JPH0746407A - Picture data compressing device and picture data restoring device - Google Patents

Abstract

PURPOSE:To efficiently restore a picture with subjectively less deterioration by adopting a nonlinear quantization technique depending on a visual intensity characteristic. CONSTITUTION:Picture data are divided into blocks each comprising a picture element matrix of 8X8 (longitudinal X lateral) for each frame, subjected to the DCT processing by a transforming section 12 and DCT coefficient data of 8X8 are generated. A quantization section 13 quantizes each block of the coefficient data. A spatial frequency quantization table(SQT) and an intensity quantization data table(DQT) are used for the quantization. A 2-dimension matrix obtained by the quantization section 13 is subjected to zigzag scanning in a scanning section 14, and transformed into a one-dimensional vector. Furthermore, a coding section 15 applies run length coding and Huffman coding to the vector. As a result, since a nonlinear quantization technique depending on the visual intensity characteristic is provided by improving a linear quantization technique utilizing the spatial frequency characteristic of the vision, a picture with subjectively less deterioration is efficiently decoded.

Description

Detailed Description of the Invention

[0001]

The present invention relates to, for example, JPEG (Jo
int Photographic Experts
Group) Still image compression algorithm, MPEG (M
moving Picture Experts Grou
p) A compression device for image data applicable to a moving image compression algorithm and the like, and a decompression device for the compressed image data.

[0002]

2. Description of the Related Art Quantization is one of the techniques for compressing image data. This quantization includes linear quantization (linear quantization) used in JPEG and non-linear quantization (non-linear quantization) used in coding television signals for broadcasting. In this linear quantization processing, data conversion is performed based on a predetermined quantization table. Therefore, all the coefficient data within the predetermined range after the DCT (Discrete Cosine Transform) processing are converted into data of the same value. By reducing the number of digits of the coefficient data in this way, the total data information amount of the coefficient data is reduced. In the former case, the quantization step is constant regardless of the intensity level of the image data. In the latter, the quantization step is changed according to the intensity level.

In order to increase the compression rate while suppressing the deterioration of image quality, it is preferable to use non-linear quantization. When the luminance level is low in the image data, the deterioration of the image quality becomes conspicuous when the quantization step is increased. When the brightness level is high, the deterioration of image quality is less noticeable even if the quantization step is increased. A relationship of visual characteristics approximated using a logarithmic function is established between the brightness level and the quantized data.

In the conventional non-linear quantization, since the quantization step is calculated from the brightness level, complicated calculation is performed. Alternatively, a quantization table utilizing visual characteristics approximated by a pseudo logarithmic function is used to reduce the calculation amount.

[0005]

However, in the former case, since complicated logarithmic operation is required, the processing takes time. Therefore, it is difficult to process image data of a moving image in real time, for example. Further, in the latter case, it is not always possible to efficiently realize the ideal characteristics of the visual characteristic function. Therefore, in order to obtain a given compression rate, it is not possible to achieve non-linear quantization that matches the visual characteristics, and in the restored image, visual image quality deterioration may become more apparent than when the visual characteristics match. is there.

[0006]

SUMMARY OF THE INVENTION Therefore, the present invention efficiently realizes a non-linear quantization having a visual characteristic approximated by a logarithmic function by a simple operation, thereby suppressing the deterioration of the image quality and improving the compression ratio of the image data. To provide a compression device,
Its purpose is. Moreover, it aims at providing the image data decompression | restoration apparatus which decompresses the compressed data.

[0007]

According to a first aspect of the present invention, image data representing an intensity level is represented by A · X ⁿ (n is a natural number), and the quantization step is changed as n becomes larger. An image data compression device that quantizes the image data based on this quantization step.

The invention described in claim 2 is the image data compression apparatus according to claim 1, wherein the image data is represented by a floating point representation, and the image data is quantized using the value n of the exponent part.

According to a third aspect of the present invention, there is provided an image data restoration device for restoring the image data by dequantizing the quantized image data.

[0010]

According to the image data compression apparatus of the first aspect, when the image data is represented by A · X ⁿ , there is a linear function relationship between the logarithm of the image data and n. By using the quantization step determined based on the value of n, it is possible to perform quantization that approximates the visual characteristic. That is,
The compression rate can be increased while suppressing the deterioration of image quality.
Representing image data in the form A · X ⁿ is, for example, X =
The case of 2 is easier than the logarithmic operation, as can be seen from the fact that it is expressed as it is in the computer. Also, since the range of n is at most the logarithmic value of the maximum value of the data, the size of the table can be small when creating a table for nonlinear quantization. Therefore, the nonlinear quantization process can be efficiently performed in a short time.

More specifically, for example, a digital image represented by two-dimensional Cartesian coordinates is input, and this digital image is divided into blocks each having vertical and horizontal 8 × 8 pixels, and a discrete cosine transform is applied to each block. (DCT) is performed. DCT value (8
The following quantization is performed on the (× 8 matrix). That is, an intensity quantization table and a spatial frequency quantization table are prepared. The intensity quantization table has a value corresponding to the value of the exponent part of the DCT value. Spatial frequency quantization table is 8 ×
It has a value corresponding to each coordinate of the DCT values of 8 matrices. The value of the intensity quantization table is obtained corresponding to the value of the 8 × 8 matrix. The product of the value of this intensity quantization table and the value of the corresponding spatial frequency quantization table is obtained. This product divides the original value of the 8x8 matrix. Or
The mantissa of the original value is divided by this product. It is possible to further compress the image data by performing the quantization as described above, and then performing the zigzag scan, the run length encoding, and the Huffman encoding, which are well known in JPEG, on the quantized coefficient data.

According to the invention described in claim 2,
In quantization, blocks are divided, and when DCT is performed, for example, floating point arithmetic is performed on DCT coefficients.
The conversion result is also expressed in floating point. These values are stored as different bits in the binary number separately from the mantissa part and the exponent part. As a result, the value of the exponent part can be easily obtained. It is also easy to use the mantissa part. Then, the value quantized in this way is, for example, scanned and expressed as a one-dimensional vector, and further, run length coding and Huffman coding are performed. The exponent part is similarly encoded.

Further, in the image data decompression device according to the invention described in claim 3, the compressed data is dequantized to decompress the image data. For example, Huffman decoding, run length decoding,
By performing inverse quantization and inverse DCT, the image data is expanded and the image is restored.

[0014]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to FIGS. In this embodiment, the present invention is applied to an image data compression device.

In FIG. 1, 11 is an image input unit, 12 is a conversion unit, 13 is a quantization unit, 14 is a scanning unit, and 15 is an encoding unit. Each of these units can be configured using a workstation, for example. The image input unit 11 inputs image data from a magnetic disk, a magneto-optical disk, or the like in which a plurality of image data is written, to the workstation body. Then, this image data is divided into blocks each including a vertical and horizontal 8 × 8 pixel matrix for each frame.

The image data divided into these blocks is
In the conversion unit 12, DCT processing is performed, and 8 × 8 D
CT coefficient data is created. For example RGB or Y
Color image data represented by MCK or the like is separated into three components, that is, luminance data and two color difference data. Luminance data is a signal representing the brightness of an image. The color difference data is a signal representing the chromaticity of the image.
An orthogonal function conversion process such as DCT is performed on the block of luminance data to convert it into a block composed of 8 × 8 coefficient data. Similarly, each block of color difference data is converted into a block composed of 8 × 8 coefficient data.

Then, the quantization unit 13 quantizes each block of the DCT coefficient data. At the time of quantization, the spatial frequency quantization table "spat
ial q table ”and the intensity quantization table“ d ”
energy q and "table". FIG. 5 shows an example of the spatial frequency quantization table, and FIG. 6 shows an example of the intensity quantization table.

For example, as shown in FIG. 2, in the quantizer 13, 6 of a certain input block is calculated based on the equation (1).
From four data {x [i, j]}, the function extrac
The exponent part data e [i, j] is obtained by texp (). The intensity quantization table “density” is calculated from the exponent part data e [i, j]. q "table"
As shown in Expression (2), d [i, j] is obtained, and the spatial frequency quantization table “spatial q tabl
e [], q [i,
j] is required.

Further, as shown in FIG. 2 (b), the equation (1
m) is used to obtain the mantissa part m [i, j] of x [i, j]. Instead of x [i, j] in Expression (3), Expression (3
In m), the mantissa m [i, j] is used.

As described above, the two-dimensional matrix values q [i, j] and e [i, j] obtained by the quantizing unit 13 are zigzag-scanned by the scanning unit 14 to generate a one-dimensional vector value ql [k]. ], El [k], respectively. That is, k is given to each pair [i, j] in the equation (4) of FIG. 3, and ql is given by the equations (5) and (6).
[K] and el [k] are obtained, respectively.

The values of ql [k] and el [k] are run-length coded and Huffman coded in the coding section 15 shown in FIG. Specifically, the run-length encoding unit calculates q2 [l] and e2 from equations (7) and (8).
Data with [l] are obtained. These data consist of runs and levels for each l.

Further, in the Huffman coding section,
As shown in (9) and (10), Huffman code was used.
Encoding is done. As a result, the quantized q [i,
j] is the Huffman coding table “huffman”.
Huffman code q Huffman code for h [l]
Table "huffman Huffman code by "e"
e h [l], respectively.

FIG. 7 is a block diagram showing an image data restoration device according to the present invention. This device, as shown in the figure,
The decoding unit 71 that decodes the encoded image data, the dequantization unit 72 that dequantizes the decoded data, the inverse transformation unit 73 that performs the inverse transformation, and the image restored in this way The image data output unit 74 for outputting

That is, the image data encoded by the compression device is input to the decompression device via the communication system.
The input data is separated into encoded image data and other information such as table information, a frame header, and a scan header, and MCU (Multi) is separated from these data.
apple Component Unit or M
A data block called an inimum coded unit), a Huffman table, a quantization table, a sampling factor, a component identifier, etc. are extracted. Huffman decoding means, according to the Huffman table,
Huffman decoding is performed for each component of the MCU in the encoded image data. Different Huffman tables can be used for different components to be decoded. The Huffman decoding process is performed for each component of the encoded coefficient data in 1MCU, and the decoded coefficient data is generated.

The inverse quantization means inversely quantizes the decoded coefficient data. This inverse quantization process is performed for each component according to the quantization table determined at the time of encoding.

The IDCT means performs IDCT processing on the coefficient data after the inverse quantization processing. In the IDCT method,
The data in each component of 8x8 pixels is restored.

In this way, the decoding processing of the image of one block in the encoded image data based on the JPEG algorithm is performed, and the decoded data of one block in each component is accumulated. Similarly, another block of the encoded image data is decoded, and the decoded data for one screen of each component is generated. This allows the image to be restored.

[0028]

According to the present invention, the linear quantization method utilizing the visual spatial frequency characteristic used in JPEG and MPEG is improved to realize the nonlinear quantization method depending on the visual intensity characteristic. Since it is provided, it is possible to efficiently restore an image with little subjective deterioration.

Further, the present invention uses a value of the exponent part of the intensity and a corresponding intensity quantization table to approximate the visual intensity characteristic, unlike the existing nonlinear quantization method relating to the intensity characteristic. . By using this exponent part, it is possible to perform the quantization which is adapted to the logarithmic visual intensity characteristic, and further, which is adapted to the visual intensity characteristic by using the intensity quantization table. Also, the range of values taken by the exponent part of the intensity is logarithmically smaller than the range taken by the original image, so the size of the intensity quantization table becomes smaller accordingly, contributing to efficient encoding. can do.

[Brief description of drawings]

FIG. 1 is a block diagram showing an image data compression apparatus according to an embodiment of the present invention.

FIG. 2 is a block diagram showing an outline of arithmetic processing in a quantization unit according to an embodiment of the present invention.

FIG. 3 is a block diagram showing an outline of arithmetic processing in a scan unit according to an embodiment of the present invention.

FIG. 4 is a block diagram showing an outline of arithmetic processing in an encoding unit according to an embodiment of the present invention.

FIG. 5 is a diagram showing an example of a spatial frequency quantization table according to an embodiment of the present invention.

FIG. 6 is a diagram showing an example of an intensity quantization table according to an embodiment of the present invention.

FIG. 7 is a block diagram showing a schematic configuration of an image data restoration device according to another embodiment of the present invention.

[Explanation of symbols]

 11 image data input unit 12 conversion unit 13 quantization unit 52 inverse quantization unit 53 inverse conversion unit 54 image data output unit

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI technical display location G06T 5/00

Claims (3)

[Claims] 1. The intensity of image data is represented by A · X ⁿ (n is a natural number), the quantization step is increased as n increases, and the image data is quantized based on this quantization step. An image data compression device characterized by the above. 2. The image data compression apparatus according to claim 1, wherein the image data is represented by a floating point representation, and the image data is quantized based on the value of the exponent part. 3. An image data restoration device, which restores image data by dequantizing the quantized image data.