JPH09307898A - Image data compression processing method - Google Patents

Abstract

PROBLEM TO BE SOLVED: To compress image data at a high compression rate without deteriorating image quality of an important part of an original image so as to keep high image quality in the image data compression processing method. SOLUTION: Weblet transformation 2 is applied to original image data 1 representing an original image to obtain image data 3 for each of a plurality of frequency bands. Then histogram analysis is applied to the image data 3, importance is recognized (4) in the image based on the analysis result, and labelling processing 5 is applied to data whose importance is recognized depending on the importance and an quantization 6 is conducted to higher important part based on the result of the labelling processing 5. Coding 7 is applied to the image data 3 to which quantization 6 is applied.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to an image data compression processing method, and more particularly, to an image data compression processing method capable of obtaining a high data compression rate by using multi-resolution conversion.

[0002]

2. Description of the Related Art Since an image signal carrying a halftone image such as a TV signal has an enormous amount of information, a wide band transmission line is required for its 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. Further, recently, for example, recording of a halftone image on an optical disk or a magnetic disk has been widely performed, and in this case, image data compression is widely applied for the purpose of efficiently recording an image signal on a recording medium. ing.

As one of such image data compression processing methods, conventionally, when image data is stored or transmitted, the image data is compressed by predictive coding to reduce the amount of data. The compressed image data (compressed image data) is subjected to decoding processing and expanded when the image is played back, and then stored, transmitted, etc., and based on the expanded image data (expanded image data). A method of reproducing a visible image is adopted.

A method utilizing vector quantization is known as one of image data compression processing methods. This method divides the two-dimensional image data into blocks of K samples and predefines K vector elements to create a codebook composed of a plurality of different vectors. The vector corresponding to the set of image data and the minimum distortion is selected, and the information indicating the selected vector is encoded in association with each block.

Since the image data in the blocks as described above have a high correlation with each other, one of the vectors in which a relatively small number of image data in each block are prepared is prepared.
It is possible to give a fairly accurate representation by using one. Therefore, transmission or recording of image data can be performed by transmitting or storing a code indicating this vector instead of actual data, so that data compression is realized. For example, the amount of image data for 64 pixels in a halftone image of a 256-level (= 8 bit) density scale is 8 × 64 = 512 bits, and these 64 pixels are regarded as one block, and each image data in the block has 64 elements. If a codebook is prepared in which 256 types of such vectors are prepared, the data amount per block is the data amount for vector identification, that is, 8 bits, and the data amount is 8 / ( 8 x 6
4) = 1/64 can be compressed.

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 reconstructed data for each block,
The original image is reproduced by using the reconstructed data.

Further, as one of the methods for improving the compression rate in the case of performing the data compression by the above-mentioned predictive coding,
It is conceivable to reduce the bit resolution (density resolution) of the image data together with the predictive coding process, that is, perform the quantization process of coarsely quantizing the image data.

Therefore, the applicant of the present invention has proposed an image data compression processing method by interpolation coding, which is a combination of the above-described prediction coding method and quantization method (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 interpolation prediction coding processing based on the main data, that is, interpolation data is the main data. The image data is compressed by performing interpolative prediction based on the above, and performing variable length coding (conversion into a signal whose code length changes depending on the value) such as Huffman coding with respect to the prediction error.

Further, when compressing image data, it is naturally desirable that the compression rate is high. However, it is technically difficult to expect a large improvement in the compression rate in the above-mentioned interpolation coding, and therefore, in order to achieve a higher compression rate, the image data number reduction processing for reducing the spatial resolution is combined with the above-mentioned interpolation coding. It is possible to match.

Therefore, the applicant of the present invention has proposed an image data compression processing method capable of achieving a higher compression rate while maintaining high image quality by combining the above-described interpolation coding and the image data number reduction processing (special feature). Kaihei 2-280462).

On the other hand, as a method for processing the above-described image data, the image is converted into a multi-resolution image for each of a plurality of frequency bands, a predetermined process is performed on the image in each frequency band, and the image is again processed. A method called multi-resolution conversion for obtaining a final processed image by performing inverse multi-resolution conversion has been proposed. Known methods for this multi-resolution conversion are wavelet conversion, Laplacian pyramid, Fourier conversion and the like.

Here, the wavelet transform will be described.

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

This wavelet transform uses a function h as shown in FIG. 8 as a basis function.

[0015]

[Equation 1]

Since the signal is converted into a frequency signal for each of a plurality of frequency bands in the equation, the problem of false vibration unlike the Fourier transform does not occur. That is, if the filtering process is performed by changing the period and reduction ratio of the function h and moving the original signal, it is possible to create a frequency signal adapted to a desired frequency from a fine frequency to a coarse frequency. For example, as shown in FIG. 9, the signal Sorg
Is wavelet transformed and inverse wavelet transformed for each frequency band, and the signal Sorg is Fourier transformed as shown in FIG. 10 and the signal is inverse Fourier transformed for each frequency band. Compared with the conversion, it is possible to obtain the frequency signal in the frequency band corresponding to the vibration of the original signal Sorg. That is, in the Fourier transform, the frequency band 7 corresponding to the part B of the original signal Sorg
While the vibration is generated in the portion B'of, the wavelet transform does not generate the vibration in the portion A'of the frequency band W7 corresponding to the portion A of the original signal Sorg like the original signal. Become.

Further, using this wavelet transform,
A method for compressing the above-mentioned image data 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).

According to this method, original image data representing an image is subjected to wavelet transformation to convert the original image data into image data of a plurality of frequency bands, and each image data has a high frequency which carries a lot of noise components. Image data in a band has a small number of bits, and image data in a low frequency band carrying information on a main subject is assigned a large number of bits to perform the above-mentioned vector quantization to compress the original image data. Is. According to this method, the compression rate of the original image data can be improved, and the original image can be completely restored by performing the inverse wavelet transform on the compressed image data.

On the other hand, the method of Laplacian pyramid is described, for example, in Japanese Patent Laid-Open Nos. 5-244508 and 6-301766. After the mask processing is performed, the image is sub-sampled and the number of pixels is thinned to halve it to obtain a blurred image having a size of ¼ of the original image. Interpolate the pixels of to return to the original size image,
This image is further masked by the mask described above to obtain a blurred image, and this blurred image is subtracted from the original image to obtain a detailed image representing a predetermined frequency band of the original image. By repeating this process for the blurred image obtained, N blurred images each having a size of 1/2 ^2N of the original image are created. Here, since an image masked by a mask similar to a Gaussian function is sampled, a Gaussian filter is actually used, but the same processing as when a Laplacian filter is applied is performed. A finished image is obtained. Since an image in the low frequency band having a size of 1/2 ^2N is sequentially obtained from the image of the original image size in this manner, the image obtained as a result of this processing is called a Laplacian pyramid.

Regarding this Laplacian pyramid, Burt PJ, "Fast Filter Transforms for Image
Processing ”, Computer Graphics and Image Process
ing Vol. 16, pp. 20-51, 1981; Crowley JL, Stern R.
M., “Fast Computation of the Difference of Low ・ Pa
ss Transform ”IEEETrans.on Pattern Analysis and Mac
hine Intelligence, Volume 6, Issue 2, March 1984, Mallat
SG, “A Theory for Multiresolution Signal Decompos
ition ； The Wavelet Representation ”IEEETrans.on P
attern Analysis and Machine Intelligence, Volume 11,
No. 7, July 1989; Ebrahimi T., Kunt M., “Image comp.
ression by Gabor Expansion ”, Optical Engineering,
Volume 30, Issue 7, pages 873-880, July 1991, and Pieter.
Vuylsteke, Emile Schoeters, “Multiscale Image Con
trast Amplification ”SPIEVol.2167 Image Processin
g (1994), pp551-560, for details.

By the way, in the method of compressing image data using the multi-resolution conversion as described above, if the compression rate is further improved, the image quality of the original image may be deteriorated and the image quality may be improved. There was a limit to the compression ratio. On the other hand, in the case of quantizing image data, if the quantization width at the time of quantization is made smaller, the data compression rate will decrease, but since it can be compressed in a state closer to the original image, the reconstructed image The deterioration of the image quality of is reduced. On the other hand, if the quantization width is made coarse, there will be a large error when decompressing the compressed image data, and this error will appear in the image as noise when the image is decompressed. Although the deterioration is large, the entropy at the time of encoding becomes small, so that the data compression rate can be improved.

Therefore, the applicant of the present invention recognizes the importance of each part of the image in the image data decomposed into a plurality of frequency bands by the wavelet transform, labels the image according to the importance, and determines the importance. An image data compression method has been proposed in which a high quantization area is quantized with a fine quantization width, and a low importance area is quantized with a coarser quantization width as compared with a high importance area. Kaihei 6-350989 publication). According to this method, the image data can be compressed while maintaining the image quality for the important part of each part in the image, and the image data can be compressed at a higher compression rate for the unimportant part. be able to. Therefore, the compression rate of the image data can be improved without degrading the image quality of an important part of the image.

That is, in the quantization of the high frequency component, the degree of importance is obtained from the original image or the low frequency image, and the quantization width is changed according to the obtained degree of importance. In addition, high-frequency component encoding is performed using one Huffman table, and high-frequency component decoding is performed using one Huffman table, similar to encoding. In the inverse quantization, since the quantization table is not switched at the time of the quantization, the inverse wavelet transform can be performed as it is. Therefore, it is possible to perform high-speed decoding without performing case classification at the time of decoding.

[0024]

However, in the image compression using the non-equal length code, the capacity of the compressed data cannot be determined until the processing is completed, and therefore, it is predetermined in the means for storing the data during the compression processing. The maximum value may be reached. When compressing normal image data, for example, the upper leftmost data of the image is set as a compression start point, and the data is sequentially compressed from this point. Therefore, when the upper limit of the compressed data is set, if the data is compressed in order from the compression start point, the image data will reach the upper limit in the middle of the image, and the data will be compressed for important parts in the image. It may not be possible to do it. In such a case, the compressed data will lack important part information, and even if this data is restored, the restored part lacks important parts, so it is meaningless as an image. It will not exist.

In view of the above circumstances, it is an object of the present invention to provide an image data compression processing method capable of compressing image data without losing important parts in the image.

[0026]

An image data compression processing method according to the present invention is an image data compression processing method for performing compression processing on original image data representing an image including a predetermined subject, wherein multiresolution conversion is performed on the original image data. By performing the above, the original image data is decomposed into image data for each of a plurality of frequency bands, the importance of each part of the image is recognized based on the image data of a relatively low frequency band, and each image data is At the time of compression, the image data in a relatively high frequency band is quantized in order from the above-mentioned respective parts having the greater importance, and the quantized image data is encoded. is there.

In the image data reconstructing method according to the present invention, image data in a relatively low frequency band of the encoded image data is decoded, and the image is reproduced based on the decoded image data. Of the encoded image data, the encoded image data are decoded in descending order of importance, and the decoded image data are inversely quantized and then inverse multiresolution converted. Thus, the original image data compressed by the image data compression processing method according to the present invention is reconstructed.

[0028]

The image data compression processing method according to the present invention recognizes the importance of each part of an image in image data decomposed into a plurality of frequency bands by multi-resolution conversion and compresses the data in the image when data compression is performed. Image data in a relatively high frequency band having detailed information is disclosed in the above-mentioned JP-A-6-
Unlike No. 350989, quantization and coding are performed in the order of importance. Therefore, even if the upper limit data amount of the compressed image data is set, the compression is performed in order from the most important portion. Will never be missing. Therefore, the image obtained when the compressed image data is decompressed will have complete information about the important parts.
In addition, the maximum data amount may be reached during the compression of image data, but the information that is missing in this case is relatively unimportant, so the restored image will completely cover the parts of high importance. Therefore, it is possible to maintain a high image quality for a portion having high importance. Therefore, although the information of the entire image is small, it is possible to perform compression so that the restored image has a high image quality in an important part of the image. Further, the compressed data can be recompressed by cutting off the encoded data halfway.

Further, in the image data compression processing method according to the present invention, when the image data is restored, the image data in the low frequency band is first restored, and the importance of the image is obtained in the image data in the low frequency band, Image data in a relatively high frequency band compressed in the order of importance is restored. Therefore, it is not necessary to store the information regarding the position of the importance at the time of compression, and thus the image data can be compressed at a high compression rate.

[0030]

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

FIG. 1 is a diagram showing the basic concept of the 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 performs wavelet transform 2 on original image data 1 representing an original image to obtain image data 3 for each of a plurality of frequency bands. Next, based on each image data 3, recognition 4 of importance of each part of the image is performed, and labeling processing 5 is applied to the image data 3 for which importance 4 has been recognized according to this importance. Then, based on the result of the labeling process 5, the image data in the relatively high frequency band that includes even the detailed information in the image is quantized 6 from the order of importance, and the quantized image 6 is obtained. The encoding 7 is performed on the data 3.

The details of the embodiment according to the present invention will be described below.

This embodiment is a storage system for a radiation image information recording / reproducing system using a storage 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 phosphor sheet as digital image data by laser beam scanning. In addition, as shown in FIG. 2, the radiation image is read by the stimulable phosphor sheet 10.
On the other hand, the sheet 10 is two-dimensionally scanned by moving the sheet 10 in the sub-scanning direction (vertical direction) while scanning the laser beam in the main scanning direction (horizontal direction).

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

FIG. 3 is a diagram showing details of the wavelet transform for 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, which is described in the above-mentioned document of Marc Antonini et al.

As shown in FIG. 3, the filtering process is performed in the main scanning direction of the original image data Sorg by the function g and the function h obtained from the basic wavelet function. That is, the filtering process for each column of pixels arranged in the main scanning direction by such functions g and h is performed while shifting each pixel in the sub scanning direction to obtain the wavelet transform coefficient signal Wg0 of the original image data Sorg in the main scanning direction. This is for obtaining Wh0.

Here, the functions g and h are uniquely obtained from the basic wavelet function. For example, the function h
Is shown in Table 1 below. It should be noted that in Table 1, the function h'represents a function used when performing inverse wavelet transform on image data that has been subjected to wavelet transform. Further, as shown in the following expression (2), the function g is obtained from the function h ', and the function g'for performing the inverse wavelet transform is obtained from the function h.

[0039]

[Table 1]

G ′ = (− 1) ⁿ h g = (− 1) ⁿ h ′ (2) In this way, the wavelet transform coefficient signals Wg 0, W
When h0 is obtained, the wavelet transform coefficient signal Wg0,
For Wh0, pixels in the main scanning direction are thinned out every other pixel, and the number of pixels in the main scanning direction is halved. Then, the wavelet transform coefficient signals Wg0 and Wh0 from which this pixel is thinned out.
Filtering processing is performed in each sub-scanning direction by the functions g and h, and wavelet transform coefficient signals WW ₀ and W
Obtain V ₀ , VW ₀ and VV ₀ .

Next, the wavelet transform coefficient signal W
With respect to W ₀ , WV ₀ , VW ₀ and VV ₀ , the pixels in the sub-scanning direction are thinned out every other pixel, and the number of pixels in the sub-scanning direction is halved. As a result, each wavelet transform coefficient signal VV ₀ , WV ₀ , VW ₀ , WW
The number of pixels of ₀ is 1/4 of the number of pixels of the original image data Sorg. Next, filtering processing is performed by the functions g and h in the main scanning direction of the wavelet transform coefficient signal VV ₀ .

That is, the filtering processing for each column of the pixels lined up in the main scanning direction by the functions g and h is performed while shifting each pixel in the sub scanning direction, and the wavelet transform coefficient signal VV ₀ of the wavelet transform coefficient signal in the main scanning direction is obtained. Wg1 and Wh1 are obtained.

Here, the wavelet transform coefficient signal VV ₀
Since the number of pixels in both main and sub directions is half that of the original image data, the image resolution is half that of the original image data. Therefore, by filtering the wavelet transform coefficient signal VV ₀ with the functions g and h, the wavelet transform coefficient signal representing a frequency component lower than the frequency component represented by the wavelet transform coefficient signal VV ₀ among the frequency components of the original image data. Wg1, Wh1
Is required.

When the wavelet transform coefficient signals Wg1 and Wh1 are obtained in this way, the wavelet transform coefficient signals Wg1 and Wh1 are thinned out every other pixel in the main scanning direction, and the number of pixels in the main scanning direction is further reduced to 1. / 2 Then the wavelet transform factor signals Wg1, Wh1 function g in each of the sub-scanning direction, performs a filtering process by h, the wavelet transform factor signals WW _1, W
Obtain V ₁ , VW ₁ and VV ₁ .

Next, the wavelet transform coefficient signal W
For W ₁ , WV ₁ , VW ₁ , and VV ₁ , pixels in the sub-scanning direction are thinned out every other pixel, and the number of pixels in the sub-scanning direction is halved.
And perform the process. As a result, the number of pixels of each wavelet transform coefficient signal VV ₁ , WV ₁ , VW ₁ , WW ₁ becomes 1/16 of the number of pixels of the original image data Sorg.

Thereafter, in the same manner as described above, filtering processing is performed by the functions g and h in the main scanning direction of the wavelet transform coefficient signal VV ₁ from which pixels are thinned, and the main wavelet transform coefficient signal obtained is further processed. Pixels in the scanning direction are thinned out, and the wavelet transform coefficient signal obtained by thinning out the pixels is filtered in the sub-scanning direction by the functions g and h to obtain the wavelet transform coefficient signal W.
W ₂ , WV ₂ , VW ₂ and VV ₂ are obtained.

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

Therefore, the wavelet transform coefficient signal WW _i (i = 0 to N, the same applies hereinafter) is used as the original image data Sor.
It represents a change in the frequency of g in both main and sub directions, and the larger i is, the lower the frequency of the signal becomes. The wavelet transform coefficient signal WV _i represents the change in the frequency of the image signal Sorg in the main scanning direction, and the larger i is, the lower the frequency signal becomes. Furthermore, the wavelet transform coefficient signal VW _i represents the change in frequency of the image signal Sorg in the sub-scanning direction, and the larger i is, the lower the frequency signal becomes.

FIG. 4 is a diagram showing the wavelet transform coefficient signal for each of a plurality of frequency bands. Note that, in FIG. 4, for convenience, the state up to the third wavelet transform is shown. In FIG. 4, the wavelet transform coefficient signal WW ₃ is (1/1 /
2) It has been reduced to ³ .

Next, the wavelet transform coefficient signal W of the smallest frequency band obtained by performing the wavelet transform N times is performed.
This wavelet transform coefficient signal WW based on W _N
The importance of each part included in the image carried by _N is obtained, and labeling processing is performed according to this importance. First, as shown in FIG. 5, in the case of a human chest radiographic image, the lung field part is the most important, the parts other than the lung field are not so important, and the slippery part is not important. , The wavelet transform coefficient signal WW _{N is} subjected to a histogram analysis or the like to recognize a lung field, a part other than the lung field and a snare part, and the lung field part is 1, the non-lung field part is 2, Labeling processing is performed so that a value is assigned to an important portion in order from the important portion such as 3.

That is, histogram analysis is performed on the wavelet transform coefficient signal WW _{N in} the smallest frequency band obtained by performing the wavelet transform N times, and the importance of each density band in the image is obtained based on the result of this histogram analysis. To be FIG. 6 shows the wavelet transform coefficient signal W.
It is a figure showing the histogram of W _N. In the histogram shown in FIG. 6, a shaded area A is the most important area (in the case of a chest radiographic image as shown in FIG. 5, a lung field).
Area B is the area other than the lung field and is the second most important area. The region C is a region of almost no importance corresponding to the sunken part directly irradiated with the radiation. Then, the density range corresponding to this area is subjected to labeling processing in order from the most important one.

Next, the wavelet transform coefficient signal WV
Quantization is performed on _i , VW _i , and WW _i . Here, each wavelet transform coefficient signal WV _i , VW _i , W
Since the image represented by W _i is a reduced version of the original image, the labeling applied to the wavelet transform coefficient signal WW _N also corresponds to each wavelet transform coefficient signal WV _i , VW _i , WW _i. It has become a thing. Therefore, when quantizing each of the wavelet transform coefficient signals WV _i , VW _i , WW _i , the quantization is performed in order from the most important part in the image according to the result of the labeling.

That is, in the case of the image shown in FIG. 5 described above, the importance of the lung field part is the highest, and the parts other than the lung field and the sinker part become less important in this order. Quantization is done. First, the wavelet transform coefficient signals WV _i , VW _i , and WW _i are raster-scanned, and quantization is performed on the most important part. Then, again, the wavelet transform coefficient signals WV _i , VW _i , W
Raster scan W _i, and quantize the second important part. Then, the wavelet transform coefficient signals WV _i , VW _i , and WW _i are further raster-scanned,
Quantize the least important part. By compressing the data in descending order of importance, even if the upper limit of the compressed data is set, the missing data will be the data of the less important part, Partial data will not be lost.

Here, when the data is quantized, the smaller the quantization width, the more the data can be compressed in a state closer to the original image, but the compression rate cannot be improved so much. Further, if the quantization width is made coarse, the compression rate can be improved, but there is a large error when the compressed data is restored, and there is more noise than the original image.

Therefore, in the present invention, a small number of bits are assigned to the image data in the high frequency band that carries a large amount of noise components, and a large number of bits are assigned to the image data in the low frequency band that carries the information of the main subject. The wavelet transform coefficient signals WV _i , VW _i , and WW _i do not have the same number of bits as a whole, but the number of bits is increased in the more important portions to maintain the image quality, and the image quality in the less important portions does not matter so much. Therefore, it is preferable to reduce the number of bits to improve the compression rate, and to improve the compression rate while maintaining the image quality of the main part of the image as a whole.

After each wavelet transform coefficient signal is quantized as described above, compression processing is performed by performing the above-mentioned Huffman coding, predictive coding and the like.

Although the quantization level has been described as being constant for each label, the quantization level may be changed for each frequency band. For example, in the high frequency band, more quantization bits are used. Reduce the number. Further, the number of bits may be set to 0 as the quantization level, and in this case, the code length is 0, so that a high compression rate can be realized.

The original image data Sorg encoded and compressed in this way is stored in a recording medium such as an optical disk, and is stored and transported.

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

[0060] First, with respect to the compressed original image data, by performing decoding for the Huffman coding or predictive coding, the lowest frequency band each wavelet transform factor signal VV _N described above in, VW _N, WV _N, Get WW _N.

Next, the wavelet transform coefficient signal WW
_{For N} , the importance is recognized by histogram analysis or the like as described above, and labeling processing is performed according to the importance. Then, the wavelet transform coefficient signal WW _N
For a compressed wavelet transform coefficient signal in a higher frequency band than this, by referring to the result of this labeling, the high frequency information stored sequentially can be sequentially read out and made to correspond to the coordinates in the image. . In this way, when the importance of the wavelet transform coefficient signal WW _N in the relatively low frequency band is recognized again and the coordinates of the coded information are obtained, it is important to compress the wavelet transform coefficient signal. It is not necessary to encode information (for example, coordinate values) regarding the position of the highly frequent portion. Therefore, the image can be compressed at a higher compression rate.

[0062] Then, the wavelet transform factor signals VV _N obtained by decoding is performed, VW _N, WV
Inverse wavelet transform is applied to _N and WW _N.

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

[0064] As shown in FIG. 7, each wavelet transform factor signals _{VV N, VW N, WV N} , the WW _N performs processing spacing of one pixel among pixels arranged in the sub-scanning direction (× in FIG. 2). Followed by different function h 'is the function h described above wavelet transform factor signals VV _N this spaced in the sub-function g mentioned above the wavelet transform factor signal VW _N in the sub-scanning direction
Filtering processing is performed by a function g ′ different from.
That is, the function g ', h' wavelet transform factor signals by VV _N, filtering processing for each pixel of a row aligned in the sub-scanning direction VW _N in the main scanning direction is performed while Shifts pixel by pixel, the wavelet transform factor signals VV _N , VW
_The inverse wavelet transform coefficient signal W is obtained by doubling and adding _N inverse wavelet transform coefficient signals.
get hN ′.

The reason why the function for performing the wavelet transform is different from the function for performing the inverse wavelet transform is as follows. Wavelet transform and inverse wavelet transform have the same function,
That is, it is difficult to design orthogonal functions, and it is necessary to loosen any of the conditions of orthogonality, continuity, function shortness, and symmetry. Therefore, a function that satisfies other conditions is selected by relaxing the orthogonality condition.

As described above, in the present embodiment, the functions h and g for performing the wavelet transform and the functions h'and g'for performing the inverse wavelet transform are bi-orthogonally different. Thus, the wavelet transform factor signals VV _i, V
The original image data can be completely restored by inverse wavelet transforming W _i , WV _i , and WW _i with the functions h ′ and g ′.

On the other hand, in parallel with this, the wavelet transform coefficient signal WV _N in the sub-scanning direction is given by the function h ', and the wavelet transform coefficient signal WW _N is given in the sub-scanning direction by the function g'.
The performs filtering process, to obtain an inverse wavelet transform factor signals of the wavelet transform factor signals WV _N, WW _N, obtain the inverse wavelet transform factor signals WGN 'by adding to 2 times this.

Next, the inverse wavelet transform coefficient signal W
With respect to hN 'and WgN', a process of spacing one pixel between pixels arranged in the main scanning direction is performed. Then, the inverse wavelet transform coefficient signal WhN 'is filtered in the main scanning direction by the function h', and the inverse wavelet transform coefficient signal WgN 'is filtered in the main scanning direction by the function g', and the inverse wavelet of the wavelet transform coefficient signals WhN 'and WgN' is obtained. An inverse wavelet transform coefficient signal VVN _-1 'is obtained by obtaining a transform coefficient signal, doubling it, and adding it.

Next, the inverse wavelet transform coefficient signal VVN _-1 'and the wavelet transform coefficient signals VWN _-1 , W
1 between pixels arranged in the sub _- scanning direction for V _N-1 and WW _N- 1
Performs a process for spacing pixels. After that, the inverse wavelet transform coefficient signal VVN _-1 'is processed in the sub-scanning direction by the above-mentioned function h'.
_N-1 is filtered in the sub _- scanning direction by the function g'described above. That is, the filtering processing is performed for each row of pixels of the wavelet transform coefficient signals VVN _-1 ', VWN _-1 by the functions g', h ', which are arranged in the sub _- scanning direction, while shifting each pixel in the main-scanning direction. transform factor signals VV _N-1 to obtain a ', to give the inverse wavelet transform factor signal VW _N-1, the inverse wavelet transform factor signals WhN-1 by adding to 2 times this'.

On the other hand, in parallel with this, the wavelet transform coefficient signal WV _N-1 is filtered in the sub _- scanning direction by the function h ', and the wavelet transform coefficient signal WW _N-1 is filtered in the sub _- scanning direction by the function g'. , to obtain a wavelet transform factor signals WV _N-1, the inverse wavelet transform factor signal WW _N-1, to obtain an inverse wavelet transform factor signals WGN-1 'by adding to 2 times this.

Next, the inverse wavelet transform coefficient signal W
1 between pixels lined up in the main scanning direction for hN-1 'and WgN-1'
Performs a process for spacing pixels. Thereafter, the inverse wavelet transform coefficient signal WhN-1 'is filtered in the main scanning direction by the function h', and the inverse wavelet transform coefficient signal WgN-1 'is filtered in the main scanning direction by the function g', and the wavelet transform coefficient signal WhN-1 'is obtained. , WGN-1 to obtain a 'give inverse wavelet transform factor signals, inverse wavelet transform factor signal VV _N-2 by adding to 2 times this'.

Thereafter, the inverse wavelet transform coefficient signal VV _i ′ (i = −1 to N) is sequentially created, and finally the inverse wavelet transform coefficient signal VV ₋₁ ′ is obtained. This final inverse wavelet transform coefficient signal VV ₋₁ ′ is the original image data S
Image data representing org.

The wavelet transform coefficient signal VV _-1 ′ thus obtained is sent to an image reproducing device (not shown) for reproduction of a radiation image.

The reproducing device may be a display device such as a CRT or a recording device for performing optical scanning recording on the photosensitive film.

Here, in the reproduced image, when the data is compressed, the information that is missing when the maximum data amount is reached is the information that is not so important, and the information that is highly important remains completely. It is what Therefore, the information about the entire image is missing,
Since no information is lost in the portion of the reproduced image having high importance, high image quality can be maintained.
Therefore, even with a small amount of information, it is possible to perform compression to obtain a high-quality restored image for an important part of the image. In addition, the compressed data can be recompressed by stopping the encoded data halfway. Furthermore, for applications in which compressed data is stored as a temporal file on the hard disk for a certain period of time, and then it is possible to increase the compression rate and record it on an optical disk, it is sufficient to record the data truncated from the beginning up to a certain code amount. , It becomes unnecessary to combine compressed data and further compress it by another method.

In comparison with the above-mentioned JP-A-6-350989,
The present invention is different in the following points. That is, in the present invention, in the encoding of the high frequency component, the encoding is performed in order from the one having the highest importance, and in the decoding of the high frequency component, the low frequency image is decoded to obtain the importance, and the importance is determined. The reference design high frequency component is sequentially read and decoded. In inverse quantization, if the quantization table is switched and quantization is performed, high frequency components are created while switching the quantization table. If the quantization table is not switched, the inverse wavelet transform is performed as it is. To do.

In the above-described embodiment, the functions h and h'for performing the wavelet transform are shown in Table 1.
However, the present invention is not limited to this, and those shown in Tables 2 and 3 below may be used.

[0078]

[Table 2]

[0079]

[Table 3]

In addition to this, any function may be used as long as it can perform wavelet transformation, and for example, a function which is not symmetric but orthogonal but not symmetrical may be used.

Further, as shown in Tables 1, 2, and 3, n
Not only the function symmetrical about the axis of = 0, but n = 0
The wavelet transform may be performed using a function that is asymmetric with respect to the axis of. When the wavelet transform is performed using the left-right asymmetric function as described above, the inverse wavelet transform is performed by using the function obtained by inverting the left-right function of the wavelet-transformed function with respect to the axis of n = 0. That is, with respect to the left-right asymmetric functions g and h, the functions g ′ and h ′ that perform the inverse wavelet transform are g [n] = g ′ [− n] h [n] = h ′ [− n] (3) However, [-n] represents left-right inversion.

Is obtained.

In the above-described embodiment, labeling is performed according to the importance obtained as a result of the histogram analysis, but the image data is divided into a plurality of density bands based on the result of the histogram analysis. Then, the importance level may be determined for each density band, and when scanning the density band having the high importance level during the raster scan for quantization, the quantization may be performed in order from the density band having the high importance level.

Further, in the above-mentioned embodiment, the image is converted into the multi-resolution image by the wavelet transform, but the present invention is not limited to this, and the image is converted by the above-mentioned Laplacian pyramid method or Fourier transform. The resolution may be converted into multiple resolutions.

Furthermore, in the above-mentioned embodiment,
Although the embodiment of compressing the original image data representing the radiation image has been described, the image compression method according to the present invention can be applied to a normal image.

For example, a description will be given of an embodiment in which an image of a 35 mm negative film in which a person or the like is recorded as a main subject is compressed. First, the negative film is read by a digital scanner, image data representing this image is obtained, and this image is obtained. Wavelet transformation is performed by filtering the data with the functions g and h as described above. Next, the wavelet transform coefficient signal obtained by performing the wavelet transform is subjected to circular pattern matching or filtering by a skin color filter to recognize a human face. Here, for an image including a person, the information of the face portion is the most important and the information of the background portion other than this is unnecessary, so that the face portion is 1 and the other portions are 2 Perform labeling processing.

Next, based on the result of this labeling processing, the image data is compressed by performing quantization in order from the face portion and then performing encoding.

Further, the original image data can be reconstructed by decoding the compressed image data in the same manner as in the above-mentioned embodiment, further performing the inverse quantization and then performing the inverse wavelet transform.

By performing the compression process in this way,
As for an important part of a normal image, the data compression rate can be improved while maintaining the image quality without losing information.

Further, in the above-described embodiment, the importance of each part of the image is recognized based on the wavelet transform coefficient signal having the lowest frequency band. However, in each wavelet transform coefficient signal and original image data, Since all the positions of the images carried by each data correspond, the importance of each part of the image is recognized based on the original image data or the coefficient signal other than the wavelet transform coefficient signal with the lowest frequency band. Is also good.

[Brief description of 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 wavelet transform coefficient signal.

FIG. 5 is a diagram showing a radiation image.

FIG. 6 is a diagram showing a histogram of a wavelet transform coefficient signal WW _N.

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

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

FIG. 9 is a diagram for explaining 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 VV _i for performing a wavelet _{transform, VW i, WV i, WW} i (i = 1~n) wavelet transform factor signals

Claims (2)

[Claims] 1. An image data compression processing method for performing compression processing on original image data representing an image including a predetermined subject, wherein multiresolution conversion is performed on the original image data,
The original image data is decomposed into image data for each of a plurality of frequency bands, the importance of each part of the image is recognized based on the image data in a relatively low frequency band, and when the image data is compressed, a comparison is made. The image data compression processing method is characterized in that the image data in a relatively high frequency band is quantized in order from the above-mentioned respective parts having the highest importance, and the quantized image data is encoded. 2. The image data in a relatively low frequency band of the encoded image data is decoded, the importance of the image is recognized based on the decoded image data, and the encoding is performed. The image data compression processing method according to claim 1, wherein the decoded image data are decoded in descending order of importance, the decoded image data are dequantized, and inverse multiresolution conversion is performed. A method for reconstructing image data, comprising reconstructing the original image data compressed by.