JPH0822543A - Compressed image transforming device - Google Patents

Abstract

PURPOSE:To provide the device, which subjects compressed image data of a JPEG image or the like to the image processing for enlargement, reduction, or the like at a high speed, with a low cost by controlling the quantizing method so that pseudo filtering processing is performed at the time of expansion and compression in an image data expansion part and an image data compression part. CONSTITUTION:A quantization control part consists of a quantization table setting part 5 to perform the quantization control which weakens the high band signal component in a image dat expansion part 1, and a signal processing part consists of a thinning reduction part 3 to perform the reduction processing of the thinning system. Image data expanded in the image data expansion part 1 is temporarily stored in an image memory 2 as image element data of RGB. The thinning reduction part 3 reads out contents in the image memory 2 at a high speed and writes them in different image elements of the image memory 2 again. A quantization table setting part 5 sets the quantization table values for data expansion in the image data expansion part 1 again so that the filtering processing acts.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a compressed image conversion device for converting a digitized image signal at high speed.

[0002]

2. Description of the Related Art In recent years, signals such as images and sounds have been processed as digital signals. Behind this is progress in semiconductor technology such as LSI, recording technology such as hard disk and optical disk, and transmission technology such as optical fiber. More recently, advances in digital compression algorithms for signals themselves have also played an important role.

Taking the case of an image signal as an example, MPEG (M
oving Picture Coding Experts Group) and JPEG (Joint
A compression method combining discrete cosine transform and entropy optimum coding, such as Photographic Coding Experts Group), is normally used, and a signal can be compressed to several tens of minutes.

By compressing signals in this way, the speed of storage, transmission, etc. is increased, and new application areas such as image databases and videophones are expanding.

On the other hand, the compressed signal is difficult to be subjected to various processes unless it is expanded and decoded. Taking image signals as an example,
Most of the main processing such as color conversion, size change, and pixel ratio conversion is performed after decompression. At the research level, there are attempts to perform signal processing while compressed, but they have not been put to practical use. In many cases, once decompressed, the signal is not recompressed and as a result is only used for output purposes.

Since conventional AV equipment has not processed the signal data so much, it may be considered as a transition to a simple digital system. However, in the multimedia era where the computer technology is further integrated in the future, the processing of the compressed signal is performed. Is expected to become a major issue.

[0007]

In order to simplify the description of the problem of the compressed image conversion apparatus for converting a digitized image signal at high speed, processing of a still image will be described here.

Still images are a field in which a wide range of applications such as image databases can be expected. Now, as an example, a case where photographic film images are stored in a database will be described. For example, in the case of 35mm film, the maximum resolution is said to be about 3000 x 2000 pixels, and assuming high-quality use such as print output, it is naturally necessary to store data at the maximum resolution. high. On the other hand, when searching and displaying image data using a computer as a terminal, the display limit is approximately 1000 × 1000 pixels, and even when a high-definition monitor or the like is used, it is approximately 2000 × 1000 pixels. The performance-to-price ratio of terminals and printing devices that display these compressed images improves year by year, but it is difficult to change the stored image data according to the performance of the output device.
Therefore, it is necessary to consider the resolution of the image data stored and managed separately from the performance of the display and output devices.

Specifically, in the image database, it is desirable to input and manage the image data at the highest resolution required for output, perform processing according to the characteristics such as resolution of the output device at the time of output, and output. Be done.

This is consistent with the centralized management of data, which is an important concept of a database. Specific processing includes reduction, color conversion, and pixel ratio conversion (square pixel conversion, etc.). In reality, a large-capacity disk is required to store image data in high resolution, and it takes a long time to transfer a large amount of data to the display memory. Is an indispensable technology.

For still images, a compression method called JPEG is becoming the standard, and LSIs for performing compression and decompression are available at low cost, so that the recent spread of image files and databases is remarkable.

Therefore, in order to perform the ideal image management as described above, a technique capable of easily realizing the processing of the compressed image is required. Taking the reduction of a compressed image as an example, in the conventional technique, in order to reduce the image data that has been compressed once, the image data that has been decompressed once is reduced by performing averaging processing over time by a normal computer operation. Or, it was necessary to speed up the averaging process with dedicated large-scale hardware.

In reality, as a means for increasing the speed by using hardware of an appropriate scale, there is a method called skipping and reading data. For example, when a rectangular image is reduced to ½ in height and width, skipping is performed by skipping every other row in the vertical direction first, and then skipping by reading every other row in the horizontal direction. This is a method that can be implemented by software. This method is not limited to 1/2, but can be applied to the diagonal components in a staircase pattern.
There is a big problem regarding the quality of the generated image, such as (jaggies) or a gradation difference in the part where the gradation changes smoothly called gradation.
At present, it is not always widely used.

Although the above is an example at the time of reduction, even at the time of enlargement, speeding up can be achieved by a method of repeatedly reading the same data instead of skipping reading. As with the reduction, there was a problem in terms of image quality.

The present invention solves the above problems, and a compressed image conversion apparatus which realizes image processing such as enlargement / reduction of image data and data compression with a simple structure without significantly impairing the image quality. For the purpose of providing.

[0016]

In order to solve the above problems and achieve the object, the present invention provides an image data decompression unit for decompressing compressed image data at high speed, and an image decompressed by the image data decompression unit. An image memory that temporarily stores data, a signal processing unit that reads image data from the image memory, performs signal processing and stores the signal in the image memory, and an image data compression unit that reads image data from the image memory and performs data compression. The image data decompression unit and the image data compression unit further include a quantization control unit that controls a quantization method so that a pseudo filtering process is performed when the decompression and compression processes are performed.

According to the first embodiment of the present invention, the quantization control unit is composed of a quantization table setting unit, which performs quantization control for weakening the high frequency signal component in the image data expansion unit, The signal processing unit is composed of a thinning and reducing unit, and performs thinning-out reduction processing.

According to a second embodiment of the present invention, the quantization control section is composed of a quantization table setting section, and performs quantization control for weakening a high frequency signal component in the image data compression section, The signal processing unit is composed of a repetitive enlarging unit and performs a repetitive enlarging process of a data repeating method.

[0019]

According to the present invention, the function of realizing pseudo filtering by appropriately controlling the quantization method in the frequency domain of the compression device and the simple high-speed image processing function of the algorithm such as the conventional decimation and repetition method are provided. By combining them, the image quality can be greatly improved even with a simple image processing algorithm.

That is, when the image data is reduced, the compressed image is decompressed and stored in the image memory. In the decompression process, the quantization method is controlled so as to reduce the high frequency components unlike the compression. As a result, the high frequency components of the restored image itself are weakened, resulting in a blurred image. Even if the blurred image is subjected to the thinning-out process, the high-frequency component is small, and therefore the image quality is not significantly deteriorated.

On the other hand, when the image data is enlarged, the image data is decompressed as usual and stored in the image memory.
Subsequently, enlargement by the iterative method is performed as in the conventional case. In this state, the image quality is low in the above-mentioned meaning. Then, re-compression is performed. At this time, the quality of the compressed image becomes blurry, that is, smooth by reducing the amount of high-frequency components used for quantization as in the case of the reduction. .

Besides, it is possible to combine with various image processings described in the embodiments.

[0023]

1 is a block diagram showing the configuration of a compressed image conversion apparatus according to a first embodiment of the present invention. This is an example of converting a compressed image into a compressed image that has been reduced to 1/2.

In FIG. 1, 1 is an image data decompression unit for decompressing JPEG compressed image data 6 and 2 is an image for temporarily storing the image data decompressed by the image data decompression unit 1 as RGB pixel data. The memory 3 reads out the contents of the image memory 2 at a high speed, and again a thinning reduction unit as a signal processing unit for writing to different pixels of the image memory, and 4 a portion of the image memory 2 reduced by the thinning reduction unit 3 An image data compression unit 5 for compressing image data is a quantization table setting unit as a quantization control unit for resetting the quantization table value at the time of data expansion in the image data expansion unit 1 so that the filtering process works.

Before explaining the operation of this embodiment, JPEG compression will be briefly described as basic knowledge.

Two-dimensional DCT (discrete cosine transform) is used in JPEG compression, and the natural image is converted into frequency domain data which is easier to compress than the spatial domain. Since the two-dimensional DCT calculation requires a large amount of calculation, it is spatially divided into blocks of about 8 × 8 pixels and then applied to each block. When this 8 × 8 pixel block is used, a total of 64 data are obtained.

Each of the obtained data is expressed as a coefficient of the inverse DCT at the time of inverse transformation (at the time of expansion), and indicates that a large number of frequency components having large coefficient values are included (for example, Rafael C. Digital Ima by Gonzalez / Paul Wintz
ge Processing p.121).

The compression is carried out by reducing the allocation of the number of bits when quantizing this coefficient. A parameter that controls this reduction method is called a quantization parameter. Usually, the allocation is increased for low-frequency components and the allocation is reduced for high-frequency components.

According to the present invention, by changing the setting of the quantization parameter, not only simple compression but also equivalent filtering is performed. For example,
When using a quantization table that externally specifies the number of bits assigned to each coefficient as a quantization parameter,
A table corresponding to 64 pieces of data per block is created. If the number of allocated bits for the data corresponding to the high frequency component in this table is forcibly set to 0, the high frequency component is removed. However, this processing increases distortion between blocks, so an extreme value such as 0 is not optimal.

Not only is the number of bits specified as a quantization parameter, but the quantized component is, for example, a fraction.
If it is possible to specify a process that reduces the value of, spatial low-pass filtering can be realized without increasing block distortion.

The above is the main points of this embodiment, and the operation will be described with reference to the drawings.

Now, as the JPEG compressed image data 6, 25
It is assumed that a compressed image of 6 pixels × 256 pixels is input to the image data decompression unit 1. With 256 pixels x 256 pixels, 32 blocks of 8 x 8 pixels are arranged vertically and horizontally, so
It consists of a total of 1024 blocks. The compressed image is obtained by quantizing the DCT result for each of these blocks, performing zigzag scanning, and Huffman coding (entropy coding) the result. Since the Huffman coding is not directly related to the present invention, its explanation is omitted here. Quantized D corresponding to each block of 1024
The CT coefficient is inverse DCT transformed by the image data decompression unit 1.
It is (decompressed), generated as an image image of 256 pixels × 256 pixels, transferred to the image memory 2, and stored.

As the quantization table used for this quantization, the table used for compression is used in normal decompression, but in the present embodiment, the high frequency component is removed by the quantization table setting unit 5 as described above. , That is, the allocated bit number 0 is used. With this setting, the decompressed image image becomes blurred.

Subsequently, the image image in the image memory 2 is reduced to 1/2 by the thinning reduction unit 3. This 1/2 reduction is realized by reading and rewriting from the image memory by the following algorithm.

(1) For the original data (256 pixels × 256 pixels), only the odd rows of the original data are sequentially read from the first row,
Writing is performed on the original data in order from the position of the first line of the original data. As a result, the size is reduced by half in the vertical direction.

(2) With respect to the reduced data (256 pixels × 128 pixels) of the above (1), only the odd columns of the reduced data are sequentially read from the first column, and the reduced data are sequentially read from the position of the first column. Write in. As a result, the size is reduced by half in the horizontal direction.

This algorithm is very simple and can be easily accelerated by hardware. However, it has not been used so far due to the image quality defect such as the above-mentioned jaggedness, but the present invention makes it difficult to cause jaggedness because the image image is blurred beforehand when the data is expanded.

The image data that has been reduced to ½ in the vertical and horizontal directions in this way is then read out by the image data compression section 4 and subjected to ordinary DCT conversion, quantization and Huffman coding, and processed JPEG compressed image data. 7 is obtained.

FIG. 2 is a block diagram showing the arrangement of a compressed image conversion apparatus according to the second embodiment of the present invention, which is an example in which a compressed image is doubled.

In FIG. 2, 11 is an image data decompression unit for decompressing JPEG compression image data 6, and 12 is an image for temporarily storing the image data decompressed by the image data decompression unit 11 as RGB pixel data. A memory, 13 is a repetitive enlargement unit as a signal processing unit that reads out the contents of the image memory 12 at high speed and writes again to different pixels of the image memory, and 14 is a portion of the image memory 12 enlarged by the repetitive enlargement unit 13. The image data compression unit for compressing the image data of 15 is a quantization table setting unit as a quantization control unit for setting the quantization table value at the time of data compression in the image data compression unit 14 so that the filtering process works. .

Next, the operation of the second embodiment will be described with reference to the drawings.

Now, the JPEG compressed image data 6 is 25
It is assumed that a compressed image of 6 pixels × 256 pixels is input to the image data decompression unit 11. With 256 pixels x 256 pixels, 32 blocks of 8 x 8 pixels are arranged vertically and horizontally, so
It consists of a total of 1024 blocks. The compressed image is obtained by quantizing the DCT result for each of these blocks, performing zigzag scanning, and Huffman coding (entropy coding) the result. Since the Huffman coding is not directly related to the present invention, its explanation is omitted here. Quantized D corresponding to each block of 1024
The CT coefficient is inverse DCT transformed by the image data decompression unit 11.
It is (decompressed), generated as an image image of 256 pixels × 256 pixels, transferred to the image memory 12, and stored.

The quantization table used for this quantization is a normal expansion setting, and the original image is restored. Subsequently, the image image in the image memory is doubled by the repetitive enlarging unit 13. The enlargement is realized by reading and rewriting from the image memory by the following algorithm.

(1) For the original data (256 pixels × 256 pixels), all rows of the original data are sequentially read from the last row, and the same data is written in a row that is a multiple of the read row and the row before that. This doubles vertically.

(2) With respect to the expanded data (256 pixels × 512 pixels) of the above (1), all columns of the expanded data are sequentially read from the final column, and are the same as the double column of the read column and the column before it. Write the data. As a result, it is enlarged twice in the lateral direction.

This algorithm is very simple as in the case of the reduction, and the speed can be easily increased by hardware, but on the other hand, there are image quality defects such as the above-mentioned jaggedness. Therefore, this image quality is improved by the image data compression unit 14.

Specifically, the image data that has been doubled vertically and horizontally is read by the image data compression unit 14 and DCT-converted. In the quantization table used at this time, the quantization table setting unit 15 sets the bit allocation to 0 so that high frequency components do not exist. Since the DCT is an orthogonal transform, this operation does not affect other components, but it can introduce distortion in the image.

As described above, by carefully devising the quantization method at the time of compression / expansion as seen in the first and second embodiments, even if the original image quality is greatly deteriorated, such as thinning reduction and repeated enlargement. It is possible to reduce the influence on the quality of the compressed image.

The present invention can also be used as an applied embodiment to improve the image quality when an image is displayed on an interlaced scan monitor. This is achieved by blurring the image in the vertical direction in order to prevent flicker that is characteristic of interlaced monitors such as televisions. In other words, it is effective even when enlargement or reduction is not performed.

It can be used not only for blurring but also for the purpose of emphasizing high frequency components. For example, if the low frequency component is dropped as the processing before the contour extraction, the subsequent processing becomes easy.

In the embodiment, the simple example of the quantization control in which the bit allocation of the high frequency component of the quantization table is set to 0 has been shown, but the filter function for controlling the DCT coefficient should be specified externally. If the control is performed, the distortion caused by the quantization control can be significantly reduced.

In this embodiment, JPEG is used as an example, but the present invention can be similarly applied to compression using another DCT system such as MPEG or another orthogonal transform system.

[0053]

As described above, according to the present invention, the JPE
The image compression / expansion function that combines a space and frequency conversion system and quantization for a two-dimensional image such as G is combined with a simple image processing unit with an algorithm such as thinning-out reduction and repeated enlargement, and a quantum at the time of compression / decompression is combined. By appropriately controlling the conversion, a compressed image can be processed with a simple configuration and good image quality can be obtained.

[Brief description of drawings]

FIG. 1 is a block diagram showing a configuration of a compressed image conversion apparatus according to a first embodiment of the present invention.

FIG. 2 is a block diagram showing a configuration of a compressed image conversion device according to a second embodiment of the present invention.

[Explanation of symbols]

1, 11 ... Image data expansion unit, 2, 12 ... Image memory,
3 ... thinning reduction unit, 4, 14 ... image data compression unit,
5, 15 ... Quantization table setting unit, 6 ... JPEG compressed image data, 7 ... Processed JPEG compressed image data,
13 ... Repeated expansion section.

─────────────────────────────────────────────────── ─── Continuation of the front page (51) Int.Cl. ⁶ Identification code Internal reference number FI Technical display location H04N 1/393 1/41 Z 7/24 H04N 7/13 Z

Claims (3)

[Claims] 1. An image data decompression unit for decompressing compressed image data at high speed, an image memory for temporarily storing the image data decompressed by the image data decompression unit, and image data read from the image memory. A signal processing unit that performs signal processing and stores it in an image memory, an image data compression unit that reads image data from the image memory and performs data compression, and an image data decompression unit and when performing decompression and compression processing in the image data compression unit. A compressed image conversion apparatus, comprising: a quantization control unit that controls a quantization method so that a pseudo filtering process is performed. 2. The quantization control unit includes a quantization table setting unit, performs quantization control for weakening a high frequency signal component in the image data expansion unit, and the signal processing unit includes a thinning reduction unit. 2. The compressed image conversion apparatus according to claim 1, wherein the compressed image conversion apparatus performs thinning-out reduction processing. 3. The quantization control unit includes a quantization table setting unit, performs quantization control for weakening a high frequency signal component in the image data compression unit, and the signal processing unit includes a repetitive expansion unit. The compressed image conversion apparatus according to claim 1, wherein the compressed image conversion apparatus performs a data repetition type enlargement process.