fixedly, but if it is possible to increase or decrease each divisor of the quantization table 204, the amount of code can be increased overall. Depending on the application, it is possible to improve the image quality or increase the transmission speed.

Furthermore, like the system described above, the quantization table 204 may be switched in stages based on block distribution information. In this case, in order to match the contents of the quantization table between the transmitting section and the receiving section, it is necessary to transmit information indicating the state of the quantization table together with the transmission data.

Note that in the above embodiments, two-dimensional transformation of images has been described, but the present invention can be similarly applied to one-dimensional transformation encoding. Furthermore, a similar configuration can be adopted for audio signals. Concerning transformation methods, in addition to using cosine transformation, which is a typical orthogonal transformation, other orthogonal transformations such as Fourier transformation, Hadamard transformation, Slant transformation, and Barr transformation, as well as transformation methods expressed by non-orthogonal linear regular matrices, Conversion may also be used. Alternatively, a nonlinear transformation that includes an inverse transformation can also be used.

[Effects of the Invention] As described above, according to the present invention, in a transform encoding method that adaptively selects an optimal transform method from a plurality of transform methods, quantization is performed based on dispersion information within a block. By determining the method and by selecting the transform result with the smallest code amount of the variable length coding result after quantization, the transmission efficiency can be sufficiently increased.

[Brief explanation of the drawing]

FIG. 1 is a block diagram of a still image transmission system according to an embodiment of the present invention, FIG. 2 is a diagram showing the contents of a bit pattern table in the same system, and FIG. 3 is a block diagram of a still image transmission system according to another embodiment of the present invention. FIG. 4, a block diagram of the image transmission system, is a diagram showing the contents of a quantization table in the system. 11.21... Transmission unit, 12.22... Transmission path, 1
3.23...Reception unit, 100,200...Block divider, 101,201...Converter, 102...Dispersion z1' calculation circuit, 103,115,202,215...
・Coefficient memory, 104.203...Quantizer, 105
゜113...Bit pattern table, 106゜20
7...Memory, 107,112,212...Inverse digitizer, 108...Error evaluation circuit, 1.09,208...
... Minimum value detector, 110,209 ... Multiplexer, 1
11.210...Divider, 114,214...Inverse transformer, 116,216...Block combiner, 04°2
13... Quantization table, 205... Variable length encoder, 206... Code amount counter, 211... Decoder. Applicant's agent Patent attorney Takehiko Suzue (a) 00゜(b) 'ot. (c)=10'' (d) 11J 2nd rlA Figure 1 (a) (b) Figure 4

Claims (2)

[Claims] (1) A block dividing means for dividing information to be transmitted into a plurality of blocks; a converting means for performing multiple types of transformation on each block divided by the block dividing means; a calculation means for obtaining block dispersion information indicating the degree of dispersion; and quantization for selecting one quantization method from a plurality of quantization methods to quantize each of the conversion results based on the block dispersion information obtained by the calculation means. a detection means for detecting a transformation result that minimizes the encoding error of the quantization result of the quantization means; and a detection means for detecting the transformation result detected by the detection means and the type of transformation used to obtain the transformation result. 1. A transform encoding method characterized by comprising means for multiplexing and transmitting information. (2) block division means for dividing information to be transmitted into a plurality of blocks; transformation means for performing multiple types of transformation on each block divided by this block division means; quantization means for quantizing each of the transformation results; means for variable-length encoding each quantization result of the quantization means; and a transformation result that minimizes the code amount of each variable-length code obtained by this means. and a means for multiplexing and transmitting the conversion result detected by the detection means and information indicating the type of conversion used to obtain the conversion result. method.