JP3217507B2 - Image compression device - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention divides image information into a plurality of blocks, performs discrete cosine transform for each block, converts the data into frequency data, and quantizes each frequency data.
The present invention relates to an image compression apparatus for compressing by allocating a Huffman code thereafter.

[0002]

2. Description of the Related Art One of the techniques for compressing image information with high efficiency is a compression method based on DCT (Discrete Cosine Transform).

The DCT method is an irreversible method in which an original image cannot be completely restored from compressed data because it includes quantization. However, since a high compression rate can be achieved, image information, especially a medical image of a huge amount of data, is required. It is effective for transfer and storage, and tends to dominate international standard methods.

[0004] The compression method based on the DCT method is as follows. First, an input image (original image) is divided into a plurality of blocks with a block size of n × n pixels, and each block is two-dimensionally divided.
The data is subjected to CT conversion, and converted into n × n DCT coefficients per block.

The DCT coefficients are quantized at a different step size for each coefficient position using a quantization table, and each quantized data is rearranged one-dimensionally by zigzag scanning.

The DCT coefficients are divided into zero (ineffective coefficient) and non-zero (effective coefficient), and are separately encoded. That is, the effective coefficient is queried in the Huffman code table, and a variable-length code (Huffman code) corresponding in advance based on the general frequency of occurrence of each coefficient, that is, a Huffman code with a short bit length is assigned to a high-frequency code. Codes with a long bit length are assigned to those that are assigned and have a low appearance frequency.

On the other hand, the invalid coefficient is encoded by a zero run-length encoding method. That is, first, the number of consecutive zeros (hereinafter referred to as “zero run length”) is counted. The zero run length is referred to a zero run length Huffman code table created exclusively for zero run length, and a Huffman code corresponding in advance based on the general occurrence frequency of each zero run length, that is, the occurrence frequency, Are assigned a Huffman code with a short bit length, and those with a low appearance frequency are assigned a code with a long bit length. However, this compression method has the following problems.

First, the frequency of occurrence of each zero run length is:
The threshold, the block size, the step size of the quantization, and the characteristics of the input image are variously changed, so that it is difficult to create a Huffman code table that can appropriately respond to all the input images and achieve a high compression ratio. Therefore, the compression effect by the zero run-length encoding does not always appear well.

That is, if the zero run length of the current input image having a high frequency of occurrence is a low frequency of a general frequency of occurrence, a Huffman code having a long bit length is assigned, or the frequency of occurrence of the current input image is low. This is because if the zero run length is high at a general occurrence frequency, a Huffman code having a short bit length may be assigned.

Second, if the Huffman code table for the effective coefficient and the Huffman code table for the zero run length are commonly used, the allocated bit length may be long, so a dedicated Huffman code table for the zero run length is required. It is necessary to have a separate Huffman code table for the effective coefficient, which increases the circuit scale.

Third, when the block size is increased, for example, for a high compression ratio, the maximum length of the zero run length is correspondingly increased, and the types of Huffman codes to be handled are increased. The average bit length becomes longer, and as a result, the compression ratio may decrease.

[0012]

SUMMARY OF THE INVENTION The present invention has been made in order to cope with the above-described circumstances, and the first object of the present invention is to always achieve a stable compression ratio, and the second is to provide a zero run length. By using a part of the Huffman code table for the effective coefficient for the Huffman code table, there is no need for a Huffman code table dedicated to zero run length,
A third object is to provide an image compression apparatus capable of reducing the circuit scale by sharing the Huffman code table between the effective coefficient and the invalid coefficient.

[0013]

According to the present invention, there is provided an input means for inputting image information, a means for dividing the image information into a plurality of blocks of a predetermined size, a frequency conversion process and a quantization for each block. Means for arranging and outputting each symbol obtained by sequentially executing the processing in one dimension,

A first encoding means for allocating the symbol to an invalid coefficient indicating zero and an effective coefficient other than zero, and referencing the effective coefficient to a predetermined code table to obtain a code; A second encoding unit that counts a continuous number and assigns any one of the same bit length codes of a part of the code table to the continuous number;
And means for one-dimensionally arranging and outputting the output of the second encoding means as before.

[0015]

According to the image compression apparatus of the present invention, the image information input via the input means is obtained by dividing the image information into a plurality of blocks of a predetermined size and sequentially performing the frequency conversion processing and the quantization processing on each block. Each symbol is arranged one-dimensionally, these symbols are assigned to an invalid coefficient indicating zero and an effective coefficient other than zero, the effective coefficient is queried by referring to a predetermined code table, and the number of consecutive invalid coefficients is determined. By assigning any one of the same bit length codes of the code table to the above, a stable compression ratio can always be achieved, a Huffman code table dedicated to zero run length can be eliminated, and the Huffman code table is used as an effective coefficient. And the invalid coefficient, the circuit scale can be reduced.

[0016]

An embodiment of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of an image compression apparatus according to one embodiment of the present invention. In FIG. 1, reference numeral 1 denotes an input terminal 1 for inputting an analog image signal. The input processing device 2 digitizes the analog image signal input from the input terminal 1 and time-multiplexes and outputs a luminance signal and a color difference signal as appropriate. The blocking circuit 3 divides the digital image signal from the input processing device 2 into a plurality of blocks of a predetermined size (n × n pixels).

Discrete cosine transform circuit (DCT circuit) 4
Performs a two-dimensional DCT (Discrete Cosine Transform) on each block from the blocking circuit 3 to convert it into n × n DCT coefficients representing horizontal and vertical spatial frequencies.

The quantization circuit 5 linearly quantizes the output from the discrete cosine transform circuit 4 with a different step size for each coefficient position using an appropriate quantization table.
The T coefficient is converted into a quantum symbol (hereinafter simply referred to as a symbol), and each symbol is arranged one-dimensionally by, for example, zigzag scan and output.

The encoding circuit 6 discriminates the symbol from the quantization circuit 5 into an invalid coefficient indicating zero in the distribution circuit 7 and an effective coefficient other than zero, and outputs the effective coefficient to the Huffman encoding circuit 8.
Further, the invalid coefficient is converted to a zero run-length Huffman coding circuit 10.
The Huffman code table 9 stored in the ROM, for example, is shared, and a Huffman code is assigned to each and output.

This zero run-length Huffman coding circuit 1
0 counts the number of consecutive invalid coefficients (zero symbols) (zero run length) from the quantization circuit 5 and is a specific Huffman code in the Huffman code table, that is, a Huffman code having the same bit length, for example, 6 bits. Assign one of Therefore, even if the Huffman code table 9 is shared with the Huffman encoding circuit 8, the Huffman code can be distinguished from the Huffman code of the effective coefficient. The problem of not appearing well can be avoided, and a stable compression ratio can always be achieved. The zero run length Huffman code is added with a zero run length precursor to add
Sent to

The frame generation circuit 11 arranges the Huffman codes from the encoding circuit 6 one-dimensionally according to the original pixel positions, and appropriately adds a control word such as a synchronizing signal so as to be suitable for transmission or recording. After the framing, the data is output to the transmission path or the recording device 12. Next, the operation of the present embodiment configured as described above will be described. The block size is 4 × 4 and the bit length is 6
The following description is based on the assumption that seven types of Huffman codes of bits are applied to zero run length. FIG. 2 is a diagram showing each data in each processing from block processing to one-dimensional arrangement through discrete cosine transform and quantization processing. For example, an analog image signal input to the input processing device 2 via the input terminal 1 is digitized there, and a luminance signal and a color difference signal are time-multiplexed as appropriate.

The digitized image signal is sent to a blocking circuit 3, where it is divided into a plurality of blocks having a block size of 4 × 4 pixels as shown in FIG. (B) shows the spatial coordinate data (pixel value array data) of a certain block shown by oblique lines in (a). Hereinafter, the flow of processing will be described using this block as an example.

The spatial coordinate data of this block is sent to a discrete cosine transform circuit 4, where a two-dimensional DCT (discrete cosine transform) is executed, and as shown in FIG.
4 × 4 D representing horizontal and vertical spatial frequencies
It is converted into a CT coefficient (frequency data).

The frequency data is linearly quantized by a quantization circuit 5 at a different step size for each coefficient position using an appropriate quantization table, and is converted into a quantum symbol (hereinafter simply referred to as a symbol). Quantized data as shown in (d) is generated. Here, the first stage of compression is completed. The quantized data is output in a one-dimensional array as shown in (f) by, for example, a zigzag scan as shown in (e). The one-dimensionally arranged quantized data is sent to the encoding circuit 6, where it is first classified by the distribution circuit 7 into an invalid coefficient indicating zero and an effective coefficient other than zero.

The effective coefficients are sent to a Huffman encoding circuit 8, where they are referred to a Huffman code table 9 stored in, for example, a ROM, and are assigned Huffman codes.

On the other hand, the invalid coefficient (zero) is sent to a zero-run-length Huffman encoding circuit 10, where the number of continuous (zero-run-length) is counted, and a Huffman code table used as an effective coefficient according to the zero-run-length is used. 9 is used to assign the appropriate Huffman code.

As shown in FIG. 3, the Huffman code used at this time uses a partially specified Huffman code in the Huffman code table 9, that is, seven types of Huffman codes C1 to C7 having the same bit length of, for example, 6 bits. I do. Therefore, even if the Huffman code table 9 is shared with the Huffman encoding circuit 8, the Huffman code can be distinguished from the Huffman code of the effective coefficient. The problem of not appearing well can be avoided, and a stable compression ratio can always be achieved.

FIG. 4 is a diagram showing an encoded zero run length data format. As shown in FIG.
The zero-run-length Huffman code has a fixed length and is sent to the frame generation circuit 11 with a fixed-length zero-run-length pre-cursor added immediately before it.

FIG. 5 (a) shows a data format of a zero run length in which six zeros in the one-dimensional quantized data shown in FIG. 2 (f) are continuous. In this case, a Huffman code C6 corresponding to six consecutive zero run lengths is assigned.

Here, the Huffman code includes 1 to 7
C1 to C corresponding to each of the zero run lengths
There are seven types up to 7. If a zero run length in which eight or more zeros are consecutive occurs, the seven types of Huffman codes C1 to C7 are appropriately combined. FIG. 5B is a diagram illustrating an example of a combination of Huffman codes when 30 zeros are consecutive. FIG. 5 (b)
In this case, the Huffman code C7 corresponding to seven consecutive zero run lengths is repeated four times, and a Huffman code C2 corresponding to two consecutive zero run lengths is added. As described above, even if the number of Huffman codes is reduced, it is possible to cope with zero run lengths of various lengths. In addition, by reducing the types of Huffman codes in this way, it is possible to minimize the pressure on the effective coefficient allocation range.

The Huffman codes coded by the coding circuit 6 are arranged one-dimensionally according to the original pixel positions in the frame generation circuit 11, and are appropriately added with control words such as synchronization signals, and then transmitted or transmitted. After being subjected to framing suitable for recording, it is output to the transmission path or the recording device 12.

As described above, according to this embodiment, a partially specified Huffman code in the Huffman code table, that is, one of seven types of Huffman codes having the same bit length of, for example, 6 bits is assigned to the zero run length. So that it can be distinguished from the Huffman code of the effective coefficient,
Therefore, the Huffman code table can be shared between the effective coefficient and the zero run length, and as a result, a Huffman code table dedicated to zero run length can be eliminated, and the circuit scale can be reduced at the same time. In addition, a stable compression ratio can always be achieved.

Further, since the number of Huffman codes to be assigned to the zero run length is relatively small, for example, seven,
There is also an effect that the effective coefficient allocation range is not significantly affected. The present invention is not limited to the above-described embodiments, and can be implemented with various modifications.

[0034]

As described above, the present invention provides an input means for inputting image information, a means for dividing the image information into a plurality of blocks of a predetermined size, a frequency conversion process and a quantum Means for arranging and outputting one-dimensionally each symbol obtained by sequentially executing the conversion process,

First encoding means for allocating the symbol to an invalid coefficient indicating zero and an effective coefficient other than zero, and referencing the effective coefficient to a predetermined code table to obtain a code; A second encoding unit that counts a continuous number and assigns any one of the same bit length codes of a part of the code table to the continuous number;

Means for arranging the output of the first encoding means and the output of the second encoding means in a one-dimensional manner as before and outputting the image information. Are divided into a plurality of blocks, and the symbols obtained by sequentially executing the frequency conversion process and the quantization process for each block are arranged one-dimensionally, and these symbols are converted into an invalid coefficient indicating zero and an effective coefficient other than zero. A code can be obtained by referring to a predetermined code table for distribution and effective coefficients, and one of the same bit length codes in the above code table is assigned to the continuous number of invalid coefficients, so that stable compression is always achieved. Rate can be achieved, the Huffman code table dedicated to zero run length can be eliminated, and the Huffman code table is shared by the effective coefficient and the invalid coefficient. It can be small.

[Brief description of the drawings]

FIG. 1 is a block diagram of a first embodiment of the present invention.

FIG. 2 is a diagram showing each data in each processing from block processing to one-dimensional arrangement through discrete cosine transform and quantization processing.

FIG. 3 is a diagram showing a correspondence between a zero run length and a Huffman code.

FIG. 4 is a diagram showing a format of compressed data of a zero run length part.

FIG. 5 is a diagram showing an example of compressed data of a zero run length part.

[Explanation of symbols]

DESCRIPTION OF SYMBOLS 1 ... Input terminal, 2 ... Input processing circuit, 3 ... Blocking circuit, 4 ... Discrete cosine transformation circuit, 5 ... Quantization circuit, 6
... Encoding circuit, 7 ... Distribution circuit, 8 ... Huffman encoding circuit, 9 ... Huffman code table, 10 ... Zero run length Huffman encoding circuit, 11 ... Frame generation circuit, 12
... Transmission path or storage device.

Claims (1)

(57) [Claims] An input unit for inputting image information; a unit for dividing the image information into a plurality of blocks of a predetermined size; and a frequency conversion process and a quantization process for each of the blocks. Means for arranging and outputting each symbol one-dimensionally, allocating the symbols to an invalid coefficient indicating zero and an effective coefficient other than zero, and referencing the effective coefficient to a predetermined code table to obtain a code. A second encoding unit that counts the number of consecutive invalid coefficients and assigns one of the same bit length codes of a part of the code table to the number of consecutive invalid coefficients; An image compression apparatus comprising: an encoding unit; and a unit configured to output the output of the second encoding unit in a one-dimensional array as before.