JP3282138B2 - Image data compression processing method and image data reconstruction method - Google Patents

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method for compressing image data, and more particularly, to a method for compressing image data for reducing the amount of data of an original image by using a wavelet transform, and a method for regenerating compressed image data. It concerns the method of configuration.

[0002]

2. Description of the Related Art An image signal carrying a halftone image, such as a TV signal, has an enormous amount of information, and therefore requires a wide-band transmission line for transmission. Therefore, conventionally, attention has been paid to the fact that such an image signal has high redundancy, and various attempts have been made to compress the image data by suppressing this redundancy. Also, recently, recording halftone images on, for example, optical disks and magnetic disks has been widely performed. In this case, image data compression has been widely applied for the purpose of efficiently recording image signals on a recording medium. ing.

As one of such image data compression methods, conventionally, when image data is stored or transmitted, the image data is subjected to compression processing by predictive coding to reduce the data amount. Perform storage, transmission, etc.
At the time of image reproduction, a method is employed in which the compressed image data (compressed image data) is decoded and decompressed, and a visible image is reproduced based on the decompressed image data (decompressed image data). Have been.

As one of the image data compression methods,
A method using vector quantization is known. This method divides the two-dimensional image data into blocks of K samples, and includes, in a codebook composed of a plurality of different vectors created in advance by defining K vector elements, in each of the blocks, A vector corresponding to a set of image data and a minimum distortion are respectively selected, and information indicating the selected vector is encoded in association with each block.

[0005] Since the image data in the above-described blocks have a high correlation with each other, the image data in each of the blocks is represented by one of a relatively small number of vectors prepared.
One can be used to indicate fairly accurately. Therefore, since the transmission or recording of the image data can be performed by transmitting or storing a code indicating this vector instead of the actual data, data compression is realized. For example, the image data amount for 64 pixels in a halftone image of a density scale of 256 levels (= 8 bits) is 8 × 64 = 512 bits, and this 64 pixels is regarded as one block, and each image data in the block is represented by 64 elements. Represented by a vector consisting of
If you create a codebook prepared as follows,
The data amount per block is the data amount for vector identification, that is, 8 bits, and eventually the data amount is 8 / (8 ×
64) = 1/64.

After the image data is compressed and recorded or transmitted as described above, the vector element of the vector indicated by the vector identification information is used as reconstructed data for each block.
Using this reconstructed data, the original image is reproduced.

[0007] One of the methods for improving the compression ratio in the case of performing the above-described data compression by predictive coding is as follows.
It is conceivable to reduce the bit resolution (density resolution) of the image data together with the predictive encoding process, that is, to perform a quantization process for coarsely quantizing the image data.

Therefore, the applicant of the present invention has proposed an image data compression method by interpolation coding which combines the above-described method by predictive coding and the method by quantization (Japanese Patent Laid-Open No. 62-247676). In this method, image data is divided into main data sampled at appropriate intervals and interpolation data other than the main data, and the interpolation data is subjected to interpolation prediction coding processing based on the main data, that is, the interpolation data is converted into the main data. , And compresses image data by performing variable-length coding such as Huffman coding (conversion to a signal whose code length changes depending on a value) on a prediction error.

When compressing image data, it is naturally desirable that the compression ratio be high. However, it is technically difficult to desire a large improvement in the compression rate in the interpolation coding. Therefore, in order to achieve a higher compression rate, the image data number reduction processing for reducing the spatial resolution is combined with the interpolation coding. It is conceivable to match them.

Therefore, the applicant of the present application has proposed an image data compression method that combines the above-described interpolation coding and the image data number reduction processing and can achieve a higher compression ratio while maintaining higher image quality (particularly). JP-A-2-280462).

On the other hand, a method called wavelet transform has been proposed as a method for processing the above-mentioned image data.

Here, the wavelet transform will be described.

The wavelet transform has recently been developed as a method of frequency analysis, and is applied to stereo pattern matching, data compression, and the like (OLIVIER RIOUL and MARTIN VETTERLI; Wavelets a).
nd Signal Processing, IEEESP MAGAZINE, P.14-38, OCTOB
ER 1991, Stephane Mallat; Zero-Crossings of aWavele
t Transform, IEEE TRANSACTIONS ON INFORMATION THEOR
Y, VOL. 37, NO. 4, P. 1019-1033, JULY 1991).

In this wavelet transform, a function h as shown in FIG. 8 is used as a basic wavelet function.

(Equation 1) Since the signal is converted into a frequency signal for each of a plurality of frequency bands in the following formula, a problem of spurious vibration such as Fourier transform does not occur. That is, if the filtering process is performed by changing the period and the reduction ratio of the function h and moving the original signal, a frequency signal suitable for a desired frequency from a fine frequency to a coarse frequency can be created. For example, as shown in FIG. 9, the signal Sorg is subjected to the wavelet transform and the signal subjected to the inverse wavelet transform for each frequency band, and as shown in FIG. 10, the signal Sorg is subjected to the Fourier transform and the inverse Fourier transform is performed for each frequency band. In terms of signals, the wavelet transform can obtain a frequency signal in a frequency band corresponding to the vibration of the original signal Sorg as compared with the Fourier transform. That is, in the Fourier transform, a vibration occurs in a portion B 'of the frequency band 7 corresponding to the portion B of the original signal Sorg, whereas in the wavelet transform, a portion of the frequency band W7 corresponding to the portion A of the original signal Sorg. No vibration occurs in A 'as in the case of the original signal.

Also, using this wavelet transform,
A method for compressing the image data described above has been proposed (Marc Antonini et al., Image Coding Using Wavelet).
Transform, IEEE TRANSACTIONS ON IMAGE PROCESSING
, VOL.1, NO.2, p205-220, APRIL 1992).

In this method, original image data representing an image is
By sequentially performing wavelet transform while sampling at a predetermined interval by a predetermined basic wavelet function, the original image data is converted into a plurality of image data having different combinations of frequency bands from a high frequency band to a low frequency band in both main and sub directions. Decompose, reduce the number of bits or set the number of bits to 0 for high frequency band image data that carries a lot of noise components to these image data,
The original image data is compressed by allocating a large number of bits to the low frequency band image data carrying the information of the main subject and performing the above-described vector quantization. According to this method, the compression ratio of the original image data can be improved, and the inverse wavelet transform is sequentially performed on the compressed image data while supplementing the data portion decimated at a predetermined sampling interval. By applying
The original image can be completely restored.

In the above-described method of compressing image data using the wavelet transform, the original image data is divided into image data from a high frequency band to a low frequency band by repeating the wavelet transform and sampling at predetermined intervals. Therefore, aliasing occurs when sampling is performed. Here, aliasing refers to a phenomenon in which high-frequency components of original data are mixed with low-frequency components. Therefore, by sampling the image data, the high frequency components of the sampled image data are mixed with the low frequency components. For example,
When the sampling interval described above is sampled for each pixel of the image data, the Nyquist frequency (when sampling a signal having a limited frequency band at a fixed interval, the maximum sampling interval that can uniquely describe the original signal waveform) Reciprocal of value)
Is mixed with the low frequency component, and the Nyquist frequency of the image data is １ ／ of the Nyquist frequency of the original image data. When such aliasing occurs, when the wavelet-transformed image data is subjected to the inverse wavelet transform, the sampled data cannot be completely restored as the original image,
There is a problem that an artifact (false image) occurs in a portion of the sampled image data.

This problem can be solved by appropriately selecting a basic wavelet function used when performing a wavelet transform on the original image data. That is, by appropriately selecting the basic wavelet function, when performing inverse wavelet transform on the wavelet-transformed image data, the effects of aliasing can be compensated for each other in the low frequency band and the high frequency band. Artifacts do not occur due to aliasing caused by sampling data.

[0019]

However, the method of Antonini et al. Described above performs quantization by reducing the number of bits to 0 or less for image data in a high frequency band among image data obtained by wavelet transform. Therefore, the amount of image data to be compensated for the aliasing required when performing the inverse wavelet transform on the image data becomes zero or small. Therefore, although the method of Antonini et al. Described above can improve the compression ratio of image data, it is not possible to appropriately compensate for aliasing when reconstructing the compressed image data, and The above-described artifact occurs.

Such a problem can be solved by increasing the filter size of the basic wavelet function when performing the wavelet transform, thereby increasing the degree of freedom of the frequency transform response and reducing the aliasing. However, when the filter size of the basic wavelet function is increased, the operation time for performing the wavelet transform becomes longer, and high-speed compression processing cannot be performed.

In view of the above circumstances, the present invention can compress image data at a high compression rate, and can perform processing at a higher speed without a problem of artifacts occurring in a reconstructed image. An object of the present invention is to provide a compression processing method and an image data reconstruction method.

[0022]

According to the image data compression processing method of the present invention, the original image data representing the image is sampled at predetermined intervals and sequentially subjected to a wavelet transform, whereby the original image data is reduced from a high frequency band to a low frequency band. Sequentially decomposing into a plurality of coefficient image data representing different frequency bands up to the frequency band, and dividing the plurality of coefficient image data at least with respect to the coefficient image data of the highest frequency band by a smaller number of bits than the coefficient image data of the other frequency band. An image data compression processing method for quantizing and removing noise in high frequency components of the original image data by encoding the quantized coefficient image data and performing compression processing on the original image data, wherein the coefficient image The lower the data frequency band, the larger the filter size that reduces aliasing. It is characterized in applying the wavelet transform function as a basic wavelet function.

In the image data reconstructing method according to the present invention, the encoded coefficient image data is decoded, and the filter size for reducing aliasing decreases as the frequency band of the decoded coefficient image data decreases. The invention is characterized in that the original image data compressed by the image data compression processing method according to the present invention is reconstructed by performing an inverse wavelet transform using a large function as a basic wavelet function.

[0024]

According to the image data compression processing method of the present invention, the image data compression processing method of Antonini et al., Which compresses image data by performing a wavelet transformation, when performing a wavelet transformation while sampling original image data. For the coefficient image data in the low frequency band, the filter size of the basic wavelet function used when performing the wavelet transform is increased as compared with the coefficient image data in the high frequency band. That is, since the image data in the high frequency band contains many components such as noise and the number of data to be processed is large, even if aliasing occurs, the image quality of the original image is not so affected. The compression processing is performed at high speed by performing a wavelet transform using a small basic wavelet function. On the other hand, the coefficient image data in the low frequency band carries important information and the number of data to be processed is small, so that the aliasing is reduced by performing a wavelet transform using a basic wavelet function having a large filter size. It is.

For this reason, even when the compressed reduced image data is reconstructed, no artifact due to aliasing occurs in the image of the frequency band carrying important information.
In addition, high-frequency image data having a large number of data to be processed can be subjected to high-speed wavelet transform. Therefore, as a whole, an image free from artifacts can be reconstructed, the image data can be compressed at a high speed, and the image data can be compressed at a high compression ratio.

[0026]

Embodiments of the present invention will be described below with reference to the drawings.

FIG. 1 is a diagram showing a basic concept of an embodiment of an image data compression processing method according to the present invention.

As shown in FIG. 1, the image data compression processing method according to the present invention is based on a function having a larger filter size in a lower frequency band according to the method of Antonini et al. The wavelet transform 2 is performed as a wavelet function to obtain coefficient image data 3 for each of a plurality of frequency bands. Next, quantization 4 is performed on the coefficient image data 3 obtained by the wavelet transform 2 with a smaller number of bits as the frequency band becomes higher.
The encoding 5 is performed on each of the quantized image data 3.

Hereinafter, the embodiment of the present invention will be described in detail.

This embodiment is directed to a radiographic image information recording / reproducing system using a stimulable phosphor sheet recorded in, for example, JP-A-55-12492 and JP-A-56-11395. It is intended to read a radiation image of a human body recorded on a body sheet as digital image data by laser beam scanning. The radiation image is read by moving the sheet 10 in the sub-scanning direction (vertical direction) while scanning the stimulable phosphor sheet 10 with a laser beam in the main scanning direction (horizontal direction) as shown in FIG. This is performed by scanning the sheet 10 two-dimensionally.

Next, wavelet transform is performed on the original image data.

FIG. 3 is a diagram showing details of the wavelet transform on the original image data Sorg.

In the present embodiment, the orthogonal wavelet transform in which each coefficient of the wavelet transform is orthogonal is performed, and is described in the above-mentioned document by Marc Antonini et al. Further, in this embodiment,
For the coefficient image data in the highest frequency band among the coefficient image data obtained by the wavelet transform, quantization is performed with the number of bits set to 0.

First, as shown in FIG.
basic wavelet functions determined by the basic wavelet function as a reference in the main scanning direction of the org (hereinafter simply referred to as function) performs a filtering process by the g _N and functions h _N. That is, the filtering process for each row of pixels arranged in the main scanning direction by the functions g _N and h _{N is} performed while shifting one pixel at a time in the sub scanning direction, and the coefficient image data of the original image data Sorg in the main scanning direction is obtained. Is what you want.

Here, the functions g _N and h _N are uniquely obtained from the reference basic wavelet function, and the filter size is increased as the coefficient image data becomes low-frequency band data by wavelet transform. is there. The function h _N are those shown in Table 1 below. In Table 1, it is assumed that N = 2 (a function having the smallest filter size, that is, a basic wavelet function serving as a reference) to N = 10 (a function having the largest filter size). Here, g _N are those obtained from h _N, the relationship between g _N and h _N denote the following equation (2).

[0036]

[Table 1] g _N = (− 1) ⁿ h _N (2) The above-mentioned original image data Sorg is subjected to a filtering process using the functions g ₂ and h ₂ having the smallest filter sizes, and the coefficient image data W of the original image data is obtained.
It is assumed that g0 and Wh0 are obtained. Here, the function h ₂
Is shown in FIG. 4 together with its frequency emphasis characteristics.

[0037] In this way, when the coefficient image data Wg0, Wh0 determined by the function g _2, h _2, the coefficient image data Wg0, Wh0, sampling the main scanning direction of the pixel on every other pixel in the main scanning direction Halves the number of pixels. This sampling, of the Nyquist frequency of the filtered image data by the function h _2,
Aliasing occurs in which high-frequency components having a Nyquist frequency of 0.5 or more are mixed with low-frequency components having a Nyquist frequency of 0.5 or less. Next, filtering processing is performed by the functions g ₂ and h _{2 in} the sub-scanning direction of each of the coefficient image data Wg 0 and Wh 0 in which this pixel has been thinned out, and the coefficient image data WW ₀ and WV
₀ , VW ₀ and VV ₀ are obtained.

Next, the coefficient image data WW ₀ , W
For V ₀ , VW _0, and VV ₀ , pixels in the sub-scanning direction are sampled every other pixel, and processing is performed to reduce the number of pixels in the sub-scanning direction to half. This sampling causes aliasing as described above. Thereby, each coefficient image data VV ₀ , WV ₀ , VW ₀ ,
The number of pixels of WW ₀ is 1/4 of the number of pixels of the original image data Sorg.
Becomes Next, the coefficient image data V is calculated using functions g ₃ and h ₃ having filter sizes larger than the functions g ₂ and h _2.
Performs filtering processing by the main scanning direction V _0.
Incidentally, the function h ₃ functions g ₃ is the filter size is a 6 As shown in Table 1 is obtained from the function h _3.

That is, the filtering processing for each row of the pixels arranged in the main scanning direction is performed by the functions g ₃ and h ₃ while shifting one pixel at a time in the sub-scanning direction.
And requests the coefficient image data Wg1 and Wh1 in the main scanning direction V _0.

Here, the number of pixels of the coefficient image data VV ₀ is equal to that of the original image data by sampling in both main and sub directions.
Since it is 1/2, the frequency band of the image is half that of the original image data. Therefore, by performing a filtering process to the coefficient image data VV ₀ the function g _3, h _3, coefficient image representing a low frequency component than the frequency component represented by the coefficient image data VV ₀ among the frequency components of the original image data data Wg1 , Wh1 are obtained.

As described above, the coefficient image data Wg1, Wg
When h1 is obtained, pixels in the main scanning direction are sampled every other pixel in the coefficient image data Wg1 and Wh1,
The number of pixels in the main scanning direction is further reduced to half. Next, filtering processing is performed by the functions g ₃ and h _{3 in} the sub-scanning direction of the coefficient image data Wg 1 and Wh ₁ , and the coefficient image data WW ₁ , WV ₁ , VW ₁ and VV
_{Get 1} .

Next, the coefficient image data WW ₁ , W
For V ₁ , VW ₁ , and VV ₁ , a process is performed in which pixels in the sub-scanning direction are sampled every other pixel, and the number of pixels in the sub-scanning direction is halved. Thus, the number of pixels of each of the coefficient image data VV ₁ , WV ₁ , VW ₁ , and WW ₁ is 1/16 of the number of pixels of the original image data Sorg.

[0043] In the same manner as described above, the function g in the main scanning direction of the coefficient image data VV ₁ which pixels are thinned out
_3, h ₃ performs filtering processing by a large function g _4, h ₄ is the filter size than samples the main scanning direction of the pixel of the further coefficients obtained image data, the coefficient image data obtained by thinning out the pixels, sub Filtering is performed in the scanning direction by the functions g ₄ and h ₄ , and the coefficient image data WW ₂ , WV ₂ ,
VW ₂ and VV ₂ are obtained.

By repeating such a wavelet transform N times, the coefficient image data WW _{0 to} WW _N , W
_{V 0 ~WV N, VW 0 ~VW} N, and VV
_{Get N.} Here, coefficient image data WW _N , WV _N , VW obtained by the N-th wavelet transform
_N, VV _N, as compared with the original image data by sampling the number of pixels in the main sub both directions (1/2) because it is a ^N, each coefficient image data has a low enough frequency bands N is large, the original image The data represents low-frequency components among the frequency components of the data.

Therefore, the coefficient image data WW _i (i
= 0 to N, the same applies to the following), which represents a change in the frequency of the original image data Sorg in both the main and sub directions. The coefficient image data WV _i are those showing a change in frequency of the main scanning direction of the image signal Sorg, i becomes as low-frequency signals and larger. The frequency in the main scanning direction is lower than the frequency in the sub-scanning direction. As the coefficient image data VW _i is represents a change in the sub-scanning direction of the frequency of the image signal Sorg,
The larger the value of i, the lower the frequency of the signal, and the frequency in the sub-scanning direction is lower than the frequency in the main scanning direction.

FIG. 6 is a diagram showing coefficient image data for each of a plurality of frequency bands. Note that FIG. 6 shows the state up to the state where the third wavelet transform is performed for convenience. In FIG. 6, the coefficient image data WW
Reference numeral ₃ denotes an original image in which the main and sub directions are reduced to (1/2) ³ .

By increasing the filter sizes of the functions g _N and h _N as shown in Table 1 as the number of wavelet transforms increases, aliasing that occurs when sampling pixels of coefficient image data is reduced. can do. For example, a function h as shown in FIG.
_{In 10} , the filter size is 20, and the frequency emphasis characteristic is as shown in FIG. 5 (b). That is, in the function h ₁₀ is the Nyquist frequency 0.5 or more enhancement degree is smaller by comparing the degree of enhancement function h ₂ shown in Figure 4 (b). Therefore, high frequency components mixed with low frequency components having a Nyquist frequency of 0.5 or less by sampling pixels are reduced, and aliasing is reduced.

On the other hand, when the filter size of the basic wavelet function is increased, the amount of computation for performing filtering is increased and the processing time is prolonged. Since it is smaller than the coefficient signal of the band, the delay of the compression processing due to the longer processing time does not matter.

Next, the wavelet re-transformed coefficient signals WV _i , VW which have been subjected to the wavelet transform again
_i (i = 1 is _excluded), WW i, quantization is performed for VV _i.

Here, among the coefficient image data, the coefficient image data in the high frequency band carries unnecessary information such as noise, and the coefficient image data in the low frequency band includes important information such as a main subject. Therefore, the higher the frequency of the coefficient signal, the lower the number of bits in the quantization. That is, as shown in FIG. 6, the coefficient image data WW ₁ , WV ₁ , V
And 0 bits for W _1, coefficient image data WW
₂ is 1 bit, coefficient image data WV ₂ , VW
_The quantization is performed with 2 bits for 2 and with 8 bits for coefficient image data of more than 2.

Here, when quantizing data, the data can be compressed in a state closer to the original image as the number of bits is higher, but the compression ratio cannot be improved so much. Further, the compression ratio can be improved by reducing the number of bits, but the error when the compressed data is restored is large, and the noise is larger than that of the original image.

Therefore, in the present invention, a small number of bits are assigned to image data in a high frequency band carrying a large amount of noise components, and a large number of bits are assigned to image data in a low frequency band carrying information of a main subject. Therefore, the higher the number of bits in the important part, the higher the number of bits to maintain the image quality, and the less important parts, because the image quality is not so important, the number of bits is reduced, and the overall image quality of the main part of the image is compressed while maintaining The rate is improved.

After quantizing each coefficient image data in this way, the above-described encoding such as Huffman encoding and predictive encoding is performed to perform compression processing.

Although the quantization level has been described as being fixed for each label, the quantization level may be changed for each frequency band. Decrease the number. In addition, the number of bits may be set to 0 as the quantization level. In this case, the code length becomes 0, so that a high compression rate can be realized.

The original image data Sorg thus encoded and compressed is stored on a recording medium such as an optical disk, for example, and is stored and transported.

Next, a method for reconstructing compressed data will be described.

Firstly, with respect to the compressed original image data, by performing decoding for the Huffman coding or predictive coding, each coefficient image data WV i described _above, VW _i,
Get WW _i .

[0058] Then, the coefficient image data _WV i obtained by decoding is _performed, VW i, _WW i, V
Performs inverse wavelet transform on V _i.

FIG. 7 is a diagram showing details of the inverse wavelet transform.

As shown in FIG. 7, first, a process is performed on each of the coefficient image data VV _N , VW _N , WV _N , and WW _N with an interval of one pixel between pixels arranged in the sub-scanning direction (× 2 in FIG. 7). Display). Then function h described above coefficient image data VV _N this spaced in the sub-scanning direction
_The coefficient image data V is calculated by a function hN ′ different from _N
The W _N performs a filtering process with a different function g _{N 'is} a function g _N described above in the sub-scanning direction. That is, the coefficient image data VV based on the functions g _N ′ and h _N ′
_N and VW _N are subjected to filtering processing for each pixel in a row arranged in the sub-scanning direction while shifting one pixel at a time in the main scanning direction to obtain inverse coefficient image data of the coefficient image data VV _N and VW _N , and doubling this. Then, the inverse coefficient image data WhN 'is obtained.

Here, the relationship between the functions g _N , h _N and the functions g _N ′, h _N ′ is expressed by the following equation (3).

H _N ′ [n] = h _N [−n] g _N ′ [n] = g _N [−n] (3) where [−n] represents a left-right half-turn with respect to the center axis of the function. , Functions g _N and h _N perform inverse wavelet transform using functions g _N ′ and h _N ′ that are asymmetrical about the center axis.

On the other hand, in parallel with this, the coefficient image data W
V _N is filtered in the sub-scanning direction by a function h _N ′, and the coefficient image data WW _N is filtered in the sub-scanning direction by a function g _N ′, so that the coefficient image data WW _N , WW
_The inverse coefficient image data WgN 'is obtained by obtaining the inverse coefficient image data of _N , doubling it and adding it.

Next, the inverse coefficient image data WhN ', WgN'
Is performed with an interval of one pixel between pixels arranged in the main scanning direction. Then the 'function h _N a in the main scanning _direction' inverted coefficient image data WhN, inverse coefficient image data WGN '
Is filtered in the main scanning direction by a function g _N ′ to obtain inverse coefficient image data of the coefficient image data WhN ′ and WgN ′, and by doubling and adding the result, the inverse coefficient image data VV _N−1 ′ is obtained. obtain.

Next, the inverse coefficient image data V
V _N−1 ′, coefficient image data VW _N−1 , WV
_{For N−1} and WW _N−1 , processing is performed to leave an interval of one pixel between pixels arranged in the sub _- scanning direction. After that, the inverse coefficient image data VV _N-1 '
_{With N−1} ′, filtering processing is performed on the coefficient image data VW _N−1 in the sub _- scanning direction with a function g _N−1 ′. That is, for each row of pixels in the sub _- scanning direction of the coefficient image data VV _N-1 ′ and VW _N−1 by the functions g _N−1 ′ and h _N−1 ′ having a filter size smaller than the functions h _N and g _N. Is performed while shifting pixel by pixel in the main scanning direction to obtain coefficient image data VV.
_N-1 ', VW The inverse coefficient image data WhN is obtained by obtaining the inverse coefficient image data of _N-1 and doubling and adding it.
Get -1 '.

On the other hand, in parallel with this, the coefficient image data W
The function _{h N-1} 'to V _N-1 in the sub-function coefficient image data _{WW N-1} in the sub-scanning direction g
_N-1 to obtain a _'performs filtering processing by, give inverse coefficient image data of the coefficient image data WV _{N-1, WW N-1,} the inverse coefficient image data WGN-1 by adding to 2 times this' .

Next, the inverse coefficient image data WhN-1 ', WgN
With respect to -1 ', processing is performed to leave an interval of one pixel between pixels arranged in the main scanning direction. Then, inverse coefficient image data WhN-1 ′
Is filtered in the main scanning direction by a function h _N-1 ′, and the inverse coefficient image data WgN−1 ′ is filtered in a main scanning direction by a function g _N−1 ′ to obtain coefficient image data WhN-1 ′ and WgN−.
The inverse coefficient image data VV _N−2 ′ is obtained by obtaining the inverse coefficient image data of 1 ′, doubling and adding the result.

Hereinafter, the inverse coefficient image data VV _i '
(I = −1 to N), and finally inverse coefficient image data VV ₋₁ ′ is obtained. This final inverse coefficient image data VV
_-1 'is image data representing the original image data Sorg.

The coefficient image data V thus obtained
V ₋₁ ′ is sent to an image reproducing device (not shown) and is used for reproducing a radiation image.

This reproducing apparatus may be a display means such as a CRT or a recording apparatus for performing optical scanning recording on a photosensitive film.

In this way, the original image data Sorg is subjected to wavelet transform to obtain image data for each of a plurality of frequency bands, and of this image data, data of the high frequency band is again subjected to wavelet transform to obtain a plurality of frequency bands. Of the important part of the data is quantized by increasing the number of bits, and the non-essential part is quantized by decreasing the number of bits. The data compression ratio can be improved while maintaining the image quality.

Here, each coefficient image data VV _N , VW
_N, WV _N, an interval of one pixel among the pixels for WW _N, the function g _{N ',} h _N' but data is formed for the portion of the pixel spaced by performing an inverse wavelet transform by However, since this is restored and formed by extracting the high-frequency component mixed with the low-frequency component of the image data by the aliasing described above, the data is different from the data of the original image, that is, Artifacts are occurring. However, for coefficient image data in the low frequency band,
Since the aliasing is small and the artifact is also small, an image with little artifact can be obtained for image data in a low frequency band that carries important information.

In the above-described embodiment, the functions g _N and h _N for performing the wavelet transform are shown in Table 1 below.
Is used, but the present invention is not limited to this, and any function may be used as long as the coefficient image data in the lower frequency band has a larger filter size.

In addition, any other function that can perform the wavelet transform may be used. For example, a bi-orthogonal function instead of an orthogonal function may be used.

Further, as shown in Table 1, FIG. 4 and FIG. 5, the wavelet transform may be performed using not only a function which is not symmetrical about the central axis but also a function which is symmetrical about the central axis. It is. When the wavelet transform is performed using the symmetric function as described above, the function that performs the wavelet transform and the function that performs the inverse wavelet transform have the same shape.

In the above-described embodiment, the embodiment in which the original image data representing the radiation image is compressed has been described. However, the image compression processing method according to the present invention can be applied to a normal image. .

For example, an embodiment for compressing an image of a 35 mm negative film in which a person or the like is recorded as a main subject will be described. First, the negative film is read by a digital scanner, and image data representing the image is obtained. As described above, the wavelet transform is performed by performing the filtering process using the functions g _N and h _N having the larger filter size in the lower frequency band.

Next, as in the above-described embodiment, the image is quantized by using a low bit number for the high frequency band portion and a high bit number for the low frequency band portion, and is coded as necessary. Compress data.

The original image data can be reconstructed by decoding the compressed image data in the same manner as in the above-described embodiment and further performing inverse wavelet transform.

As described above, by performing the compression processing,
While maintaining the image quality of important parts even for normal images,
The data compression ratio can be improved.

In the above embodiment, when quantizing the coefficient image data, the number of bits is set to 0 for the coefficient signal in the high frequency band. However, the present invention is not limited to this. Any number of bits may be used as long as they are lower than the number of bits used when quantizing the coefficient signal.

[0082]

As described above in detail, the image data compression processing method according to the present invention, when obtaining coefficient image data for each of a plurality of frequency bands by wavelet transform, aliases the coefficient image data in a lower frequency band. The wavelet is applied using a function having a large filter size to reduce the wavelet as a basic wavelet function. For this reason, the occurrence of aliasing decreases in coefficient image data in a low frequency band that carries important information, and it is possible to reduce artifacts caused by aliasing when reconstructing this image data.

For image data in a high frequency band where the number of data to be filtered is large, artifacts are generated but they do not carry very important information. Therefore, the filtering speed is improved by a frequency having a small filter size. Things. Therefore, by quantizing the image data of the high frequency band with a smaller number of bits than the image data of the low frequency band, the image data can be obtained in a good image without artifacts while improving the data compression ratio. Can be compressed, and the time for performing the compression process can be shortened.

[Brief description of the drawings]

FIG. 1 is a diagram showing a basic concept of an image data compression processing method according to the present invention.

FIG. 2 is a diagram showing a method of reading image data used in the present invention.

FIG. 3 is a diagram showing details of a wavelet transform.

FIG. 4 is a diagram showing a basic wavelet function with a small filter size.

FIG. 5 is a diagram showing a basic wavelet function having a large filter size.

FIG. 6 is a diagram showing coefficient image data.

FIG. 7 is a diagram showing details of an inverse wavelet transform;

FIG. 8 is a diagram showing a basic wavelet function used for wavelet transform.

FIG. 9 is a diagram illustrating a wavelet transform.

FIG. 10 is a diagram for explaining a Fourier transform;

[Explanation of symbols]

10 stimulable phosphor sheet h, h ', g, g ' function for performing a wavelet transform _{VV i, VW i, WV i} , WW i (i = 1~
n) Coefficient image data

──────────────────────────────────────────────────続 き Continuation of the front page (56) References JP-A-63-136890 (JP, A) Marc Antonini et al, Image Coding Using Wavelet Transform, IEEE Transactions on Image Processing, Image Processing. 1, A pili 1992, p. 205-p. 220 (58) Field surveyed (Int.Cl. ⁷ , DB name) H04N ^7/ 24-7/68

Claims (2)

(57) [Claims] 1. A method in which original image data representing an image is sequentially subjected to wavelet transform while being sampled at predetermined intervals, so that the original image data is converted into a plurality of coefficient image data representing different frequency bands from a high frequency band to a low frequency band. The plurality of coefficient image data are quantized with a smaller number of bits than the coefficient image data of the other frequency bands at least for the coefficient image data of the highest frequency band, and the quantized coefficient image data is encoded. In the image data compression processing method of removing noise in high frequency components of the original image data and compressing the original image data, a filter that reduces aliasing as the frequency band of the coefficient image data is lower A wavelet function is used as a basic wavelet function. Image data compression processing method characterized by applying a conversion. 2. An inverse wavelet transform in which the encoded coefficient image data is decoded, and a function with a larger filter size for reducing aliasing as the frequency band of the decoded coefficient image data is lower is used as a basic wavelet function. 2. An image data reconstruction method, wherein the original image data compressed by the image data compression processing method according to claim 1 is reconstructed.