JPS63109662A - Picture data compressing method - Google Patents

Abstract

PURPOSE:To eliminate the influence of a peripheral edge to a statistic, by finding the statistic of each component by using part or all of the transformational coefficient matrices of blocks which are not in the peripheral area of a picture in transformational coefficient matrices. CONSTITUTION:Orthogonal transformation is performed on all blocks in a picture matrix A and calculation of statistics of transformational coefficients is carried out to the blocks in another picture matrix B only. Namely, the transformational coefficient matrix alpha(i, j) (where i, j=0, 1, 2,..., m-1) of each block is found by dividing the picture matrix A of M X N picture elements into blocks of m X m picture elements and performing discrete two-dimensional orthogonal transformation to each block. In this case, however, the statistic of each component is calculated on the transformational coefficient matrices alpha(i, j) of blocks which belong to the picture matrix B and, at the same time, to which statistic calculation is performed, only. By owitting the blocks in the peripheral area of film from the calculation of the statistic of each components of transformational coefficient matrices in such way, bad influences of a peripheral edge to a statistic can be eliminated.

Description

DETAILED DESCRIPTION OF THE INVENTION [Field of Industrial Application] The present invention relates to a method of compressing radiation image data by block transform encoding, and in particular, when determining statistics of a set of transform coefficients in order to investigate image characteristics. This invention relates to an image data compression method that calculates statistics using transformation coefficients near the center of an image.

[Background of the invention]

Transform coding using orthogonal transforms such as cosine transform and Hadamard transform is well known as a data compression technique for digital image data with gradation.

This transform encoding is a method in which an image is divided into small blocks of m×m pixels, orthogonal transform is performed for each block to obtain a transform coefficient matrix, and the number of quantization levels is determined for each component of the matrix.

To determine the number of quantization levels, statistics for each component, such as variance and standard deviation values indicating variations in transformation coefficients and numerical data, are used. In the conventional method, all transformation coefficient matrices are subject to statistical calculation.

Incidentally, in medical X-ray film images, there are often non-exposed areas (excluded areas) in the periphery of the film, and the optical density of this area is much lower than that of the exposed areas.

In addition, the optical density of the exposed area adjacent to the non-exposed area is
It is often very large, and therefore the boundary between the non-exposed and exposed parts of the film often forms a sharp edge (hereinafter referred to as a peripheral edge) with a large density difference.

Therefore, even in a digital image obtained by reading an X-ray film image containing these peripheral edges with an image reading device and performing A/D conversion, there is a very high probability that the peripheral edges will be included.

In the transform coefficient matrix obtained by orthogonally transforming a digital X-ray image containing such peripheral edges into blocks, the transform coefficient matrix of the block containing peripheral edges has transform coefficients whose absolute values are larger than those of other blocks. The probability that it will appear is extremely high.

Therefore, when such image data is subjected to data compression using transform encoding, transform coefficients with large absolute values caused by peripheral edges have a very large influence on the above-mentioned statistical calculation, and some This may cause components to be assigned a higher number of quantization levels than necessary, or the compression ratio may decrease.

Furthermore, during medical X-ray photography, since the necessary image information (information on the region of interest) is taken near the center of the film, information on the periphery of the film is often unnecessary. In other words, there are many cases where image information does not exist in the periphery of the film. In data compression of medical images, the purpose of calculating statistics during compression processing is to obtain statistics of the image information part or region of interest. However, it is not desirable to calculate statistics.

Furthermore, if statistics are calculated for all transformation coefficient matrices, the calculation time will be enormous, which will hinder speeding up of the compression process.

[Purpose of the invention]

An object of the present invention is to eliminate the influence of peripheral edges on statistics, obtain more accurate statistics for a region of interest, and speed up compression processing.

[Structure of the invention]

To this end, the present invention provides an image data compression method in which digitized radiographic image data is divided into blocks and orthogonally transformed encoded. In addition, statistics of each component are determined using part or all of the transform coefficient matrix of blocks other than the peripheral part of the image among the transform coefficient matrices obtained by performing orthogonal transformation on each block.

〔Example〕

In FIG. 1, A is an image matrix of digital image data, and the matrix size is MXN pixels. This image matrix A has a matrix size of m x
It is a collection of blocks of m pixels. B is an image matrix that does not include blocks at the periphery of the image, and its matrix size is M'XN' pixels. Here, MSN,
M', N', m are positive integers, k2, k: l, k
For 4, ks (4 for k2≧, k, ≧ks), l m=2 N=, ・m M= k, ・m N'=4 ・m M'=kS −m be satisfied.

In this embodiment, orthogonal transformation is performed on all blocks in image matrix A, and statistics calculation of transformation coefficients is performed only on blocks in image matrix B.

Statistics calculation may be performed for all blocks in image matrix B, but in order to further reduce calculation time, it is also possible to reduce the number of target blocks by thinning out.

FIGS. 2 and 3 show how blocks in image matrix B are thinned out. The lattices in FIGS. 2 and 3 indicate the boundaries of blocks, and block b represented by diagonal lines and block C represented by black are blocks in image matrix B, and blocks that are neither of the above. a is a block outside the image matrix B.

Here, block C is a block in image matrix B that performs statistical calculation, and block b is a block in image matrix B that does not perform statistical calculation. Therefore, in Figure 2, the thinning is 1:4, and in Figure 3, it is 1:2.
This indicates that the data will be thinned out.

In this example, image matrix B is set for image matrix A, statistics are calculated for all blocks in image matrix B or blocks obtained by thinning, and this result is used to Target all blocks for compression.

A detailed explanation will be given below according to the flow shown in FIG.

! Blocking the Image Matrix An image matrix A of MXN pixels is divided into blocks of m x m pixels.

(21, 2i - Orthogonal/reversal processing) A discrete two-dimensional orthogonal transformation (Hadamard transformation, slant transformation, cosine transformation, etc.) is applied to each block, and the transformation coefficient matrix α (i, j) (i ,j=0
．． 1, 2, 3. ..., m-1). Here, α
Classify (i, j) as follows.

α(i, D = αm (i, j): Transform coefficient matrix of blocks that do not belong to image matrix B.

= αb (i, j): Transform coefficient matrix of a block that belongs to the image matrix and does not perform statistical calculation.

=α((x+j) 'Transformation coefficient matrix of a block that belongs to image matrix B and performs statistical calculation.

The statistics of each component are calculated only for the transformation coefficient matrix αc (i, j) for each dish. In this example, the statistics to be calculated are the standard deviation value σc (IIj) when the average value of the conversion coefficients of each component is assumed to be O, and the absolute value P'C of the conversion coefficient that maximizes the absolute value of each component. (IIJ) and their σc (1+
J), P c (1+J), Sc by the following formula
Find (i, j).

SC(11j) = P c (1+j) /σc (
IIj) In this example, the statistics σc (1+J), Pc (1+J) calculated for αc (IIJ)
, Sc (1+J) for all α(IIJ).

Using the two transformation coefficient matrix σc (inD) of the quantization level in Llfl, the transformation coefficient matrix α(i, j
) is the number of quantization levels K(i,
D, determine the number of quantization bits β(i,j). The concentrated region is defined as the number of quantization levels K (a positive constant X uniquely determined by LD) when the conversion coefficient is regarded as a continuous variable Z.
For (K(i, D ), (X l −X (K(i
, j))≦ZIX(K(i,j))). Here, X (K (i, D ) is the judgment level when the distribution of each component of the transformation coefficient matrix α (i, j) is regarded as a standard normal distribution, and the quantization of Max is applied (Zk ) (k=0.1.2.3...., K(I
IJ) ), the output level is (Qk) (k=0.1
．． 2. ...When expressed as K(i, D-1) (however, Zv
t-+ <q+-+<7. kSZo--oo, Z k
(in j) = ” ”), for a certain constant T, 2 Zr-+ -Zx-z≦X (K(i, D)≦
It is a constant that satisfies (γ+1)Zx-+r'Zx-z. This constant γ is preferably a real value satisfying 1≦T≦3.

Next, the number of quantization levels K(i, j),! The number of child bits β(i, j) is calculated using the following formula.

(if(i,j) ≠(0,0) )=C(if(i
, j) = (0,0)) Here, d is a variable that controls the value of the number of quantization bits β(i, j), and increasing this d increases the compression rate, and decreasing it increases the quality of the reproduced image. has the property of becoming closer to the original image.

The function 1nt(X) is a function that truncates the real number X below the decimal point and sets 1nt(x)=0 when X is negative.

C is a constant indicating the number of quantized bits of the DC component, and if the original image is composed of n bits/pixel, usually C>n
Set it so that

e is a constant that takes either 0 or -1, and determines whether the number of output levels in the concentrated area is even or odd.

(5) The quantification level 'Sc (
1, J), the number of quantization levels K' (i, j) and the number of quantization bits β' (IIJ) of the exceptional region in each component of the transform coefficient matrix α (ITJ) are determined. The exceptional region is the following defined by the above-mentioned X (K(i, j)):
(K(i,D)).

In this example, a case will be explained in which −-like quantization is used as a quantization method in an exceptional region.

The number of quantization bits of the concentrated region in each component is β(i, j
), the -like quantization width is expressed as W(β(i,j)
).

w (x,)≧w (xi) (if XI <
Xi) Next, the number of quantization levels K' (Lj),
The number of quantization bits β' (11j) is obtained from the following formula.

β' (IIJ) = Int' (j! Ogg
Here, the function Int'(x) is Int'(
X) =x+l (decimal point of x≠O]=x
(Defined as decimal point of x = 03.

(6), the configuration is corrected. If each component of the normalized transformation coefficient matrix is α' (IIJ), normalization is performed using the following equation.

α' (i, j) = α(LD /σc (IIJ)(
if (i, D ≠ (0, 0) ) (7), transform coefficient matrix α' (LD ((i, j)
≠ (0, 0)), the number of quantization levels K(i, j) (or β(i, j)), the number of quantization levels K' (i,
Quantize and encode with reference to D (or β'(i, j)).

<Example 1 of encoding method> A code sequence in a concentrated area and a code sequence in an exceptional area are determined independently. Each code sequence may be either a fixed length code or a variable length code.

Assume that the number of quantization levels in the concentrated area is an odd number.

The output code sequence in the concentrated region is divided into multiple data codes (codes uniquely determined for the output level) and one flag code (indicating that the transform coefficient to be quantized did not exist in the concentrated region). It consists of the following symbols.

All output code sequences in the exception area are composed of data codes.

If the transform coefficient to be quantized exists within the concentrated area, a data code is output; if the transform coefficient does not exist, a flag code is output.

Only when the transform coefficient to be quantized does not exist within the concentrated region, the data code of the exceptional region is added.

Example 2 of coding method One variable-length code sequence is used for two areas, a concentrated area and an exception area. In this case, a shorter code is assigned to an output level with a high probability of occurrence.

The number of quantization output levels in the concentrated region is an odd number,
Either an even number is fine.

The code series is entirely composed of data codes.

Up to 11 pieces of information necessary for restoring the original image, such as the number of quantization bits β(t+j), β' (i, j), and the standard deviation value σ.

(i, j), Sc (information such as i, D is efficiently encoded and used as overhead.

The overhead is combined with the beginning of the encoded data obtained by quantizing and encoding the transform coefficients to produce final compressed data.

〔Effect of the invention〕

From the above, the present invention has the following advantages.

(2) By excluding blocks at the periphery of the film from the calculation of the statistics of each component of the transformation coefficient matrix, it is possible to eliminate the negative influence of the peripheral edges on the statistics.

■ During X-ray imaging, the region of interest is imaged at the center of the film, so by excluding blocks at the periphery of the film from the calculation of statistics for each component of the conversion coefficient matrix, more accurate statistics can be obtained for the region of interest. can be found.

(2) The compression process can be sped up by excluding blocks at the periphery of the film from the statistical calculation of each component of the transformation coefficient matrix and by thinning out the blocks to be further calculated.

[Brief explanation of the drawing]

Figure 1 shows an image matrix, Figure 2 is an enlarged view of the lower left part of Figure 1 and shows one example of thinning out, Figure 3 shows another example of thinning out by the same amount, and Figure 4 is a flowchart of block transform encoding.

Claims (2)

[Claims] (1) In an image data compression method in which digitized radiographic image data is divided into blocks and orthogonally transformed encoded, all divided blocks are divided into blocks in the image periphery and blocks not in the image periphery, and The statistics of each component are determined by using part or all of the transform coefficient matrix of blocks other than the image periphery among the transform coefficient matrices obtained by performing orthogonal transformation on each block. Image data compression method. (2) The above-mentioned statistics are used to compress the data of the transform coefficient matrix of the blocks in the peripheral part of the image and the data of the transform coefficient matrix of the blocks not in the peripheral part of the image. image data compression method.