JPH066829A - Method and apparatus for compressing image data - Google Patents

Abstract

PURPOSE: To reduce the effect of image data compression on an image reconstructed while achieving the degree of desired compression by changing a luminance component compression factor, a color component compression factor and the ratio of both of them. CONSTITUTION: A luminance component data input 60 is supplied to a decorrelator 12' and a color component data input 62 is supplied to a decorrelator 12", and compression is individually performed to the input data of the luminance component and the color component. A combined-bit counter 72 detects the quantities of the compressed data in a luminance component compressed data output-terminal 68 and a color component compressed data output-terminal 70. Then, a feedback signal 74 showing the sum value of these two quantities is generated, and this signal 74 is transmitted to a quantization controller 76. The controller 76 individually changes the degree of a luminance component quantization performed by a luminance component quantizer 64 and the degree of a color component quantization performed by a color component quantizer 66. Thus, the combined data rate of the compressed data output of the luminance component and the color component becomes equal to or approximate to a desired value.

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 image data.

[0002]

Quantizers are used in known compression devices to selectively remove information from the luminance and color components of input image data as part of the compression process. The higher the degree of quantization, the higher the degree of compression. Generally speaking, the compatibility of information content and compression changes depending on the part of the input data, so it is necessary to decide in advance how much quantization should be set and how much compression should be performed. Is impossible.

For some systems this is not a problem. For example, in the image data compression device proposed by the Joint Photographic Experts Conference currently under consideration by the International Standards Committee, the input image data is used for a specific image and using a quantizer and an entropy encoder. It is compressed to a certain limit, which changes depending on the parameter being set.
The compressed image data was processed and displayed in a non-real time manner, so that varying degrees of compression did not pose a particular problem.

In contrast to the above cases, the problem of varying the degree of compression achieved is when the input image data is compressed to fill a fixed size data block or when the compressed data has similar size constraints. It becomes serious when you obey. It is necessary to store each block of compressed image data in a magnetic track with a fixed maximum capacity, or use a compressed data block in a real-time transmission system while processing each block of data a fixed maximum number of times. Therefore, the maximum size of compressed data is decided in advance. In such systems, there is no flexibility in the maximum size of a block that can fit in compressed image data. If you compress the data too much to fit into a fixed size block, the data will overflow and you are unavoidable.

On the other hand, it is desirable not to compress the input image data more than necessary. This is to suppress deterioration of an image reconstructed from image data compressed in the data compression process as much as possible.

[0006] In the data compression system proposed in British patent application 9119985.1, the degree of compression performed by the data compression system is controlled on a block-by-block basis by dynamic quantization control.

[0007]

In image data compression systems, it has always been a challenge to reduce the effect of image data compression on the reconstructed image while achieving the desired degree of compression.

It is therefore an object of the present invention to reduce the effect of image data compression on a reconstructed image while achieving the desired degree of compression.

[0009]

According to the present invention, there is provided a data compression apparatus for compressing the luminance component and color component of input image data to create compressed image data having an overall (total) compression coefficient. Provided. The data compression apparatus includes a brightness component compression unit for compressing a brightness component of input image data by a brightness component compression coefficient, and a color component compression unit for compressing a color component of the input image data by a color component compression coefficient, Feedback means for supplying a feedback signal according to the overall compression coefficient, coupled with the luminance component compression means and the color component compression means, and responsive to the feedback signal, the luminance component compression coefficient, the color component compression coefficient,
And a control means for controlling the overall compression coefficient by changing the ratio between the luminance component compression coefficient and the color component compression coefficient.

It is to be noted in the present invention that the effect of changes in the compression of the luminance component of the image data on an individual may be different from the effects of the same change on the compression of the color component of the image data on the individual. This fact is used in the present invention as follows. That is, in order to change the overall compression coefficient of the image data compression apparatus, the luminance component compression coefficient, the color component compression coefficient, and the ratio of the luminance component and the color component compression coefficient can be changed. As described above, the luminance component compression coefficient and the color component compression coefficient used to set the overall compression coefficient to a specific value are related to each other by a predetermined law, and the purpose thereof is reconfigured. The purpose is to reduce the effect of image compression.

The term "compression coefficient" is used in the following meaning. That is, when the compression factor is high, the overall degree of compression is high and the resulting amount of compressed data is small. When the compression factor is low, the overall degree of compression is low and the resulting amount of compressed data cannot be too small.

By determining two of the above-mentioned three parameters (luminance component compression coefficient, color component compression coefficient, and ratio of luminance component compression coefficient to color component compression coefficient),
The remaining parameters can be determined.

In the preferred embodiment, the control means varies the luminance component compression factor such that it is less than or equal to the color component compression factor. In particular, if the overall compression factor values are within one or more predetermined ranges (eg, for very large or very small overall compression factors), the control means causes the luminance component compression factor to be the color component compression factor. It is preferable to control so that the luminance component compression coefficient is lower than the color component compression coefficient when the overall compression coefficient value is not within the one or more predetermined range. This configuration limits the amount of compressed data that is formed and sacrifices color data quality by increasing the color component compression factor, leaving more room for perceptually more important luminance data. is doing.

In the preferred embodiment, each of the luminance component compression means and the color component compression means includes a decorrelator for converting the image data into conversion data representing a plurality of frequency components in the spatial frequency domain, and the conversion data. To the entropy encoder for encoding the quantized transform data.

According to the present invention, different degrees of compression must be applied to the luminance and color components of the input image data, and thus different decompression must be applied to the compressed luminance and color components. . To control this effect,
In the preferred embodiment of the present invention, means is provided for associating the compressed image data with a data header (header) representing the corresponding luminance component compression coefficient and color component compression coefficient.

When compressing a block of input image data to form a block of compressed data, the feedback means preferably comprises a compressed bit counter for storing the bit count of the block of compressed data. .
For a block of input image data of known size, this bit count is directly related to the overall compression factor.

The present invention provides a decompression device for decompressing (decompressing) compressed data including a luminance component and a color component, as seen from the second viewpoint. The data decompression device reads a data header associated with the compressed image data and applies individual luminance component compression and color component compression to each of the luminance component and the color component of the compressed image data. In response to the header reading means, the luminance component expanding means for expanding the luminance component of the compressed image data by an amount corresponding to the degree of compression of the luminance component, and the color component for responding to the header reading means. Luminance component decompression means for decompressing the color components of the compressed image data by an amount corresponding to the degree of compression of the.

According to a third aspect, the present invention provides a method of compressing the luminance and color components of input image data to produce compressed image data having one overall compression factor. This method compresses a luminance component of input image data by a luminance component compression coefficient, compresses a color component of input image data by a color component compression coefficient, a luminance component compression coefficient, a color component compression coefficient, and a luminance component. Controlling the overall compression factor by varying the ratio of compression factor to color component compression factor.

According to a fourth aspect, the present invention provides a method for decompressing compressed image data containing luminance and color components. The method reads a data header associated with the compressed image data and directs applying individual luminance and color component compressions to each of the luminance and color components of the compressed image data. From the steps of expanding the luminance component of the compressed image data by an amount corresponding to the degree of compression of the luminance component, and expanding the color component of the compressed image data by an amount corresponding to the degree of compression of the color component. Become. The present invention is particularly useful when applied to an image data storage device. Hereinafter, embodiments of the present invention will be specifically described with reference to the accompanying drawings in which the same constituent elements are designated by the same reference numerals.

[0020]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENT FIG. 1 shows an apparatus for separating and compressing intra-picture frequencies of a video signal in the two-dimensional spatial frequency domain. The video signal is in digital form and consists of a series of multi-bit (eg 8 bit) samples or words, each representing a pixel of the luminance and chrominance components of the image to be scanned, and through the input 10 a decorrelator 12 is input. The frequency-separated video signal is supplied to the quantizer 14 by the decorrelator 12, and is supplied to the entropy encoder 20 via the data sequencer (ordering device) 18. They together compress the frequency separated video signals supplied by the decorrelator 12 to form a compressed signal at the output 19. The compressed signal is transmitted and stored. After transmission or storage, the compressed signal is entropy decoded, reordered, using the inverse parameters of the parameters used in decorrelation, ordering, quantization, and entropy coding, respectively. Stretched through inverse (non) quantization and correlation operations,
It is almost restored to its original shape.

The decorrelation operation performed by the decorrelator 12 is a two-dimensional spatial frequency domain processing of an image (eg a field or frame of a video signal) due to the high correlation of adjacent pixels in the image. It is based on the fact that the amount of information needed to display an image can be reduced by creating frequency separated signal parts that represent different components of the image. In particular, the parts of the frequency separated signal represent different spatial frequency components of the image.

The ordering (changing the order) operation will be described in detail later. The act of quantizing is a lossy act that intentionally discards frequency data that is considered redundant or less important because the image is fully perceived by human mental visual tissues. This achieves signal compression. This quantizer 14
The compression can be performed by two methods. That is,
The quantizer reduces the number of levels to which the data input to it is assigned and increases the probability that a zero-valued sample will continue to occur in the data it outputs. The ability for signal compression enhancement enhanced by the operation of the quantizer is performed by the entropy encoder 20. That is, the quantizer 14 reduces the information content, which inevitably enables the entropy encoder to reduce the bit (data) rate.

In addition, (lossless) compression and bit (data) rate reduction are provided by the entropy coder 20 and produced by the quantizer 14 using methods known here, eg variable length coding. The data is encoded such that more probable (more frequently occurring) data items form shorter output bit sequences than less probable (less frequently occurring) data items. In this context, decorrelation behavior changes the probability distribution that any particular signal level will occur, which is about the same as the different levels considered before decorrelation, so that the probability of one level occurring is It has the effect of making the form of a much higher probability in comparison.

The compression / encoding device shown in FIG. 1 can be embodied in various ways using various forms of decorrelation. An increasingly generalized embodiment is so-called transform coding, and in particular, a transform form called discrete cosine transform is used. The use of the Discrete Cosine Transform for decorrelation is actually designed for the example compressor shown in FIG. 1 proposed by the Joint Photographic Experts Meeting and currently being considered by the International Organization for Standardization. According to the decorrelation transform technique, the signal is processed by a linear transform operation (decoration) before the quantization and coding operations. The disadvantage of this conversion technique is that it is impractical in terms of the amount of data involved, as the entire image (eg the entire field) has to be converted. Therefore, the image (field) must be divided into blocks (e.g., 8x8 representing pixels) and each of them must be transformed.
That is, transform coding is complicated and can only be used in block units.

A method proposed recently as compression / encoding in the frequency domain is a band division encoding method. In this method, the decorrelator 12 in the apparatus of FIG.
Is a spatial (two-dimensional) band division device that divides an input video signal into a plurality of unrelated division bands each including a spatial frequency component of the image in each one of a plurality of regions of the two-dimensional frequency space of the image. Contains. This sub-band is selectively quantized by quantizer 14 according to its respective position in the sensitivity spectrum of human mental and visual tissues. That is, in this case, the decorrelation is performed by distributing the energy of the entire image to the divided bands in the two-dimensional spatial frequency domain. Band-division filtering is believed to provide better decorrelation than conversion techniques. Similarly, unlike transform techniques, it is not limited to block-wise operations. Band division filtering can be applied directly to the video signal.

FIG. 2 shows a band division coding apparatus in which an input video signal passes through a low pass decimation (decimation) filter 22 and a high pass decimation filter 24. The two resulting output signals are
It represents different parts of the frequency spectrum of the input signal. The two signals are then quantized, ordered and entropy coded as described in FIG. The split band components of the input signal are transmitted or stored for later reproduction. The transmission and storage of the sub-band components is indicated by dashed line 26 in FIG.

When the divided band is restored from the recording medium,
These are passed through a corresponding matched filter to recover the original frequency components. These adaptive filters are a low pass interpolation filter 30 and a high pass interpolation filter 28. The outputs of the interpolation filters 28, 30 are added by the adder circuit 32 to form the original input video signal. Figure 2
FIG. 6 is a diagram showing how an input video signal is decomposed into two divided bands. In fact, the input video signal is decomposed into more sub-bands. FIG. 3 shows how the input signal is decomposed into eight sub-band components and recombined to form the output signal.

The filter in the band division encoding device is
It has a finite impulse response filter with appropriate delay and weighting factors to enable horizontal and vertical frequency decomposition. Different filter morphologies for subband frequency separation are known, and some possible filters have been identified by the ASSP-in "IEEE Transactions on Acoustics Speech and Signal Processing" published in June 1986. 34, pages 434-441, entitled "Exact Reconstruction Techniques for Band Division Coding of Tree Structures".

FIG. 4 is a diagram showing the decorrelator 12 in FIG. 1 in more detail. The decorrelator comprises a horizontal filter device 46, an intermediate field memory 48, a diversion sequencer (address generator) 50, a vertical filter device 52, an output field memory 54, and an output sequencer (address generator) 56. Band division filtering is performed individually. Thus, in FIG. 4, the filtering in the two orthogonal image directions, namely in the horizontal direction (the scanning direction of the image in normal video) and in the vertical direction, is one-dimensional by the horizontal and vertical filter devices 46 and 52. The filtering operations are completely independent of each other.

The horizontal filter device 46 and the vertical filter device 52 have substantially the same structure. Therefore, only the structure of the horizontal filter device 46 will be described in detail. The filtering action is divided into eight horizontal and vertical divided bands,
That is, it is an operation of forming a square array of 64 (8 × 8) divided bands. The 64 divided bands have the same size. The horizontal filter device 46 is shown in FIG.
It is preferable that the tree-like structure, that is, a hierarchical structure as shown in FIG.
The first stage consists of a low pass filter (LF) and a high pass filter (HF), each followed by a decimator (not shown). The LF filter, the HF filter, and the decimator together form a quadrature mirror filter (QMF). Each filter may be a conventional type finite impulse response filter. In use, the lines of the field of the input digital video signal are sample-wise added to the first stage and low-pass and high-pass filtered by LF and HF, respectively. in this way,
LF and HF form an output consisting of a low-pass filtered and a high-pass filtered version of the input line signal, respectively.
This output represents the spatial frequency components of the line in the upper and lower half of the horizontal spatial frequency range. That is, in the first step, the input line is horizontally divided into two divided bands. The decimator decimates (subsamples) each output by a factor of 2, making the total number of samples output by the decimator (all) equal to the total number of samples in the line.

The second stage is that there are two QMFs each as in the first stage, and that the output from each decimator of the first stage is sent as an input to each one of the two QMFs. Except for this, it has a similar structure to the first stage. Thus, the second stage forms four outputs representing the spatial frequency content of the line in four equal quarter horizontal spatial frequency ranges. That is, in the second step, the two divided bands formed by dividing the input line in the first step are further divided to form four divided bands in the horizontal direction. The second stage four decimator decimates (subsamples) each output at a rate of two. This causes the total number of samples output from the second stage decimator (total) to be equal to the total number of samples in the line.

The third stage is that there are four QMFs each as in the first stage and that the output from each decimator of the second stage is sent as one input to each of the four QMFs. And has a structure similar to the first stage. Thus, the third stage consists of eight equal eighths.
Form eight outputs representing the spatial frequency content of the line in the horizontal spatial frequency range of That is, in the third stage, the four divided bands formed by dividing the input line in the second stage are further divided to form eight divided bands in the horizontal direction. The third stage eight decimator decimates (subsamples) each output at a rate of two. This makes the total number of samples output from the third stage decimator (total) equal to the total number of samples in the line.

The eight outputs of the third stage or horizontal filter device 46 are sent to the intermediate field memory 48 for the first time.
It is stored in the positions corresponding to 1/8 of each line. The above process of horizontal filtering is repeated for all other lines of the field of the input digital video signal. As a result, the field of the input digital video signal filtered in the horizontal direction (only) into eight divided bands is stored in the intermediate field memory 48. Each line of the fields stored in the intermediate field memory 48 contains 8 respectively horizontal spatial frequency information representing one of the 8 sub-bands of the horizontal spatial frequency range of the image represented by the original field. It is divided into individual parts. The horizontally filtered field stored in the intermediate field memory 48 can be considered as divided into eight columns.

The horizontally filtered field stored in the intermediate field memory 48 is (converting sequencer 5
Supplied to the vertical filter device 52 (under control of 0),
Here, it is vertically filtered into eight sub-bands by the same method as that in which the horizontal filter device 46 filters the eight sub-bands in the horizontal direction. The horizontally and vertically filtered fields are supplied line by line to the output field memory 54 and sent to the quantizer 14. The memory 54 can be thought of as being divided into 64 (8.times.8) arrays, each of which stores 64 divided bands. Thus, successive fields of the input digital video signal are band-division filtered, appropriately filtered, delayed by two fields and sent to the quantizer 14.

The conversion sequencer 50 creates a read address for the intermediate field memory 48 and controls the reading of its contents into the vertical filter device 52 as follows. As mentioned above, the signal stored in the intermediate field memory 48 comprises the lines of the original field,
Each is horizontally divided into eight divided bands. That is, the signal stored in the intermediate field memory 48 can be considered to consist of eight columns, as already mentioned. To allow the signal stored in the intermediate field memory 48 to be vertically filtered by hardware of similar construction (vertical filter device 52) used to filter it horizontally, convert it. Must. That is, it must be rotated 90 degrees so that when read into the vertical filter device 52, it forms eight rows (rather than columns). Conversion sequencer 5
0 addresses the intermediate field memory 48 to accomplish this.

Due to the nature of the filtering formed by the combination of the horizontal filter device 46 and the vertical filter device 52, the data stored in the output field memory 54 will be somewhat scrambled, so that It is reordered by the output sequencer 56 before being sent to the subsequent parts.

FIG. 5 illustrates the various sub-band components that are formed when the input video signal is decomposed into eight frequency components in the horizontal and vertical directions (which provides the data to the output sequencer 56). And after being rearranged,
It can be considered as the data stored in the output field memory 54. ). Each of these sub-bands or sub-images consists of an array of data elements and is represented by one of the blocks in FIG. The upper left block represents the dc divided band. This is not necessarily an exact representation of the constant part of the signal whose frequency is strictly zero, but it is the lowest band both horizontally and vertically. This dc sub-band will contain most of the dc information of the original input video signal. The relative importance of the image perceived by the viewer in the remaining sub-bands varies. Generally speaking, the high frequency sub-bands are of less importance to the ultimate viewer perception. In FIG. 5, the frequency represented by a particular sub-band component increases towards the bottom and right side of the array of blocks.

FIG. 6 shows the mental and visual reaction of a human to image components of changing spatial frequency.
As can be seen from this, the human perception level first rises and then decreases at an almost constant rate as the spatial frequency increases. The data compression apparatus utilizes this characteristic, recognizing that the higher the spatial frequency component is, the higher the lossy loss of information and the higher the loss of information, the more severely the reconstructed image is not perceived.

FIG. 7 shows a quantization matrix applied to different divided bands by the quantizer 14 shown in FIG. The operation method of the quantizer is to divide the value of each element in each divided band block by the corresponding quantized value and adopt the integer part of the result. This quantized value is the one entered in the quantization matrix (α ₀₀ ~ α
₇₇ ) and a quantization scale (compression) factor (Qs) described in detail later. That is, the value read from the dc luminance division band in the upper left corner of the matrix is divided by the corresponding quantized value of Qs × α ₀₀ and the integer is taken. Similarly, a value read from another sub-band is divided by a quantized value Qs × α _mn suitable for this sub-band,
Its integer value is taken. In general, the smallest quantized value is
Occurs just below and to the right of the dc luminance split band. This is because the human psycho-perceptual system is most sensitive in this divided band. The values for the quantisation matrix are determined by a process of trial and error by seeing which value gives the best image. Alternatively, it can be generated by expanding the curve in FIG. 6 into three dimensions to create a curved surface. This curved surface is a curved surface formed by a locus formed by rotating the curve of FIG. 6 around the sensitivity coordinate axis by 90 degrees.
The subband frequency blocks are in different regions of the spatial frequency plane below this curved surface. The volume between each frequency block and this curved surface can be calculated by integrating below the curved surface to determine the relative quantized value.

FIG. 8 is a block diagram showing the outline of a data compression apparatus of the present invention capable of individually and dynamically controlling the quantization of the luminance component and the color component. The luminance (Y) component data 60 is supplied to the decorrelator 12 ′, and then supplied to the luminance component quantizer 64. The output of the luminance component quantizer is sent to the entropy encoder 2 through the data sequencer 18 '.
0 ', which outputs the luminance (Y) component compressed data to the output terminal 68. Similarly, the color (U / V) component data input 62 is supplied to the decorrelator 12 ″ and from there to the color component quantizer 66. The output from the color component quantizer 66 is the data sequencer 18 ″. Is supplied to the entropy encoder 20 ″ through the encoder and outputs the color (U / V) component compressed data to the output terminal 70. The luminance component quantizer 64 and the color component quantizer 66 are the quantizers described above. The input data of the luminance component and the color component are individually compressed in this manner, and the compressed data outputs 68 and 70 of the luminance component and the color component are combined by a combiner (not shown). Combined into a single data stream for further processing.

Combined bit counter 72 detects the amount of compressed data at luminance component compressed data output 68 and color component compressed data output 70 and produces a feedback signal 74 indicative of the sum of these two amounts. This feedback signal 74 is sent to the quantization controller 76, which
Reference numeral 6 individually changes the degree of the luminance component quantization performed by the luminance component quantizer 64 and the color component quantization performed by the color component quantizer 66 to obtain combined data of the compressed data output of the luminance component and the color component. Make the rate equal or close to the desired value. The quantization controller 76 and the means by which the quantization controller 76 changes the degree of quantization performed by the luminance component quantizer 64 and the color component quantizer will now be described in more detail.

FIG. 9 is a graph showing a relative change performed by the quantization controller 76 with respect to the quantization scale factor Qs used in the luminance component quantizer 64 and the color component quantizer 66. As described above with reference to FIG. 7, the degree of quantization applied to each divided band changes depending on the control of the quantization matrix. The overall degree of quantization of the entire block of compressed data (ie 64 sub-bands) varies depending on the quantization scale factor (Qs). Each data value being quantized is divided by both the appropriate value from the quantization matrix and the value of Qs. When Qs is high, the overall degree of quantization is increased, the degree of compression is also increased, and the resulting blocks of compressed data are smaller. When Qs is low, the overall degree of quantization is reduced, the degree of compression is also reduced, and the resulting blocks of compressed data are larger.

FIG. 9 shows two curves, that is, a luminance component curve (solid line) and a color component curve (broken line). The vertical axis shows the value of Qs and the horizontal axis shows schematically the corrective action taken by the quantization controller 76 to change the overall compression factor. Moving the horizontal axis to the right increases the overall compression factor. Moving the horizontal axis to the left also reduces the overall compression factor. The value of Qs used by the luminance component quantizer 64 and the color component quantizer 66 always corresponds to the position of the vertical axis of each point on the luminance component curve and the color component curve. These two points are at the same position on the horizontal axis.

The common origin 78 of the luminance component curve and the color component curve corresponds to the value of Qs = 1. Moving from the origin 78 to the right, it can be seen that Qs increases at the same rate as both the luminance component compression and the color component compression. Point 8
At 0, the luminance component and color component curves separate such that the Qs value for color component compression increases more rapidly than the Qs value for luminance component compression. This separation continues to point 82. From point 82 to point 84, the value of Qs for luminance component compression continues to increase, but Qs for color component compression
The value remains constant. From point 84, the two curves again increase at the same rate.

From the graph of FIG. 9, the overall compression factor is
It can be seen that control is performed by changing Qs (luminance component compression coefficient) for luminance component compression, Qs (color component compression coefficient) for color component compression, and a ratio of Qs for luminance component compression and Qs for color component compression. Will. When two of these parameters are fixed,
The third parameter is fixed as well.

FIG. 10 shows in more detail a data compression device that operates to dynamically control the degree of quantization of the luminance and chrominance components individually. The apparatus comprises a luminance component compression channel 86 (surrounded by a dashed line), a corresponding color component compression channel (not shown), a compression bit counter 72 and a combiner / header inserter 88. . Input data for the luminance and color components is received as blocks of known size and compressed to form output data blocks.

In FIG. 10, the luminance component compression channel 8
The 6 and color component compression channels operate in a similar manner, but the quantisation values of their respective quantisation matrices are different and Qs varies differently, as described above with reference to FIG. . However, from the perspective of the description of FIG. 10, the operation of the rest of the luminance and chrominance component channels is the same, so the apparatus of FIG. Only common items between and will be explained.

Luminance component compression channel 86 shown in FIG.
Can perform trial compression of the luminance component input data to find out how much compression can be performed with the current value of Qs. After checking this, generate the corrected value of Qs,
Apply the same luminance component input data to create compressed luminance data that merges with the output data stream of the data compressor.

In operation, the luminance component input data is provided to decorrelator 12 '. The output from the decorrelator 12 'is provided to a test quantizer 90, which applies a trial quantisation based on the Qs value stored in the previous Qs latch 100 for the previous block of input image data. . The test quantizer 90 operates similarly to the luminance component quantizer 64 described in connection with FIG. Similarly, the test data sequencer 92 and the test entropy encoder 94 are the data sequencer 18 'and the entropy encoder 20' in FIG.
Works the same as and measures the correct value for the degree of compression achieved.

The output from the test entropy encoder 94 for the luma component compression channel is provided to a compressed bit counter 72. The compressed bit counter 72 also receives the output of the corresponding test entropy encoder of the color component compression channel (not shown). The compression bit counter 72 measures the amount of data (bit count value) output during the test compression of the block of the input image data by the test entropy encoder in the luminance component compression channel and the color component compression channel. It detects and produces a feedback signal indicative of the sum of these two quantities. At the final stage of test compression,
The output from the compressed bit counter 72 is a programmable read only memory (PRO) of the luma component compression channel that calculates an alternative value for Qs according to the luma component curve shown in FIG.
M) 98. Similarly, the corresponding PROM 98 of the color component compression channel receives a similar output from the compression bit counter 72 and calculates an alternative Qs value to use for color component compression. The method by which the PROM 98 performs these calculations will be described with reference to FIG.

PROM 98 of luminance component compression channel 86
Outputs the corrected value of Qs to the preceding Qs latch 100 for use in the next test compression, and outputs it to the Qs latch 102 for compression of the current luminance component input data. Qs latch 1
The output from 02 is fed to a luminance component quantizer 64, where the overall degree of quantization is controlled. Decorrelator 1
When the data from 2'is test compressed, this data is also processed by delay circuit 104. This delay circuit serves to delay the arrival of this block of data at the luma component quantizer 64 until the corrected Qs value is stored in the Qs latch 102. Thus, the correction value for Qs is applied to the same data block to be corrected.

The output from the luminance component quantizer 64 is a field memory 60 for storing compressed data prior to the final format creation by the entropy encoder 20 '.
Is entropy coded by an entropy coder 20 'coupled with and output from the system.

When decompressing (decompressing) a block of compressed data having a variable degree of quantization applied to the block, the decompressor may need to respond to the degree of quantization applied to the block of data. You will understand. Therefore, the value stored in the Qs latch 102 is supplied as header data for the block of compressed data stored in the field memory 60. The combiner / header inserter 88 prepends this header data to the compressed data block. This allows the decompression device to read from the header data what quantization scale factor has been applied to each block of compressed data. Each P in the luminance component channel and the color component channel
The combination of the ROM 98 and the Qs latch 102 can be regarded as the quantization controller 76.

The output from the entropy encoder 20 'is provided to a second compressed bit counter 108, the output from which is provided continuously to an overflow detector 106.
The overflow detector 106 compares the instantaneous compression bit count value in each compression channel with a predetermined maximum size of a block of compressed data being compiled, and outputs an overflow indication signal when the predetermined maximum size is exceeded. The overflow indication signal triggers the entropy encoder 20 'to generate a code that is not normally created for compressed data so that the overflow indication signal is not confused with the original compressed data. An appropriate code can be selected by referring to the code table of the entropy encoder.

The compressed data may be recorded as individual tracks on magnetic tape for later decompression and decompression, for example. It will be appreciated that the blocks of compressed data may be processed in various other ways before decompression.

FIG. 11 illustrates a technique for the PROM 98 in the luma compression channel and the color compression channel to determine how the Qs should change between different blocks of input data. PROM 98 reads the output from compressed bit counter 72 and subtracts from it a predetermined maximum size of a block of known compressed data. This produces a difference value D. The PROM 98 then references a look-up table in which the data represented by the graph of FIG. 11 is stored. Appropriate quantization changes for the measured difference value D and the current Qs value can be read from this look-up table. This change follows the appropriate curve in FIG. When the measured difference value is positive, it means that the compressed data is too large, and a certain positive value is added to the current value Qs to create a new Qs value, which is used for the next data block. . The larger the value of Qs, the further the compression. If the difference value is negative, the block of compressed data is too small and the negative value is set to the current value Qs.
Added to. The new value of Qs is then Qs Latch 1
02.

This technique addresses both excessive and insufficient compressed block size. If the applied quantization is too tight, the compressed block will be too small and PROM 98 will decrease the value of Qs according to the appropriate curve shown in FIG. When the quantization is not so severe, the compressed block becomes too large and the PROM 9
8 will increase the value of Qs. In this way, the size of the compressed block is kept approximately equal to the predetermined maximum size. This makes it possible to prevent excesses from being applied and to avoid applying too tight a degree of quantization.

FIG. 12 shows a data decompression device having a complementary relationship with the data compression device shown in FIG. The decompression device includes a header reader, a data splitter, a luminance component decompression channel 112 and a color component decompression channel 114.
It is equipped with The extension channels for the luminance and color components are
Similar in structure and operation, only the luma component expansion channel will be described in detail. The luma component expansion channel comprises an entropy decoder and overflow detector 116, a field memory 118, a data sequencer 120, a luma component dequantizer 122 and an interpolator 124.

The compressed data from the data compression device is transferred to the data decompression device via one of the already known mechanisms (such as the magnetic storage device and the subsequent reproduction device), where it also serves as a header reader / reader. It passes through the data divider 110. The luminance component compressed data is reordered by the entropy decoder / overflow detector 116 and stored in the field memory 118. The header reader / data divider 110 reads the value of Qs included in the header data of the compressed block used for the compression of the luminance component, and stores it in the Qs latch 126. The compressed luminance data itself flows into the field memory 118 through the header reader, the entropy decoder and the overflow detector 116.

Entropy decoder and overflow detector 116
Detects the occurrence of an overflow indicator code in the data stream and overflows (excess) when the data is compressed.
Is set, all output data values that cannot be determined from the compressed data are set to predetermined values in the field memory 118, for example, all undetermined values are set to zero.

The Qs latch 126 supplies the value of Qs to the luminance component dequantizer 122 while dequantizing the compressed luminance data from each block. Therefore, a non-quantization degree that matches the quantization degree applied at the time of compression is applied to the compressed luminance component data.

Entropy decoder and overflow detector 116,
The data sequencer 120 and the interpolator 124 are the entropy encoder 20, the data sequencer 18, and the decorrelator 1.
It has a complementary relationship with 2. The luminance component dequantizer 122 is
Each compressed data value is multiplied by the appropriate value from the quantisation matrix and the Qs value from the Qs latch 126 to provide a complementary operation to that performed by the luma component quantizer 64.

The color component expansion channel 114 operates similarly. The header reader and data divider 110 is
The quantization scale factor Qs used for color component compression and stored in the header data is supplied to the color component (U / V) Qs latch 128, and the compressed color data is entropy decoded by the color component expansion channel 114. Supply to the device / overflow detector 116. The dequantization operation is performed by the color component dequantizer 130.

[0064]

According to the present invention, the luminance component compression means and the color component compression means are individually provided, and the feedback means for outputting the feedback signal according to the overall compression coefficient is provided. Further, the control means changes the luminance component compression coefficient, the color component compression coefficient, and the ratio of the two to control the overall compression coefficient. Thus, the effect of image compression on the reconstructed image can be reduced while achieving the desired degree of compression.

[Brief description of drawings]

FIG. 1 is a diagram showing an outline of a data compression device.

FIG. 2 is a diagram showing a simple example of band division encoding.

FIG. 3 is a diagram illustrating an example of high-order band division encoding.

FIG. 4 is a diagram showing an example of a two-dimensional band division decorrelator.

FIG. 5 is a diagram showing an example of a frequency-separated video signal.

FIG. 6 is a diagram showing the reaction of human mental visual tissues to images of different spatial frequencies.

FIG. 7 is a diagram showing a quantization matrix.

FIG. 8 is a block diagram showing an outline of a data compression apparatus of the present invention for individually compressing a luminance component and a color component.

9 is a graph showing changes in compression coefficient of a luminance component and a color component in the apparatus of FIG.

FIG. 10 is a block diagram showing details of a data compression apparatus of the present invention for individually compressing a luminance component and a color component.

11 is a diagram showing a technique for determining changes in compression coefficients of a luminance component and a color component in the apparatus of FIG.

12 is a diagram showing a data decompression device having a complementary relationship with the compression device of FIG.

[Explanation of symbols]

 12 ', 12 "decorrelator 20', 20" entropy encoder 64 luminance component quantizer 66 color component quantizer 72 combined bit counter (feedback means) 76 quantization controller (control means) (12 ' , 64, 20 ') Luminance component compression means (12 ", 66, 20") Color component compression means

Front Page Continuation (72) Inventor Clive Henry Gillard England RG240XQ Hampshire, Basingstoke, Cineham, Cafford Village, Saffron Claus 47

Claims (11)

[Claims] 1. A data compression device for compressing a luminance component and a color component of input image data to create compressed video data having an overall compression coefficient, wherein the luminance component of the input image data is a luminance component. Luminance component compression means for compressing according to a compression coefficient, color component compression means for compressing a color component of the input image data according to a color component compression coefficient, and a feedback signal according to the overall compression coefficient are supplied. For the color component compression coefficient of the brightness component compression coefficient, the color component compression coefficient, and the brightness component compression coefficient in response to the feedback signal. An image data compression apparatus, comprising: a control unit that controls the overall compression coefficient by changing a ratio. 2. The apparatus according to claim 1, wherein the control means changes the luminance component compression coefficient so as to be smaller than or equal to the color component compression coefficient. 3. The control means causes the luminance component compression coefficient to become equal to the color component compression coefficient for one or more predetermined ranges of the value of the overall compression coefficient, and the overall compression coefficient. 3. The apparatus according to claim 2, wherein when the compression coefficient is out of the one or more predetermined ranges, the luminance component compression coefficient is changed to be smaller than the color component compression coefficient. 4. The luminance component compression means and the color component compression means each include a decorrelator for converting image data into conversion data representing a plurality of frequency components in a spatial frequency domain, and the conversion data with respect to the conversion data. 4. The apparatus according to claim 1, further comprising a quantizer that applies a variable quantization degree and an entropy encoder that encodes the quantized transform data. 5. The apparatus according to any one of claims 1 to 4, further comprising means for associating a data header representing a luminance component compression coefficient and a color component compression coefficient corresponding to the compressed image data with the compressed image data. . 6. The block of input image data is compressed to form a block of compressed data, and the feedback means further comprises a compressed bit counter for storing a bit count of the block of compressed data. Item 1
The device according to any one of to 5. 7. A data decompression device for decompressing compressed image data including a luminance component and a color component, wherein a data header associated with the compressed image data is read and the luminance component and the color component of the compressed image data are read. And a header reading means for individually instructing the degree of compression of the luminance component and the degree of compression of the color component applied to, and expanding the luminance component of the compressed image data by an amount corresponding to the degree of compression of the luminance component in response to the header reading means. A data decompression device comprising: a luminance component decompression means for responsive to the header reading means; and a color component decompression means responsive to the header reading means for decompressing the color component of the compressed image data by an amount corresponding to the degree of the color component compression. 8. A method for compressing a luminance component and a color component of input image data to create compressed image data having one overall compression coefficient, wherein the luminance component of the input image data is a luminance component. Compressing by the compression coefficient, compressing the color component of the input image data by the color component compression coefficient, the luminance component compression coefficient, the color component compression coefficient, and the ratio of the luminance component compression coefficient to the color component compression coefficient. An image data compression method comprising the steps of controlling the overall compression factor by varying. 9. A method of decompressing compressed image data including a luminance component and a color component, comprising reading a data header associated with the compressed image data and applying the data header to the luminance component and the color component of the compressed image data, respectively. The luminance component compression degree and the color component compression degree are individually designated, the luminance component of the compressed image data is expanded by an extent according to the luminance component compression degree, and the degree according to the color component compression degree. An image data expansion method including the steps of expanding only the color components of the compressed image data. 10. An image data storage device comprising the image data compression device according to claim 1. 11. An image data storage device comprising the image data expansion device according to claim 7.