JP4466135B2 - Image processing device - Google Patents

Description

  The present invention relates to an image processing method, an image processing apparatus, and an image processing program for processing a radiographic image, and more specifically, an image processing method, an image processing apparatus, and an image processing program capable of obtaining a radiographic image suitable for diagnosis and the like. About.

  Conventionally, a difference (Sobel) filter, a Laplacian filter, and a simple average filter as a smoothing filter are often used as filters for detecting the edge of an image because of their excellent performance.

  In addition, as represented by wavelet analysis, a reconfigurable filter bank using a combination of a high-frequency filter and a low-frequency filter can also be used as a method for configuring a multi-resolution that enables operations such as edge enhancement over a plurality of frequency bands. Effective methods have been proposed. The filter bank refers to an aggregate of a plurality of filters having different characteristics.

The filters described above are described in Patent Document 1 and Patent Document 2 below.
Japanese Patent Laid-Open No. 11-272853 (first page, FIG. 1) Japanese Patent Laid-Open No. 2-12623 (first page, FIG. 1)

However, generally, a filter bank using a difference (Sobel) filter or a Laplacian filter and a simple average filter has not been configured so far.
For this reason, when configuring a filter bank having the characteristics of these filters, the reconfigurability of the filter must be sacrificed or the shape of the filter must be sacrificed to some extent.

  Furthermore, since a normal two-channel filter bank uses only one edge filter, either the Sobel type or the Laplacian type must be selected to form a filter bank, and an approximate filter bank must be configured. I had to. For this reason, the edge detection performance was not satisfactory.

  The present invention has been made in view of the above problems, and is a filter that can be reconfigured with respect to the above problems and has both edge detection performance using a Laplacian filter and a difference (Sobel) filter. It is an object of the present invention to realize an image processing apparatus that can easily handle multi-resolution analysis excellent in edge detection / smoothing capability at high speed by configuring a bank.

  The inventor of the present application has conducted extensive research to solve the above problems, and as a result, the edge detection and smoothing ability of an image by a filter bank having both edge detection performance by a Laplacian filter and a difference (Sobel) filter. The present inventors have found that an excellent multi-resolution analysis is possible, and have completed the present invention.

That is, the above-described problems are solved by the invention listed below.
(1) The invention described in claim 1 is an image processing apparatus provided with filter processing means for performing filter processing on digital data, and the filter processing means is configured in the form of a three-channel filter bank. A decomposition filter unit that decomposes digital data using a plurality of filters having different characteristics; a downsampling unit that downsamples each of the decomposition outputs of the decomposition filter unit; and the decomposition that receives the output of the downsampling unit An image processing apparatus having a reconstruction filter unit for reconstructing the digital data, wherein the decomposition filter unit includes a Laplacian filter, a difference (Sobel) filter, and a simple average filter It is.

  (2) In the invention described in claim 2, the filter processing is signal component detection for detecting a specific component included in digital data, and the signal component detection is performed by at least one of a Laplacian filter and a difference (Sobel) filter. The image processing apparatus according to claim 1, wherein the image processing apparatus is performed using one processing result.

  (3) The image processing apparatus according to claim 2, wherein the signal component detection is edge detection for detecting an edge of an image included in digital data. .

  (4) The image processing apparatus according to claim 4, wherein the signal component detection is singular point detection for detecting discontinuous points of an image included in digital data. is there.

  (5) In the invention according to claim 5, the filtering process is an evaluation of a noise amount included in digital data, and the evaluation is performed using at least one processing result of a Laplacian filter and a difference (Sobel) filter. The image processing apparatus according to claim 1, wherein the image processing apparatus is an image processing apparatus.

  (6) In the invention according to claim 6, the filter processing is component conversion processing accompanied by conversion of a specific component of digital data, and includes a Laplacian filter, a difference (Sobel) filter and a simple filter constituting the decomposition filter unit. The component conversion process is performed on at least one of the outputs obtained from the average filter, and the digital data subjected to the component conversion process is reconstructed by the reconstruction filter unit. An image processing apparatus.

(7) The invention according to claim 7 is the image processing apparatus according to claim 6, wherein the component conversion processing is enhancement processing.
(8) The invention according to claim 8 is the image processing apparatus according to claim 6, wherein the component conversion process is a noise removal process.

  (9) The invention according to claim 9 is the image processing apparatus according to any one of claims 1 to 8, wherein the digital data of the medical image is filtered.

  (1) According to the first aspect of the present invention, when filter processing is performed on digital data, a decomposition filter unit in the form of a three-channel filter bank comprising a Laplacian filter, a difference (Sobel) filter, and a simple average filter Then, the digital data is decomposed, each decomposition output of the decomposition filter unit is down-sampled by the down-sampling unit, and the decomposed and down-sampled digital data is reconstructed by the reconstruction filter unit.

  Here, various applications are possible by configuring a filter bank using a difference (Sobel) filter and a Laplacian filter excellent in edge detection, and a simple average filter that is fast in calculation, and a high-performance and high-speed image processing algorithm Is possible.

  In other words, a multi-channel filter bank having both edge detection performance by a Laplacian filter and a difference (Sobel) filter and further having a smoothing performance of a simple average filter makes it possible to perform multiplexing with excellent edge detection / smoothing capability. An image processing apparatus capable of handling resolution analysis easily and at high speed can be realized.

  (2) In the invention described in claim 2, signal component detection for detecting a specific component included in digital data is executed as the filter processing of (1), and this signal component detection is performed by a Laplacian filter. And at least one processing result of a difference (Sobel) filter.

  In other words, a filter bank using a difference (Sobel) filter and a Laplacian filter excellent in edge detection is configured, and a specific component of a signal (such as a specific shape of an image) is recognized more accurately by combining them. This is effective for detecting a specific component of a signal.

(3) In the invention according to claim 3, edge detection for detecting an edge of an image included in the digital data is executed as the signal component detection of the above (2).
In other words, a filter bank using a difference (Sobel) filter and a Laplacian filter excellent in edge detection is configured, and a specific component of a signal (such as a specific shape of an image) is recognized more accurately by combining them. Furthermore, regarding the detection of an edge as a detection of a specific component of a signal, since it is decomposed by a Sobel filter and a Laplacian filter, which have a sufficient track record as an edge detection filter, the effect is large.

(4) In the invention described in claim 4, the singular point detection for detecting the discontinuous point of the image included in the digital data is executed as the signal component detection of the above (2).
In other words, a filter bank using a difference (Sobel) filter and a Laplacian filter excellent in edge detection is configured, and a specific component of a signal (such as a specific shape of an image) is recognized more accurately by combining them. In particular, a singular point such as a discontinuous point (defect) in an image is accompanied by an abrupt change in pixel value and becomes an edge point, so that detection by this filter bank is effective.

  (5) In the invention described in claim 5, the noise amount included in the digital data is evaluated as the filtering process of (1), and this evaluation is performed using at least a Laplacian filter and a difference (Sobel) filter. It is performed using one processing result.

  Here, by using extraction of high-frequency components of an image by a filter bank using a difference (Sobel) filter and a Laplacian filter excellent in edge detection, components such as noise can be quantified.

  (6) In the invention described in claim 6, as the filter processing of (1) above, component conversion processing involving conversion of a specific component of digital data is executed, and a Laplacian filter constituting the decomposition filter section and Component conversion processing is performed on at least one of the outputs obtained from the difference (Sobel) filter and the simple average filter, and the digital data subjected to the component conversion processing is reconstructed by the reconstruction filter unit.

  Here, since the configuration of this filter bank can be completely reconfigured, the result of one of the filters in the decomposition filter unit can be obtained by applying a component conversion process to one of the signal components after decomposition and then performing a reconstruction algorithm. It becomes easy to perform image processing that particularly utilizes the.

(7) In the invention described in claim 7, the enhancement processing is executed as the component conversion processing of (6).
Here, since the configuration of this filter bank can be completely reconstructed, the result of the difference (Sobel) filter and the Laplacian filter is obtained when the reconstruction algorithm is performed after component conversion processing is performed on any signal component after decomposition. If it is reconstructed, for example, by multiplying by a constant multiple, a multi-scale enhanced image using a difference (Sobel) filter and a Laplacian filter can be easily obtained.

(8) In the invention according to claim 8, noise removal processing is executed as the component conversion processing of (6) above.
Here, since the configuration of this filter bank can be completely reconstructed, it is obtained by a difference (Sobel) filter and a Laplacian filter when a reconstruction algorithm is performed after component conversion processing is performed on any signal component after decomposition. The amount of noise in the image was measured using the obtained image (noise measurement processing), the high frequency part of the area judged to be noisy was suppressed (noise removal processing), and reconfiguration was performed to suppress noise. An image can be obtained.

(9) In the invention described in claim 9, in the above (1) to (8), the filtering process is performed on the digital data of the medical image.
Here, the difference (Sobel) filter and the Laplacian filter are filters mainly used for digital image processing, and a filter bank combining them is effective for digital image processing. By applying this filter bank to medical images, it becomes possible to perform frequency enhancement processing used in medical image processing at high speed.

DESCRIPTION OF EXEMPLARY EMBODIMENTS Hereinafter, preferred embodiments of an image processing apparatus of the best mode for carrying out the invention will be described with reference to the drawings. Note that the present invention is not limited thereby.
Hereinafter, the configuration and operation of the best mode for carrying out the present invention (hereinafter referred to as embodiment) will be described in detail with reference to the block diagram of FIG. 1 and other explanatory diagrams.

FIG. 1 is a block diagram showing the basic configuration of an image processing apparatus 100 according to an embodiment of the present invention, classified by function.
In FIG. 1, reference numeral 101 denotes a control unit as a control unit that controls each unit of the image processing apparatus 100, and performs control when performing filter processing on digital data.

  Reference numeral 110 denotes a decomposition filter unit in the form of a three-channel filter bank composed of a Laplacian filter, a difference (Sobel) filter, and a simple average filter, which decomposes input digital data (X (n)).

  Here, the decomposition filter unit 110 in the three-channel filter bank format includes a plurality of filters having different characteristics such as a Laplacian filter (H0) 111, a difference (Sobel) filter (H1) 112, and a simple average filter (H2) 113. The component decomposed by a plurality of filters having different characteristics is output.

  Reference numeral 120 denotes a three-channel downsampling unit corresponding to the decomposition filter unit 110, and downsamples and outputs each of the decomposition outputs of the decomposition filter unit 120 by the downsampling filters 121 to 123. Here, since there are 3 channels, 1/3 downsampling is executed to reduce the number of signals (number of pixels) to 1/3.

  A reconstruction filter unit 130 reconstructs (restores) the decomposed and down-sampled digital data. The 3-channel up-sampling unit corresponds to the 3-channel decomposition filter unit 110 and the down-sampling unit 120. 140 and a three-channel filter unit 150.

  Here, the upsampling unit 140 includes sampling filters 141 to 143 that perform upsampling three times corresponding to the downsampling of 1/3 in the downsampling filters 121 to 123, and is reduced to 1/3. The number of signals (number of pixels) is tripled to return to the original state.

  The filter unit 150 includes a plurality of filters having different characteristics such as a Laplacian filter (F0) 151, a difference (Sobel) filter (F1) 152, and a simple average filter (F2) 153 in a three-channel filter bank format. By combining with the upsampling unit 140, the digital data that has been decomposed and downsampled can be reconstructed (restored).

  That is, in the configuration of the three-channel filter bank, the decomposition filter unit 110 is configured using three filters having different characteristics as filter banks, and the down-sampling unit 120 performs 1/3 down-sampling. The reconstruction filter unit 130 includes an upsampling unit 140 that performs upsampling three times, and a filter unit 150 that is configured by a filter having the same filter bank characteristics as the decomposition filter unit 110. (Restoration) can be performed. It is to be noted that the calculation is performed using a matrix, so that the upsampling unit 140 that performs upsampling three times and the filter unit 150 including a filter having the same filter bank characteristics as the decomposition filter unit 110 are completely reconstructed. Proven to perform (restore).

  Here, it is assumed that the transfer function of each filter is as follows.

Given in.
Here, the transfer function of each filter can also be expressed as shown in FIG.
Then, the complete reconstruction condition with zero phase lag of the 3-channel filter bank is

Given in.
The Laplacian filter, the difference (Sobel) filter, and the simple average filter filter can be completely reconstructed.

It can be easily confirmed from the calculation result.
Further, the orthogonality is clear from the filter coefficient. For more details on these,
“Wavelet Analysis and Filter Bank I” (co-authored by G. Strang and T. Nguyen, co-translated by Shinichi Takahashi, Masaaki Ikehara, Baifukan), pages 101-191 and “Wavelet Analysis and Filter Bank II” (G. Strang, T. Nguyen co-author, Shinichi Takahashi, Masaaki Ikehara co-translation, Bafukan).

  The input digital data X (n) has a signal size (the number of signals or the number of pixels) reduced to 1/3 by the respective filters of the decomposition filter unit 11 and the downsampling filter of the downsampling unit 120. It is decomposed into seed (three lines) signals.

  Further, if the same processing is performed on the signals obtained by the simple average filter (H2) 113 and the downsampling filter 123, a finer signal can be obtained. Similarly, it is possible to decompose the signal into multiple scales.

  The three types of signals that have been decomposed and down-sampled are up-sampled by the up-sampling unit 140, and then filtered and added by the filter unit 150, so that the same state as the original signal X (n) is obtained. It is possible to reconstruct (restore) the signal X ′ (n).

  Here, assuming that the outputs of the downsampling unit 120 are X1 (n), X2 (n), and X3 (n), what is important is that X1 (n) and X2 (n) obtained by these filter processes. Is given to the original signal X (n) as a signal processed by a Laplacian filter and a difference (Sobel) filter used as a basic edge extraction filter for signal processing, and X3 (n) is a simple method for creating an unsharp signal at high speed. It is given as a signal processed by an average filter. It is already well known from various documents that these filter processes are effective filters for signal processing and image processing.

  Here, FIG. 2A is an example of an original image before being filtered by the image processing apparatus 100 of the embodiment. FIG. 2B is an example showing the state of the image after being filtered by the image processing apparatus 100.

Here, an example of an image of a portion of the output of the downsampling unit 120 (X1 (n), X2 (n), X3 (n) described above) is shown.
The image in the lower right portion of FIG. 2B represents an image that contains a large amount of high-frequency components that are down-sampled after being filtered by the Laplacian filter 111 in the vertical and horizontal directions with respect to the original image. Yes.

  The image in the center portion of FIG. 2B represents an image containing a large amount of high-frequency components in a state where the original image is down-sampled after being filtered by the difference (Sobel) filter 112 in the horizontal and vertical directions. ing.

  Further, the image in the upper left part of FIG. 2B is obtained by subjecting the original image to the multiple scale by repeatedly performing the downsampling process after being filtered by the simple average filter 113 in the vertical and horizontal directions. It is an example of the image of the state which created the structure.

It is conceivable to perform the following processing on the digital data of the image to be filtered as described above.
(1) Edge detection and enhancement processing:
Of the image data decomposed by the decomposition filter unit 110, either the Laplacian filter 111 or the difference (Sobel) filter 112 is multiplied by β and then reconstructed, whereby the Laplacian filter or the difference (Sobel) 111 filter 112 is used. A signal in which the obtained edge component is enhanced for each direction and resolution is obtained.

  In this case, the Laplacian filter 111 or the difference (Sobel) filter 112 itself may be changed to include β by the control of the control unit 101 in the configuration of FIG. 1, or the component by the control of the control unit 101 in the configuration of FIG. You may make it change the amplification degree of amplifier 161-163 of each channel in the conversion process part 160 from normal 1 to (beta).

  At this time, depending on the pixel value of the image obtained by the simple average filter 113, the enhancement correction coefficient α is determined as shown in FIG. 3, and the enhancement correction coefficient α is applied to the above-described enhancement coefficient β. By using the obtained enhancement coefficient β ′, it is possible to suppress enhancement of a low-dose portion having a large noise component in the X-ray image.

(2) Signal component detection processing:
By analyzing each signal component of the down-sampled image data decomposed by the decomposition filter unit 110, a specific signal, for example, a sudden change in signal value (image edge, image singular point (discontinuous point)) ) And the like can be detected.

  Here, for example, in the case of a step (edge) of an image, a steep signal change in a grayscale section is detected as a large signal value in the difference filter, and is detected as a sudden pixel value change across 0 in the Laplacian filter ( FIG. 5 (a)).

  Further, for example, in the case of an image line (a line orthogonal to the detection direction), a steep signal change in a grayscale section is detected as a change in a large positive and negative signal value in the difference filter, and two positive values in the Laplacian filter. This is detected as a change in pixel value in which a large negative valley occurs in the central portion between the direction peaks (FIG. 5B).

  In addition, for example, in the case of an image roof (a roof or a mountain range that is orthogonal to the detection direction), a gradual signal change in the light and shade section is detected as positive and negative wide positive and negative signal values in the difference filter. In the Laplacian filter, two positive peaks appear at the roof start point and end point, and this is detected as a pixel value change in which a large negative valley occurs at the top of the roof (FIG. 5C).

In addition, for example, in the case of a singular point (discontinuous point in an image) of an image, it is detected as a steep step or a steep line in a light and shade section.
In the above case, it is possible to detect a specific component of an image using only a difference (Sobel) filter and a Laplacian filter, but the combined use improves the detectability. That is, as described above, a filter bank using a difference (Sobel) filter and a Laplacian filter excellent in edge detection is configured, and a specific component of a signal (a specific shape of an image, etc.) is used by combining them. Can be recognized more accurately, and is effective in detecting a specific component of a signal. Furthermore, regarding the detection of an edge as the detection of a specific component of the signal, the effect is great because it is decomposed by a difference (Sobel) filter and a Laplacian filter, which have a sufficient track record as an edge detection filter.

In addition, a filter bank using a difference (Sobel) filter and a Laplacian filter, which are excellent in edge detection, is configured and used in combination to recognize a specific component of a signal (such as a specific shape of an image) more accurately. In particular, a singular point such as a discontinuous point (defect) in an image is accompanied by an abrupt change in pixel value and becomes an edge point, so that detection by this filter bank is effective.
(3) Noise evaluation and suppression:
Each signal component of the down-sampled image data decomposed by the decomposition filter unit 110 is analyzed, and first, a portion having a small change or variance in the low resolution component signal output from the simple average filter 113 (FIG. 6B). ) A broken line part) is extracted.

  Such a portion can be considered to be a peripheral portion (a portion other than the subject region) that does not include subject information in the image and has no change. For example, in the radiographic image shown in FIG. 7, it can be considered as an irradiation field area where no radiation is irradiated, or a blank area where the radiation reaches the light receiving part directly without passing through the object around the object. it can.

  Then, the variance of the signal of the high resolution component corresponding to the small change portion of the low resolution component is examined. In this way, it is possible to evaluate the amount of noise that will be included in the entire image. In order to extract a region having no change in the low resolution portion, for example, a dispersion value of a fixed region having a width A (A × A in the case of an image) is examined from the signal end and a region that is equal to or smaller than a threshold value may be taken.

  Also in this case, since two filters effective for high frequency detection are used, a method of combining these detection results or using them separately can be considered. Such a method is effective, for example, for examining the noise component of the direct radiation irradiation region or the region outside the irradiation field in the X-ray image signal, and the noise amount of the image can be estimated (evaluated) by these amounts.

Further, by reducing and reconstructing the signal component of an image that is equal to or higher than the threshold value, which is the determination, it is possible to obtain an image in which the edge component is suppressed in a noisy image.
That is, since the filter bank configuration can be completely reconstructed, it can be obtained with a difference (Sobel) filter and a Laplacian filter when a reconstruction algorithm is performed after component conversion processing is performed on one of the decomposed signal components. The noise amount of the image is measured (noise measurement processing) using the obtained image, and the high-frequency part of the area determined to be noisy is suppressed (noise) as shown in FIG. 10 (a) or 10 (b). An image with suppressed noise can be obtained by performing reconfiguration and reconstructing.

(4) Other processing:
In addition, as shown in FIG. 8, with respect to the high-frequency component signal obtained from at least one of the Laplacian filter 111 or the difference (Sobel) filter 112, the enhancement of the portion with the large signal value is suppressed, and the enhancement with the small portion is performed. After performing such nonlinear conversion, reconstruction is performed. By performing the processing in this way, it is possible to emphasize a minute structure without emphasizing more than necessary an area where sufficient contrast has already been obtained.

(5) Application examples:
In the above description, the Laplacian filter 111 and the difference (Sobel) filter 112 are filters mainly used for digital image processing, and the filter bank of this embodiment combining them is effective for various digital image processing. It is.

  By applying this filter bank to medical images, it becomes possible to perform frequency enhancement processing used in medical image processing at high speed. In general, in a medical image, information on a region to be noticed is buried and difficult to discriminate. However, by applying the image processing apparatus of this embodiment, edge detection is performed on a region or lesion to be noticed.・ Multi-resolution analysis with excellent smoothing capability can be easily and quickly handled.

It is a functional block diagram which shows functionally the whole structure of embodiment of this invention. It is process explanatory drawing which shows the flow of the whole process of embodiment of this invention. It is a functional block diagram which shows functionally the whole structure of embodiment of this invention. It is explanatory drawing which shows the mode of the process in embodiment of this invention. It is explanatory drawing which shows the mode of the process in embodiment of this invention. It is explanatory drawing which shows the mode of the process in embodiment of this invention. It is explanatory drawing which shows the mode of the process in embodiment of this invention. It is explanatory drawing which shows the mode of the process in embodiment of this invention. It is explanatory drawing which shows the mode of the process in embodiment of this invention. It is explanatory drawing which shows the mode of the process in embodiment of this invention.

Explanation of symbols

DESCRIPTION OF SYMBOLS 100 Image processing apparatus 101 Control part 110 Decomposition filter part 120 Downsampling part 130 Reconstruction filter part 140 Upsampling part 150 Filter part 160 Component conversion process part

Claims (9)

An image processing apparatus provided with a filter processing means for performing filter processing on digital data,
The filter processing means is configured in the form of a three-channel filter bank, and down-samples each of a decomposition filter unit that decomposes digital data with a plurality of filters having different characteristics, and a decomposition output of the decomposition filter unit A configuration having a downsampling unit, and a reconstruction filter unit that receives the output of the downsampling unit and reconstructs the decomposed digital data;
The decomposition filter unit includes a Laplacian filter, a difference (Sobel) filter, and a simple average filter.
An image processing apparatus. The filtering process is a signal component detection for detecting a specific component included in digital data,
The signal component detection is performed using at least one processing result of a Laplacian filter and a difference (Sobel) filter.
The image processing apparatus according to claim 1. The signal component detection is edge detection for detecting an edge of an image included in digital data.
The image processing apparatus according to claim 2. The signal component detection is singular point detection for detecting discontinuous points of an image included in digital data.
The image processing apparatus according to claim 2. The filtering process is an evaluation of the amount of noise included in the digital data,
The evaluation is performed using at least one processing result of a Laplacian filter and a difference (Sobel) filter.
The image processing apparatus according to claim 1. The filtering process is a component conversion process involving conversion of a specific component of digital data,
The component conversion process is performed on at least one of the output obtained from a Laplacian filter, a difference (Sobel) filter, and a simple average filter constituting the decomposition filter unit, and the digital data subjected to the component conversion process is reconstructed. Reconfigure in the filter part,
The image processing apparatus according to claim 1. The component conversion process is an enhancement process.
The image processing apparatus according to claim 6. The component conversion process is a noise removal process.
The image processing apparatus according to claim 6. Filtering digital data of medical images,
The image processing apparatus according to claim 1, wherein the image processing apparatus is an image processing apparatus.

Images