JPH10145783A - Image compression / expansion method and apparatus - Google Patents

Abstract

(57) [Problem] To provide an image compression / expansion method and apparatus for compressing / expanding a digital image by hierarchizing the image so that the image can be reproduced according to the performance on the reproduction side. SOLUTION: An input image is converted into a plurality of components by a resolution conversion analysis unit 11 and quantized, and then the hierarchical encoding unit 1
In step 3, it is classified by quality. The hierarchized data is transmitted by the transmission cable 30 and decoded by the hierarchal decoding unit 21. After the decoded data is subjected to coefficient reproduction, the resolution conversion / combination unit 23 combines the components to restore the image.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for compressing and expanding digital images.

[0002]

2. Description of the Related Art Digital images do not suffer from image quality deterioration during transmission and storage and are easy to process.
It is used in many fields, including medicine, broadcasting, and printing. However, the digital image has a large amount of data, and thus has a problem that a large amount of storage means is required even when transmission takes a long time or is stored. For this reason, various image compression techniques have been developed today, and it has become possible to process and transmit an image in a short time and to store the image in a storage device having a small capacity.

[0003]

At present, in a field where still images are frequently used, an image compression method having a high reproduction quality is required, and the searchability and reusability of the images stored in the database are high. It has been demanded. However, conventionally,
A compression method with a high compression rate has good searchability but poor reusability. A compression method with a low compression rate has good reusability but poor searchability. SUMMARY OF THE INVENTION An object of the present invention is to provide a method and apparatus for layering, compressing and decompressing images so that the images can be reproduced according to the performance of the reproduction side, in order to solve the above-mentioned problems.

[0004]

The object of the present invention is to convert the resolution of an image stepwise to obtain components having different numbers of pixels.
After the resolution-converted image components are hierarchically encoded to compress the image, and the compressed data is hierarchically decoded, the decoded image components having different resolutions are decoded. This is solved by synthesizing and expanding the image.

[0005]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described below.
FIG. 1 is a block diagram of an embodiment of an image compression / decompression device according to the present invention. The image compression / decompression device shown in FIG. 1 includes an image compression device 10 and an image decompression device 20.
More specifically, the image compression device 10 includes the resolution conversion analysis unit 1
1, a quantization symbol generation unit 12 and a hierarchical encoding unit 13, and the image decompression device 20 includes a hierarchical decoding unit 2
1, a quantization coefficient reproduction unit 22 and a resolution conversion synthesis unit 23. Further, the hierarchical encoding unit 12 of the image compression device 10 and the hierarchical decoding unit 21 of the image decompression device 20 are connected by a transmission cable 30 for transmitting digital data. The original image is converted by the resolution conversion analysis unit 11, encoded by the hierarchical encoding unit 13, and transmitted via the transmission cable 30. After the transmitted data is decoded by the hierarchical decoding unit 21, the resolution is converted by the resolution conversion / combination unit 23, and a reproduced image is obtained. In FIG. 1, the image compression device 10 and the image expansion device 20 are connected by the transmission cable 30, but the image compression device 10 and the image expansion device 20 may not be connected. When not connected, the image data compressed by the image compression device 10 may be stored in a storage device, and may be read by the image expansion device 20 at the time of expansion.

FIG. 2 is a block diagram showing details of the resolution conversion analysis unit 11. As shown in FIG. 2, the resolution conversion analysis unit 11 has a configuration including four high-pass filters and four low-pass filters, and eight thinning units. The high-pass filter passes only the components above the set frequency, and the low-pass filter passes only the components below the set frequency. The thinning unit has the function of thinning out both vertically and horizontally by half. I have.

FIG. 3 is a block diagram showing details of the hierarchical encoding unit 13. As shown in FIG. 3, the hierarchical encoding unit 1
Reference numeral 3 denotes a configuration including four run length encoding units.
Run length coding is an effective coding method when a certain value often appears continuously. The run length encoding unit of the present apparatus performs the run length encoding especially only for 0 (zero). A threshold T is set for each run length encoding unit,
It has a function to regard a value less than T as 0 for the input sequence, and furthermore, this makes it possible to use the code 0 to T-1 as a zero run length code.
To K-1. The run length is T
If it is larger, it is divided into multiple parts.

FIG. 4 is a block diagram showing details of the hierarchical decoding unit 21. As shown in FIG. 4, the hierarchical decoding unit 2
Reference numeral 1 denotes a configuration including four run length decoding units.
Run length decoding is to restore run length encoded data.

FIG. 5 is a block diagram showing details of the resolution conversion / combination unit 23. As shown in FIG. 5, the resolution conversion / combination unit 23 has a configuration including eight pixel restoration units and four frequency component combination units. The pixel restoration unit inserts zero-valued pixels in the vertical and horizontal directions to double the number of pixels in both the vertical and horizontal directions, ie, quadruple the number of pixels as a whole. That is,
It has a function opposite to that of the thinning unit of the resolution conversion analysis unit 11. Further, the pixel restoration unit includes a low-pass filter or a high-pass filter, and performs processing on a component whose number of pixels is doubled. In FIG. 5, pixel restoration units 230 and 23 are shown.
3, 236 and 239 have low-pass filters,
Each of the pixel restoration units 231, 234, 237, and 240 has a high-pass filter. The frequency component synthesizing section has a function of synthesizing the high frequency component and the low frequency component, and adds the outputs of the low pass filter and the high pass filter.

FIG. 6 shows a hardware configuration of the present apparatus.
The image compression device 10 and the image decompression device 20 have the same hardware configuration, and a signal processing device (DSP: digital signal processor) via a data bus / address bus 41.
42, I / O buffer memory 43, operation buffer memory 4
4, a program memory 45, an input port 46, and an output port 47 are connected. The difference between the image compression device 10 and the image decompression device 20 is only the software stored in the program memory.

Next, the processing outline of the image compression apparatus 10 is shown in FIG.
This will be described with reference to the flowchart of FIG. Image compression device 10
Then, the image data input from the input port 46 is temporarily
It is stored in the / O buffer memory 43. (S1)

Thereafter, color space conversion of the input image data is performed. The input image is represented by an RGB color system or the like. However, since the visibility is sensitive to luminance and insensitive to color difference, and since the RGB values are highly correlated, YUV is used for efficient image compression. Convert to color system. Y, U, and V represent a luminance component, a blue component, and a red component, respectively.
The conversion from the RGB color system to the YUV color system is performed by a specific conversion formula, which is a known technique and will not be described in detail here. Not only YUV but also La * b *, Lu *
Even if a color system such as v * or HSV is used, image compression can be performed more efficiently than the RGB color system. This color space conversion is not required for non-color images. (S2)

Next, the resolution conversion analysis unit 11 performs multi-resolution conversion of the image to reduce the number of pixels. For resolution conversion of an image, spatial filtering is necessary because the aliasing noise of the signal is mixed in only by simple thinning. The spatial filtering process is performed by a high-pass filter and a low-pass filter for analysis of the resolution conversion analyzer 11 shown in FIG. The input image G1 subjected to the color space conversion is input to the high-pass filter 110 and the low-pass filter 111. One of the high-frequency components is extracted by the high-pass filter 110 and then processed by the thinning unit 112 to be output as the component A. The other component is extracted by the low-pass filter 111 and then processed by the thinning unit 113 to be output as the component A ′. The component A ′ is further input to the high-pass filter 114 and the low-pass filter 115. The processing is performed in this manner, and the thinning unit 11
2, component A from the thinning unit 116, component C from the thinning unit 117, component D from the thinning unit 118,
The component E is output from the thinning unit 119. In the thinning-out section, since both the vertical and horizontal sides are thinned out to １ ／, the total number of pixels is １ ／. That is, among the five components output from the resolution conversion analysis unit 11, the components D and E are input image G1.
Becomes 1/256 of the number of pixels. This resolution conversion analysis unit 1
The resolution conversion by 1 is generally called a wavelet transform and is well known. (S3)

Next, the quantization symbol generation unit 12
ＥE are quantized. Quantization is the operation of rounding a floating-point or fixed-point real number to an integer. The pixel data originally represented by an integer by the resolution conversion by the resolution conversion analysis unit 11 becomes a real number. Real numbers require more memory than integers, so round to integers. There are two types of quantization depending on the characteristics of the pixel components. The components A to D are high-frequency coefficients, and since the distribution is biased near 0, nonlinear quantization is performed to finely quantize the small amplitude part and coarsely quantize the large amplitude part. As a result, a quantization error can be suppressed smaller than in linear quantization. Here, it is rounded to a signed integer of 8 bits in total, that is, a 7-bit amplitude code and 1 bit. Since the component E is a low-frequency coefficient including direct current,
Perform linear quantization. Here, it is rounded to an 8-bit unsigned integer. (S4)

Next, the component A
.. E are hierarchically run-length encoded to perform data compression. Since the component E is a low-pass coefficient, it is distributed almost uniformly. Since the component E is a reduced component of the input image by itself, it can be used as it is for thumbnail display or the like. Therefore, no processing is performed on the component E here, and the component A
To D. The components A to D are high-frequency coefficients, and the distribution is biased to around 0. Therefore, if the coefficients with small amplitudes are reduced, significant information compression can be expected. Therefore, a value less than the specified threshold value is regarded as 0, and zero run length encoding is performed. The zero run length coding is performed by each run length coding unit in the hierarchical coding unit 13 shown in FIG. In the example of FIG. 3, since there are four run length encoding units, the hierarchy is divided into four layers and zero run length encoding is performed. (S5)

Next, an output data aligning unit (not shown) rearranges the data in the output order so that the compressed data can be efficiently expanded on the expansion side, and then outputs the data. Up to this point, the input image G1 has been converted to the resolution conversion analysis unit 11
Are hierarchized in the spatial domain (resolution), are hierarchized in the quality domain by the hierarchical encoding unit 13, and the components A to D are converted into a total of 17 data of four layers and a component E, respectively. There are two criteria for data alignment: a case mainly based on quality hierarchy as shown in FIG. 8 (a) and a case mainly based on spatial hierarchy as shown in FIG. 8 (b). When the quality hierarchy is mainly used, control information is transmitted first as shown in FIG. The control information is, for example, the number of pixels of an image, color / monochrome distinction, a threshold value used as a layer break, and the like. Following the control information, the component E is sent. Subsequently, those having the first layer quality, such as the first layer component D, the first layer component C, the first layer component B, and the first layer component A, are transmitted in the order of smaller number of pixels. Furthermore, it is transmitted in the order of the second layer, the third layer, and the fourth layer. In the case where the spatial layering is mainly used, FIG.
As shown in (b), the control information is transmitted first and the component E is transmitted next, as in the case where the quality layering is mainly performed. Hierarchical component D,
As in the case of the third hierarchical component D and the fourth hierarchical component D, the components D are transmitted in descending order of quality. Subsequently, the components are sent in the order of component C, component B, and component A. (S6)

Next, the processing outline of the image decompression device 20 will be described with reference to the flowchart of FIG. Image expansion device 2
0, the compressed data compressed by the image compression apparatus 10 is input from the input port 46 and is temporarily stored in the I / O buffer memory 43.
To be stored. (S7)

Next, decoding of the encoded compressed data is performed by the hierarchical decoding unit 21 shown in FIG. this is,
The operation is completely the reverse of the operation of the hierarchical encoding unit 13 in step S5 in the flowchart of FIG.
The one encoded by 3d is a run length decoding unit 21d, the one encoded by the run length encoding unit 13c is a run length decoding unit 21c, and the one encoded by the run length encoding unit 13b is The run-length decoding section 21b processes the data encoded by the run-length decoding section 13a, respectively. (S8)

Next, the coefficients quantized by the quantized coefficient reproducing section 22 are reproduced. (S9)

Next, in the resolution conversion / synthesis unit 23, the inverse conversion of the multiple resolution conversion performed by the resolution conversion analysis unit 11 in step S3 is performed to increase the number of pixels. More specifically, in the resolution conversion / combination unit 23 shown in FIG. 5, first, the pixel restoration unit 230 inserts 0-valued pixels in the vertical and horizontal directions of the pixel of the component E to increase the number of pixels to four times as a whole. After that, it is processed by a low-pass filter. Similarly, the pixel of the component D is quadrupled as a whole by the pixel restoring unit 231 and then processed by a high-pass filter. The restored component D and component E are combined by the frequency component combining unit 232 to become the component C ′. Further, after the component C ′ is processed by the pixel restoring unit 233, the component C processed by the pixel restoring unit 234 and the frequency component synthesizing unit 235 are combined to become the component B ′. Similarly, the component B is processed by the pixel restoring unit 237 and then synthesized by the frequency component synthesizing unit 238 to become the component A ′. The component A is processed by the pixel restoring unit 240 and then synthesized by the frequency component synthesizing unit 241. Thus, the image G1 is restored. The resolution conversion by the resolution conversion synthesizing unit 23 is generally called an inverse wavelet transform and is known. In addition, components A ′, B ′,
Since C ′ and E are the reduced components of the image G1 by themselves, the images G2, G3, G
4. Available as G5. (S10)

Next, an image G represented by the YUV color system
1 to G5 are converted to an RGB color system. This is the reverse operation of step S2, and the equations for the conversion are well known and will not be described in detail here. This inverse color space conversion is also performed in step S2.
This is not necessary if the image is not a color image. (S11)

Finally, the images G1 to G5 are output to the display device. The output is not performed collectively, but in step S10.
Are performed in order from the one processed in. In this case, the images G5 having the smaller number of pixels are displayed in order. (S12)

Here, the hierarchical encoding in step S5 and the step S5
8 will be described using specific numerical values.
Here, the run length encoding unit 13 of the hierarchical encoding unit 13
a, A case where only the two run length coding units of the run length coding unit 13b are used, and the threshold T of the run length coding unit 13a is set to 8 and the threshold T of the run length coding unit 13b is set to 4 I'll think about it. When a sequence a as shown in FIG. 10A is input to the run length encoding unit 13a, the number [8, 11, 10, 8, 9] with the threshold T = 8 or more is left as it is, and the number less than 8 is left. Is encoded to that (the number of consecutive times -1). As a result, a first-order encoded sequence b as shown in FIG. 10B is obtained. A number less than 8 is a first-order coded residual number sequence c as shown in FIG. The primary encoded sequence b is output as it is as the data of the first hierarchy. The first-order coded residual sequence c is input to the run length coding unit 13b, coded with the threshold value T = 4, and converted into the second-order coded sequence d as shown in FIG. The second-order coded residual number sequence e as shown in e) is obtained. Here, the run length encoding unit 13
c, since the run length coding unit 13d is not used, the second-order coded sequence d and the second-order coded residual sequence e are collectively output as data of the second hierarchy.

The run length decoding unit 2 is also used in the hierarchical decoding unit 21.
1a, since only two run length decoding units of the run length decoding unit 21b are used, the second-order coded sequence d and the second-order coded residual sequence e, which are data of the second hierarchy, are run-length decoded. Transformation part 2
1b. In the run length decoding unit 21b, the secondary encoded sequence d is decoded at the threshold T = 4, and a decoded sequence f as shown in FIG. 10 (f) is obtained. Further, the first coded residual number sequence c is restored from the second coded residual number sequence e and the decoded number sequence f. Subsequently, the first encoded sequence b,
The next coded residual number sequence c is input to the run length decoding unit 21a. In the run length decoding unit 21a, the primary encoded sequence b is decoded at the threshold T = 8, and a decoded sequence g as shown in FIG. 10 (g) is obtained. Finally, the first coded residual number sequence c
The decoding sequence g is restored.

Although the present invention has been described using the preferred embodiments, the present invention is not limited to the above embodiments. For example, the number of high-pass filters and low-pass filters in the resolution converter, the number of run-length encoders in the hierarchical encoder, and the like can be changed as appropriate.

[0026]

As described above, according to the present invention, the resolution of an image is converted stepwise to obtain components having different numbers of pixels, and then the resolution-converted image components are hierarchized. Image data is compressed by compressing the data, and the compressed data is hierarchically decoded, and then the image data is expanded by combining decoded image components having different resolutions. Therefore, it is possible to select reproduction of a plurality of images having different resolutions on the decompression side.
For this reason, on the decompression side, a low-resolution image that does not take a long time to transmit can be reproduced and its contents can be checked immediately, and a short time can be used to reproduce a vivid high-resolution image.

[Brief description of the drawings]

FIG. 1 is a block diagram of an embodiment of an image compression / decompression device according to the present invention.

FIG. 2 is a block diagram showing details of a resolution conversion analysis unit;

FIG. 3 is a block diagram showing details of a hierarchical encoding unit.

FIG. 4 is a block diagram showing details of a hierarchical decoding unit.

FIG. 5 is a block diagram illustrating details of a resolution conversion / synthesis unit;

FIG. 6 is a block diagram showing a hardware configuration of an image compression / decompression device according to the present invention.

FIG. 7 is a flowchart showing the processing operation of the image compression device of the present invention.

FIG. 8 is an explanatory diagram showing the order of transmission during data transmission.

FIG. 9 is a flowchart showing the processing operation of the image decompression device of the present invention.

FIG. 10 is an explanatory diagram of data processed by a hierarchical encoding unit and a hierarchical decoding unit;

[Explanation of symbols]

 11 Resolution Conversion Analysis Unit 13 Hierarchical Encoding Unit 21 Hierarchical Decoding Unit 23 Resolution Conversion Synthesis Unit

Claims (4)

[Claims] 1. A method comprising the steps of: converting a resolution of an image stepwise to obtain components of different numbers of pixels; and hierarchically encoding each of the resolution-converted image components. Image compression method 2. A resolution conversion analysis unit for converting a resolution of an image stepwise to obtain components of different numbers of pixels, and a hierarchical encoding for hierarchically encoding each of the resolution-converted image components. Image compression apparatus characterized by comprising a section 3. An image decompression method comprising the steps of hierarchically decoding encoded data and synthesizing decoded image components having different resolutions. 4. A method according to claim 1, further comprising a hierarchical decoding unit that performs hierarchical decoding on the encoded data, and a resolution conversion combining unit that combines decoded image components having different resolutions. Image expansion device