JPH1063643A - Method for transforming picture - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make it possible to execute wavelet transformation for picture data by using small memory capacity. SOLUTION: Original picture data 1 are stored in a memory in each several lines and filtered in a main scanning direction through a low pass filter 10 and a high pass filter 11 and these filtered results are respectively stored in respective buffer memories 4, 5. The contents of the memories 4, 5 are filtered in a sub-scanning direction, succeeding unprocessed data are stored in the memories 4, 5 instead of the processed data and the memories 4, 5 are circulatively utilized, so that the picture data in each frequency band can be successively generated in each several lines.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an image conversion method, and more particularly to a method of using a memory in a filtering process when performing a wavelet transform on image data and a file format of the converted data.

[0002]

2. Description of the Related Art Wavelet transform is one of the frequency analysis methods for signals. However, it is easier to detect local change information of a signal than Fourier transform which is also widely used as a frequency analysis method. Because of their superiority, they have recently been used in various signal processing fields.

In the field of image processing, when performing different processing on image data for each frequency band, the wavelet transform is used as means for classifying image data for each frequency band. Specifically, for example, high-frequency separation for removing noise, and compression processing by reducing data in a frequency band with a lot of noise may be mentioned. Applicant also has Marc Antonin
(Marc Antonini et al., Ima
ge Coding Using Wavelet Transform, IEEE TRANSACTIO
NS ON IMAGE PROCESSING, VOL.1, NO.2, p205-220, A
Various image data compression methods based on PRIL 1992 have been proposed (JP-A-6-350989, JP-A-6-350990, and JP-A-7-350990).
23228, 7-23229, 7-79350, and Japanese Patent Application No. 8-14510).

Generally, in image processing using wavelet transform, a hierarchical expression of an image is emphasized. As a method for hierarchically converting the format of image data, image data of the entire image is stored in a memory. To
Wavelet transform processing is performed to sequentially output image components having lower resolutions into one file.

[0005]

The above conventional method requires a sufficient memory to store image data of the entire image to be processed. However, an image processing apparatus that performs such image processing generally has limited memory resources that can be used. Therefore, it is desirable that the processing can be realized using as little memory as possible.

SUMMARY OF THE INVENTION In view of the above problems, an object of the present invention is to provide a method for sequentially performing wavelet transform with a small memory.

[0007]

According to an image conversion method of the present invention, image data representing an image is subjected to a filtering process in a vertical direction and a horizontal direction of the image to perform a wavelet transform of the image data. In the method, the image data is subjected to a filtering process in the direction of the line by a low-pass filter for each of a predetermined number of partial image data of the image data, and a low-frequency component representing a low-frequency component of the partial image data is obtained. The component image data is obtained and stored in a first buffer memory having a predetermined number of lines, and high-frequency component image data representing a high-frequency component of the partial image data is subjected to a filtering process in the direction of the line by a high-pass filter. Stored in a second buffer memory having the number of lines of The stored image data is filtered by a low-pass filter and a high-pass filter having a filter size equal to or less than the number of lines of the buffer memory in a direction perpendicular to the lines of the buffer memory, respectively. Performing a wavelet transform by applying a filtering process to image data stored in the buffer memory in a direction perpendicular to the line of the buffer memory by a low-pass filter and a high-pass filter having a filter size equal to or less than the number of lines of the buffer memory. It is a feature.

At this time, the converted data obtained by the wavelet transform may be sequentially output to one file, or may be output after being classified into different files for each frequency band.

Here, one line of image data means a set of data corresponding to 1024 pixels arranged in one line, for example, when an image is composed of 1024 × 1024 pixels. The partial image data is obtained by gathering a set of this data for a predetermined number of lines. Specifically, the predetermined number of lines is desirably an even number of lines, for example, two lines or four lines. . However, this is not necessarily limited to an even number. Since the partial image data is reduced to に よ っ て by the filtering process, if the number of lines is set to an even number, data of an integral multiple of the number of lines is obtained. This is because it is convenient. Regarding the buffer memory, the capacity corresponding to one line of the image data is referred to as one line.

The low-pass filter and the high-pass filter are 1
Although a dimensional filter is desirable, there is no particular limitation. When these filters are one-dimensional filters, the filter size is specifically the filter length, the “predetermined number of lines” in the buffer memory is smaller than the number of lines that can store the entire image data, and Is greater than the filter length of the one-dimensional filter used in the filtering process performed in the direction perpendicular to the line of the buffer memory,
Alternatively, the number of lines is the same. Here, a filter length of n means that the filter is a filter that processes n pixels, and thus, if the filter is longer than the filter length or has the same number of lines, the filter is n
In the case of a filter that processes pixels, the number of lines is equal to or more than n lines.

In the wavelet transform process, the above-described process is repeatedly performed on the converted data obtained as described above, and a wavelet transform may be performed in several stages. It is applicable regardless of the number of steps.

The theory of the wavelet transform, the design method of the filter based on the wavelet function, and the filtering process by those filters are described in detail in the above cited examples, and therefore, the description is omitted in this specification. I do.

[0013]

According to the present invention, when performing a filtering process on image data and performing a wavelet transform, the image data is divided into fragments large enough to be subjected to the filtering process. In order to perform the sequential processing, the memory for storing the data in the middle of the processing may use a minimum necessary memory cyclically, and the wavelet transform can be performed with a small memory.

In the above method, the converted data of each frequency band is output alternately. The converted data may be sequentially stored in one file as it is, or may be stored in a different file for each frequency band. If the data is classified and stored, a hierarchical expression of the data becomes possible, and the data can be used for each frequency band as in the conventional method.

[0015]

DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, an image conversion method according to the present invention will be described in detail with reference to the drawings. The embodiment described below is a method applied to a radiation image recording / reproducing system proposed by the present applicant in, for example, the above-mentioned Japanese Patent Application Laid-Open No. 55-12492. This system reads a radiation image of a human body recorded on a stimulable phosphor sheet as digital image data by laser beam scanning, and performs various image processing in order to record and reproduce or store and manage the image. As one of such image processing, image data compression processing using wavelet transform is performed as shown in the cited example of the above-mentioned Japanese Patent Application Laid-Open No. 6-350989.

First, an outline of the wavelet transform processing and problems of the conventional method will be described with reference to FIG. As shown in FIG. 1, generally, the wavelet transform is performed on the original image data in a main scanning direction (hereinafter simply referred to as a main scanning direction) when the original image data is read.
In addition, in the two directions of the sub-scanning direction during reading (hereinafter, simply referred to as the sub-scanning direction), the filtering process is performed for every other pixel. At this time, a filter designed based on the basic wavelet function is used as the filter, and a filtering process is performed stepwise by a plurality of filters based on the theory of the wavelet transform. (For the theory of the wavelet transform and the design method of an appropriate filter, see the above cited example.) As shown in FIG.
By the low-pass filter 10 and the high-pass filter 11,
Each is processed every other pixel in the main scanning direction. As a result, the original image data processed by each filter is reduced to half in the main scanning direction.

Further, the image data processed by the low-pass filter 10 is processed by the low-pass filter 12 and the high-pass filter 13 every other pixel in the sub-scanning direction. As a result, each of the main scanning direction and the sub-scanning direction is reduced to half, that is, 1 /
LL image data 16 and LH image data 17 having a size of 4 are obtained. Here, L means L of the low-pass filter, H means H of the high-pass filter, and LL indicates the result processed by the low-pass filter in both the main scanning direction and the sub-scanning direction. The result processed by the high-pass filter is represented as LH.

Similarly, the image data processed by the high-pass filter 11 is processed every other pixel in the sub-scanning direction by the low-pass filter 14 and the high-pass filter 15, respectively, and the HL image data 18 and the HH image data 19 Is obtained. In the present embodiment, the low-pass filters 10, 12, and 14 are the same filters, and the high-pass filters 11, 13, and 15 are the same filters, but these filters may be different.

The LL image obtained by performing a filtering process using a low-pass filter in both the main scanning direction and the sub-scanning direction of the wavelet transform image 2 thus obtained is a reduced image of the original image 1. It is known.

In order to perform conversion by the method described above, a buffer memory capable of storing data of one surface of an original image to be processed is required. However, in general, the memory resources of an image processing apparatus are limited, and not all of them can be used only for the above-described conversion processing. Therefore, as a conversion method, the smaller the memory used, the better. Therefore, in the image conversion method of the present invention, a small amount of memory is effectively used by dividing the image data and sequentially processing the image data.

FIG. 2 is a diagram showing one embodiment of the conversion method of the present invention. In this embodiment, two lines of data of the original image data 1 are read into the memory and processed, and when the processing for the two lines of data is completed, the next 2 lines are processed.
The processing is sequentially performed by reading the lines. In FIG. 2, these two lines are represented by thin thick frames. Note that the low-pass filter and the high-pass filter used in the present embodiment are all one-dimensional filters.

First, a filtering process in the main scanning direction as shown in the figure as a filter 6 is performed on two lines of data 3 (hereinafter referred to as partial image data 3) read into the memory. This shows a filtering process performed by the low-pass filter 10 and the high-pass filter 11, respectively. As described above, since this filtering process is performed on the partial image data 3 every other pixel, the data obtained as a result of performing such a filtering process on the data of two lines is one line worth of data. Data. Therefore, a low-pass filter
The data for one line obtained by 10 is stored in the buffer memory 4 and the data for one line obtained by the high-pass filter 11 is stored in the buffer memory 5 (shown by a thin bold frame). ).

The buffer memory 4 includes a low-pass filter 12
And the number of lines of the high-pass filter 13 is larger than the length of the longer filter. For example, when the filter length of the low-pass filter 12 is 7 and the filter length of the high-pass filter 13 is 5, the number of lines in the buffer memory 4 must be 7 or more. Similarly, it is assumed that the buffer memory 5 also has a larger number of lines than the length of the longer one of the low-pass filter 14 and the high-pass filter 15.

Reading of the partial image data 3 into the memory, and a low-pass filter for the partial image data 3
After the filtering process using 10 is repeated until a predetermined amount of data is accumulated in the buffer memory 4, a filtering process in the sub-scanning direction as shown in the drawing as the filter 7 is performed. This indicates a filtering process performed by the low-pass filter 12 and the high-pass filter 13, respectively.

In the case of performing the filtering process in the sub-scanning direction, in the conventional method, the filtering process is performed for one vertical column in the sub-scanning direction, and then the next vertical column is processed. In this embodiment, each vertical column is processed. Are processed in parallel. That is, the low-pass filter 17 or the high-pass filter 13 shown as the filter 7 is moved in the main scanning direction, and data is sequentially output to a file. By making a round in the main scanning direction, all data for one line in the buffer memory 4 has been processed, so that this data can be discarded and new data can be stored, and therefore, the buffer memory 4 having a limited capacity is circulated. Can be used

It is to be noted that the cyclic use of the memory means 1
Each time the processing for a line is completed, the data is moved up by one line to add new data to the last part of the buffer memory 4, or a method of overwriting a new line in an area where an unnecessary line is stored. Is meant. In the case of overwriting, it is necessary to change the filter so as to follow the order of the lines, but since this is apparent to those skilled in the art, the details are omitted.

The processing for the data stored in the buffer memory 4 has been described above.
The same applies to the processing for the data stored in the low pass filter 12 and the high pass filter 13 in the above description.
Respectively.

In the above sequential processing, LL image data, LH image data, HL image data, and HH image data are generated for one line each time the data in the buffer memory is replaced. Therefore, if the sequentially processed data is sequentially output to a file, the file format is as shown in FIG. This file format is disjointed as image data for each frequency band, but for example, if it is a process that encodes for transmission or compression and then performs inverse wavelet transform to obtain the original image There is no particular inconvenience, and it can be said that it is superior to the conventional method because the memory required for processing can be saved.

Also, depending on the encoding method, the file format for each line as shown in FIG. 3 may be inconvenient. In such a case, for example, by making the partial image data data for 8 lines, it becomes possible to output data every 4 lines for each frequency band as shown in FIG. Can also be performed. That is, if the number of lines of the partial image data is changed as needed, it is possible to output image data of each frequency band in a certain unit.

Further, at the time of file output, if the data is classified into a plurality of files and output as shown in FIG. 5, the hierarchical representation of the image data becomes possible as in the conventional method. For example, when performing compression processing by changing the degree of compression for each frequency band, it is easier to perform processing if the files are classified into different files. In the case where only image data of a certain frequency band is particularly distinguished, for example, when LL image data is used as a reduced image, as shown in FIG. May be classified into two files and output.

The above conversion method can also be applied to a case where the wavelet conversion is performed in two or more stages. That is, the same sequential processing can be further performed on the LL image data obtained by the above processing. For example, the above processing is performed as data of four lines of the original image as partial image data, and thus obtained. By performing the same processing on the LL image data for two lines, an output as shown in FIG. 7 can be obtained. In the case where the wavelet transform is performed in multiple stages by the conventional method, a sufficient memory capacity is required to store the entire data to be processed in each stage. However, in the transform method of the present invention, the number of stages is increased. But that doesn't require much memory.

As described above, according to the image conversion method of the present invention, the wavelet transform can be performed by the sequential processing using a small amount of memory, and the practical effect is extremely large.

[Brief description of the drawings]

FIG. 1 is a diagram showing an outline of a wavelet transform process;

FIG. 2 is a diagram showing an embodiment of the conversion method of the present invention.

FIG. 3 is a diagram showing an example of a file format when storing conversion data obtained by the conversion method of the present invention in one file.

FIG. 4 is a diagram showing another example of a file format when storing conversion data obtained by the conversion method of the present invention in one file.

FIG. 5 is a diagram showing an example of a file format when storing conversion data obtained by the conversion method of the present invention in a plurality of files.

FIG. 6 is a diagram showing another example of a file format when storing conversion data obtained by the conversion method of the present invention in a plurality of files.

FIG. 7 is a diagram showing an embodiment of a multi-stage wavelet transform using the transform method of the present invention.

[Explanation of symbols]

Reference Signs List 1 original image 2 wavelet transform image 3 partial image data read into memory 4 buffer memory 5 buffer memory 6 filter (main scanning direction) 7 filter (sub scanning direction) 10 low-pass for filtering every other pixel in main scanning direction Filter 11 High-pass filter for filtering every other pixel in the main scanning direction 12 Low-pass filter for filtering image data filtered by the low-pass filter in the main scanning direction every other pixel 13 Low-pass in the main scanning direction A low-pass filter 14 for filtering the image data filtered by the filter every other pixel in the sub-scanning direction 14 For filtering the image data filtered by the high filter in the main scanning direction every other pixel in the sub-scanning direction Low-pass filter 16 LL image 17 LH image 18 HL image 19 HH image for filtering every other pixel of the image data which has been filtered by-pass filter 15 main scan direction at a high filter in the sub-scanning direction

Claims (3)

[Claims] 1. An image conversion method for performing a wavelet transform of image data by performing filtering processing on image data representing an image in a vertical direction and a horizontal direction of the image, respectively, wherein: For each of the predetermined number of lines of partial image data, a low-pass filter is applied in the direction of the line to obtain low-frequency component image data representing a low-frequency component of the partial image data, and a predetermined number of lines are obtained. The high-frequency component image data representing the high-frequency component of the partial image data is stored in a second buffer memory having a predetermined number of lines while storing the high-frequency component image data in the direction of the line by a high-pass filter while storing the high-frequency component image data in the first buffer memory. The image data stored in the first buffer memory is A low-pass filter and a high-pass filter each having a filter size equal to or less than the number of lines of the memory perform a filtering process in a direction perpendicular to the lines of the buffer memory, respectively, and apply a filtering process to the image data stored in the second buffer memory. An image conversion method, wherein a low-pass filter and a high-pass filter having a filter size equal to or less than the number of lines of a buffer memory perform a filtering process in a direction perpendicular to the lines of the buffer memory to perform a wavelet transform. 2. The image conversion method according to claim 1, wherein the converted data obtained by the wavelet transform is classified into different files for each frequency band and output. 3. The image conversion method according to claim 1, wherein the converted data obtained by the wavelet conversion is sequentially output to one file.