JP2007018379A - Image processing method and image processing device - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To emphasize edge components, such as the outline of a structure, and attenuate noise components in an image signal while suppressing the generation of noise and artifacts caused by processing. <P>SOLUTION: In an image processing device 10, according to edge information and/or noise information acquired in a reference frequency band produced by multi-resolution decomposition of an image signal, at least either noise smoothing or edge smoothing is applied to unsharp image signals in a plurality of frequency bands and the original image signal to produce edge components and/or noise components, and density-dependent correction is applied to the produced edge components and/or noise components to produce converted image signals. Differential image signals are then acquired between the produced converted image signals and the unsharp image signals in adjacent frequency bands (the highest frequency band as for the original image signal) respectively, and each differential image signal is multiplied by a predetermined conditioning factor and added to the original image signal to produce a processed image signal. <P>COPYRIGHT: (C)2007,JPO&INPIT

Description

  The present invention relates to an image processing apparatus and an image processing method for performing processing for enhancing edge components in image signals and processing for suppressing noise components.

  Conventionally, by converting an image into a multi-resolution space, the image is decomposed into a plurality of frequency bands, and enhancement processing is performed on at least one of the image signal components of each frequency band decomposed into a plurality of frequency bands. And image processing for obtaining an enhanced image signal by inversely transforming the image signal component of the frequency band subjected to the enhancement processing and the image signal component of the other frequency band.

  For example, in Patent Document 1, the degree of enhancement when performing enhancement processing on at least one frequency band of images of each frequency band decomposed into a plurality of frequency bands by conversion to a multi-resolution space is emphasized. An image processing method for determining based on image signal values of a plurality of frequency band images including at least one of frequency band images to be processed is described, and further, corresponding image signal values for the plurality of frequency band images It is described that emphasis processing is performed only on pixels having the same sign.

  In Patent Document 2, noise components and edge components of images in each frequency band that have been decomposed into a plurality of frequency bands by conversion to a multi-resolution space are separated, and smoothing for the noise components is performed for each frequency band image signal. A technique is described in which a processed image signal is obtained by performing an emphasizing process and / or an enhancement process on edge components, and a processed image signal is obtained based on the processed frequency band image signal.

Japanese Patent Application Laid-Open No. 2004-228561 performs a conversion process on a non-sharp image signal of a plurality of frequency bands created from an original image signal, and obtains a difference image signal obtained from a difference signal between the unsharp image signal and the image signal after the conversion process. Describes a technique for obtaining a processed image by adding a high-frequency component signal obtained by adding to an original image signal or its lowest frequency image signal.
JP 11-345331 A JP 2000-306089 A JP 2002-183727 A

  However, in the methods described in Patent Documents 1 and 3, a necessary signal component such as a contour in the frequency band to be emphasized cannot be specified and reliably emphasized, and the processed image is not processed. Sometimes it was unnatural. In Patent Documents 1 and 2, noise and artifacts may occur due to edge enhancement processing and noise smoothing processing.

  The present invention has been made in view of the above technical problem, and enhances edge components such as contours of structures in an image signal and attenuates noise components while suppressing generation of noise and artifacts due to processing. Is to be able to.

The invention described in claim 1
A resolution processing step of obtaining multi-resolution conversion of the input image signal to obtain non-sharp image signals of a plurality of different frequency bands; and at least one of the input image signal and / or the non-sharp image signals of the plurality of frequency bands A conversion processing step for obtaining a difference image signal by taking a difference between the converted image signal obtained by performing the conversion process and an image signal in the frequency band adjacent to the converted image signal or an image signal in the highest frequency band; and the difference image signal And a restoration processing step of obtaining a processed image by adding to the input image signal,
The conversion process step includes
An information acquisition step of acquiring edge information and / or noise information of the input image signal;
Based on the acquired edge information and / or noise information, at least one of an edge smoothing process or a noise smoothing process is performed on at least one of the input image signal and / or the unsharp image signal of the plurality of frequency bands. After performing two processes, performing a density-dependent correction process to obtain the converted image signal;
It is characterized by including.

The invention according to claim 2 is the invention according to claim 1,
The conversion step includes
Based on the acquired edge information and / or noise information, at least one of the input image signal and / or the non-sharp image signal of the plurality of frequency bands is subjected to a noise smoothing process, and the noise smoothing is performed. A step of performing density-dependent correction processing on the image signal to obtain the first converted image signal; the first converted image signal and an image signal in a frequency band adjacent to the converted image signal or an image signal in the highest frequency band; And obtaining a differential image signal in which the edge component is adjusted by applying a predetermined edge adjustment coefficient to the differential signal after obtaining a differential signal mainly composed of edge components by taking the difference of When,
Based on the acquired edge information and / or noise information, at least one of the input image signal and / or the non-sharp image signal of the plurality of frequency bands is subjected to an edge smoothing process, and the edge smoothing is performed. A step of performing density-dependent correction processing on the image signal to obtain the second converted image signal; the second converted image signal and an image signal in a frequency band adjacent to the converted image signal or an image signal in the highest frequency band; And obtaining a differential image signal in which the noise component is adjusted by applying a predetermined noise adjustment coefficient to the differential signal after obtaining a differential signal mainly composed of a noise component by taking the difference of When,
It is characterized by including.

The invention according to claim 3 is the invention according to claim 1 or 2,
The edge information includes at least one of information indicating a pixel position of an edge component, a sign of a signal value, a direction of the edge component, and a pixel position of an edge inflection point. The noise information is a noise component. And at least one of information indicating a local dispersion value, an entropy value, and a pixel position.

The invention according to claim 4 is the invention according to any one of claims 1 to 3,
In the noise smoothing process, two-dimensional smoothing is performed for pixels at positions that are not edge components, and one-dimensional only for directions other than the edge gradient direction or directions perpendicular to the edge gradient direction for pixels of edge components. It is characterized by performing a smoothing process.

The invention according to claim 5 is the invention according to any one of claims 1 to 4,
The edge smoothing process is characterized in that a one-dimensional smoothing process is performed only in the edge gradient direction on the edge component pixels.

The invention according to claim 6 is the invention according to any one of claims 1 to 5,
The multi-resolution conversion uses a Laplacian pyramid method.

The invention according to claim 7 is the invention according to any one of claims 1 to 5,
The multi-resolution conversion uses a wavelet transform.

The invention according to claim 8 provides:
Decomposition processing means for acquiring non-sharp image signals of a plurality of different frequency bands by performing multi-resolution conversion of the input image signal; and at least one of the input image signals and / or the non-sharp image signals of the plurality of frequency bands A conversion processing means for obtaining a difference image signal by taking a difference between the converted image signal obtained by performing the conversion process and an image signal in the frequency band adjacent to the converted image signal or an image signal in the highest frequency band; and the difference image signal In an image processing apparatus comprising restoration processing means for obtaining a processed image by adding to the input image signal,
The conversion processing means includes
Information acquisition means for acquiring edge information and / or noise information of the input image signal;
Based on the acquired edge information and / or noise information, at least one of an edge smoothing process or a noise smoothing process is performed on at least one of the input image signal and / or the unsharp image signal of the plurality of frequency bands. Means for obtaining the converted image signal by performing density-dependent correction processing after performing two processes;
It is characterized by having.

The invention according to claim 9 is the invention according to claim 8,
The converting means includes
Based on the acquired edge information and / or noise information, at least one of the input image signal and / or the non-sharp image signal of the plurality of frequency bands is subjected to a noise smoothing process, and the noise smoothing is performed. Means for performing a density-dependent correction process on the image signal to obtain the first converted image signal; the first converted image signal and an image signal in a frequency band adjacent to the converted image signal or an image signal in the highest frequency band; And obtaining a difference image signal in which the edge component is adjusted by applying a predetermined edge adjustment coefficient to the difference signal after obtaining a difference signal mainly composed of edge components by taking the difference of When,
Based on the acquired edge information and / or noise information, at least one of the input image signal and / or the non-sharp image signal of the plurality of frequency bands is subjected to an edge smoothing process, and the edge smoothing is performed. Means for performing a density-dependent correction process on the image signal to obtain the second converted image signal; the second converted image signal and an image signal in a frequency band adjacent to the converted image signal or an image signal in the highest frequency band; A noise component adjusting means comprising: obtaining a difference signal mainly composed of noise components by taking a difference between the difference signals and obtaining the difference image signal in which the noise components are adjusted by applying a predetermined noise adjustment coefficient to the difference signal When,
It is characterized by having.

The invention according to claim 10 is the invention according to claim 8 or 9, wherein
The edge information includes at least one of information indicating a pixel position of an edge component, a sign of a signal value, a direction of the edge component, and a pixel position of an edge inflection point. The noise information is a noise component. And at least one of local dispersion value, entropy value, and information indicating the pixel position.

The invention according to claim 11 is the invention according to any one of claims 8 to 10,
In the noise smoothing process, two-dimensional smoothing is performed for pixels at positions that are not edge components, and one-dimensional only for directions other than the edge gradient direction or directions perpendicular to the edge gradient direction for pixels of edge components. It is characterized by performing a smoothing process.

The invention according to claim 12 is the invention according to any one of claims 8 to 11,
The edge smoothing process is characterized in that a one-dimensional smoothing process is performed only in the edge gradient direction on the edge component pixels.

The invention according to claim 13 is the invention according to any one of claims 8 to 12,
The multi-resolution conversion uses a Laplacian pyramid method.

The invention according to claim 14 is the invention according to any one of claims 8 to 12,
The multi-resolution conversion uses a wavelet transform.

  According to the inventions described in claims 1 to 14, it is possible to enhance edge components such as the contours of structures in an image signal and reduce noise components while suppressing the generation of noise and artifacts due to processing. Become.

  According to the second and ninth aspects of the present invention, each of the image signal mainly composed of edge components obtained by noise smoothing and the image signal mainly composed of noise components obtained by edge smoothing is density-dependent. Since correction processing can be performed, for example, density-dependent correction processing is performed so that noise suppression processing is strongly applied in low density areas where noise is conspicuous, and edge enhancement processing is suppressed to reduce noise due to processing Density-dependent correction processing can be performed. Further, in a high density portion where noise is not conspicuous, it is possible to perform density dependence correction processing so that noise suppression processing is weakened, and density dependence correction processing so that edge enhancement processing is strong. This makes it possible to enhance edge components such as the contours of structures in the image signal and attenuate noise components while more effectively suppressing the occurrence of noise and artifacts.

<First Embodiment>
The first embodiment of the present invention will be described in detail below.
[Configuration of image processing apparatus]
First, the configuration of the image processing apparatus 10 in the present embodiment will be described.
FIG. 1 is a block diagram showing a functional configuration of an image processing apparatus 10 according to the present embodiment.

  As shown in FIG. 1, an image processing apparatus 10 includes an image input unit 11 that receives an image (original image signal Sin) from an external device, and performs edge enhancement / noise suppression (noise attenuation) processing on the input original image signal. A decomposition / conversion processing unit 12 and a restoration processing unit 13 that perform processing to obtain a processed image signal Sout, and an image output unit 14 that outputs the processed image signal Sout subjected to the restoration processing as a visible image.

  Hereinafter, processing procedures, configurations, and operations of the decomposition / conversion processing unit 12 and the restoration processing unit 13 will be described in detail.

  The decomposition / conversion processing unit 12 and the restoration processing unit 13 are configured to include a CPU (Central Processing Unit), a ROM (Read Only Memory), and a RAM (Random Access Memory), and a program stored in the CPU and the ROM. The processed image signal Sout is obtained by performing edge enhancement / noise suppression processing on the original image signal Sin by software processing through cooperation. The disassembly / conversion processing unit 12 and the restoration processing unit 13 may be configured by dedicated hardware and firmware, respectively. The decomposition / conversion processing unit 12 implements a decomposition processing unit and a conversion processing unit, and the restoration processing unit 13 implements a restoration processing unit.

The processing procedure of edge enhancement / noise suppression processing is as shown in FIG.
First, the decomposition / conversion processing unit 12 obtains an unsharp image signal having a plurality of spatial frequencies (hereinafter simply referred to as frequencies) by performing multi-resolution conversion on the original image signal Sin (step S1). Edge information and noise information in the original image signal Sin are acquired based on any one of the non-sharp image signals in a plurality of frequency bands (step S2), and based on the acquired edge information and / or noise information. Then, pixel value conversion processing (conversion processing) is performed on the original image signal Sin and the unsharp image signal in each frequency band (step S3), and the difference between the unsharp image signal in the adjacent frequency band (for the original image signal Sin) The difference image signal is obtained by multiplying the difference signal by a predetermined adjustment coefficient (difference from the unsharp image signal in the highest frequency band) (step S4). Further, the restoration processing unit 13 adds the difference image signal to the original image signal Sin, thereby obtaining a processed image signal Sout in which edge enhancement and noise are suppressed (step S5).

  FIG. 3 is a block diagram illustrating an example of the configuration of the decomposition / conversion processing unit 12 (12A). In the present embodiment, the image (original image signal Sin) input via the image input unit 11 is subjected to multi-resolution conversion by the Laplacian pyramid method to be decomposed into image signal components of a plurality of different frequency bands. Shall. The Laplacian pyramid method enables rapid processing, but other methods such as wavelet transform can also be used.

[Multi-resolution decomposition processing]
As shown in FIG. 3, when a digital original image signal Sin representing an input image (original image) is input to the decomposition / conversion processing unit 12, first, filtering is performed by a filtering unit 121 constituting a low-pass filter. .

  The original image signal Sin filtered by such a low-pass filter is sampled (down-sampled) every other pixel, thereby generating a non-sharp image signal g1. The unsharp image signal g1 has a size that is ¼ that of the original image.

  Next, the interpolation unit 122 interpolates pixels having a value of 0 at the sampled interval of the unsharp image signal g1. This interpolation is performed by inserting a row and a column having a value of 0 for each column and each row of the unsharp image signal g1. Note that the non-sharp image signal interpolated in this way is in a state in which the change in signal value is not smooth because pixels having a value of 0 are inserted every other pixel. Then, after such interpolation is performed, the low-pass filter included in the interpolation unit 122 performs filtering again to obtain an unsharp image signal g1 ′. The unsharp image signal g1 ′ has a smooth change in signal value as compared with the unsharp image signal just after the interpolation.

  This non-sharp image signal g1 'is made the same size as the original image signal by performing interpolation and filtering of 0 every other pixel after making the image 1/4, and the spatial frequency of the original image signal is also reduced. The frequency higher than half has disappeared.

  Further, the above-described unsharp image signal g 1 is subjected to filtering processing by the filtering means 121. As a result, the unsharp image signal g1 is further sampled every other pixel and converted into a 1/4 (original 1/16) unsharp image signal g2. Then, the non-sharp image signal g2 is subjected to the same processing by the interpolation means 122 to generate a non-sharp image signal g2 '.

  By sequentially repeating such processing, the unsharp image signal gk (where k = 1 to L (L is an integer of 1 or more, the same shall apply hereinafter)) and the unsharp image signal gk ′ can be obtained. The unsharp image signals g1 ′, g2 ′, g3 ′,... Are unsharp image signals in a plurality of frequency bands (resolutions) having different frequency characteristics.

[Edge / noise information acquisition processing]
When a non-sharp image signal gk ′ (here, k = 1 to L) in a plurality of frequency bands is obtained by the multiresolution decomposition process, an edge noise information acquisition process is performed.

(Obtaining edge information)
The acquisition of the edge information is performed on the non-sharp image signal gn ′ in the nth frequency band as a reference among the non-sharp image signals gk ′ in a plurality of frequency bands. FIG. 3 illustrates a case where edge information is acquired in the second frequency band.

  The nth frequency band that is the edge information acquisition band is preferably a frequency band that is not the highest frequency band with a pixel size of about 0.5 mm to 1.0 mm pitch, for example, when the downsampling rate is 1/2 for an image with a pixel size of 0.175 pitch The third frequency band is most preferable. This is because the noise / contrast ratio is the highest at a resolution of 0.5mm to 1.0mm and the edge is easy to recognize with respect to the pseudo minute edge image in which Gaussian noise is added to the edge signal of about 30step / 1 ~ 2mm. is there. The edge information includes edge component information and edge direction information.

Edge Component Information Acquisition Processing Hereinafter, edge component information acquisition processing (reference numeral 126 in FIG. 3) will be described with reference to FIGS.
First, a difference image signal Mn between the unsharp image signal gn-1 'and the unsharp image signal gn' is obtained. The edge component information Ev is acquired by applying the edge component information acquisition process (reference numeral 126 in FIG. 3) shown in FIG. 4 to the difference image signal Mn.

  First, the target pixel is set in the difference image signal Mn (step S11). Next, it is determined whether or not the signal value of the target pixel is greater than or equal to a predetermined threshold Bh (Bh is a positive signal value), and if it is determined that the signal value is greater than or equal to the threshold Bh (step S12; YES), It is determined that the target pixel is a positive edge component (step S13). If it is not equal to or greater than the threshold value Bh (step S12; NO), it is determined whether or not the signal value of the pixel of interest is equal to or less than a predetermined threshold value Bl (Bl is a negative signal value). If it is determined (step S14; YES), it is determined that the target pixel is a negative edge component (step S15).

  On the other hand, if the signal value of the pixel of interest is not greater than or equal to the predetermined threshold value Bh or less than the threshold value Bl (step S14; NO), whether or not the pixel of interest is adjacent to both positive and negative edge component pixels. Is determined to be adjacent (step S16; YES), it is determined that the pixel of interest is an edge inflection point (step S17). The edge inflection point is a point in the vicinity of a signal value 0 that becomes a positive / negative transition of the edge component, and is a point that serves as a reference for the edge.

  On the other hand, when it is determined that the pixel of interest is not adjacent to the pixels of the positive and negative edge components (step S16; NO), it is determined that the pixel of interest is a non-edge component (step S18).

  When the component of the target pixel is determined, information indicating the determination result (positive edge, negative edge, edge inflection point, or non-edge) and the pixel position are associated with each other and stored in a memory (not shown). (Step S19). When the above-described steps S11 to S19 are repeatedly executed and the determination of the components for all the pixels is completed (step S20; YES), this process ends.

  FIG. 5A is a diagram schematically showing an image signal for a certain column of the difference image signal Mn. FIG. 5B is a diagram schematically illustrating the classification of edge components and non-edge components when the edge component information acquisition process is performed on the difference image signal Mn in FIG. Pixels determined as positive edges, negative edges, and edge inflection points by the above processing are extracted as edge components.

Edge direction information acquisition process (reference numeral 127 in FIG. 3)
The edge direction information Ed is obtained by applying a Sobel filter, a Prewitt filter or the like to the unsharp image signal gn ′.

中間周波数帯域でエントロピー値が所定の値より低く、最高周波数帯域でエントロピーが大きい場合、その画素位置ではノイズ成分が支配的であり、その画素位置をノイズ成分とすることができる。 (Acquisition of noise information)
Noise information acquisition processing The noise information Ns is, for example, local in the difference image signal M1 in the highest frequency band (first frequency band) and the difference image signal Mn in the intermediate frequency band (for example, the nth frequency band). It can be obtained from the variance value, the entropy value information and the pixel position information.
The entropy value Se of an image can be obtained by the following equation (Equation 1) using the probability P (z) of a pixel taking the pixel value z and the number of gradations Z of the image (“Wiley-Interscience” published “ Digital Image Processing 3rd Edition ").

When the entropy value is lower than a predetermined value in the intermediate frequency band and the entropy is large in the highest frequency band, the noise component is dominant at the pixel position, and the pixel position can be the noise component.

[Pixel value conversion processing for unsharp image signals]
When the edge information / noise information is acquired, based on the acquired edge information and / or information, the original image signal Sin and the unsharp image signal gk (k = 1 to k = L-1) of each frequency band are obtained. Pixel value conversion processing (reference numeral 123 in FIG. 3) is performed.
The pixel value conversion process performs at least one of an edge smoothing process and a noise smoothing process. It is more preferable to perform density-dependent correction processing after performing at least one of edge smoothing processing and noise smoothing processing.

The filter used for smoothing is preferably a downsampling filter used in multiresolution conversion processing or a low-pass filter with a frequency response close to it. FIG. 6 is an example of the smoothing filter coefficient (5 tap) used when the multi-resolution decomposition downsampling is performed by the Laplacian pyramid method in which binomial filtering is performed 8 times. For example, when the number of taps is 5 taps, the filter coefficient F (x) is desirably expressed by the following functions (Equation 2) (Equation 3).

  Hereinafter, edge smoothing processing, noise smoothing processing, and density-dependent correction processing will be described.

Edge smoothing process In the edge smoothing process, an edge component is obtained in the above-described edge information acquisition process of each of the original image signal Sin and the unsharp image signal gk (k = 1 to k = L-1) in each frequency band. A one-dimensional smoothing process is performed only on the edge gradient direction by a smoothing filter (see, for example, FIG. 6B) for the pixels determined to be. As a result, an image signal mainly including a noise component and a frequency component in a frequency band one level below is obtained.

Noise smoothing process In the noise smoothing process, each of the original image signal Sin and each non-sharp image signal gk (k = 1 to k = L-1) in each frequency band is not edged in the above-described edge information acquisition process. A pixel determined to be a component is subjected to a two-dimensional smoothing process using a smoothing filter (for example, see FIG. 6A), and a pixel determined to be an edge component is subjected to a smoothing filter (for example, FIG. 6). As shown in (b), a one-dimensional smoothing process is performed only in a direction other than the edge gradient direction or in a direction perpendicular to the edge gradient direction. As a result, only an image signal composed mainly of an edge component and a frequency component in a frequency band one level below is obtained.

Density-dependent correction processing Density-dependent correction processing refers to image signal density (image signal value) and contrast using a correction function shown in FIG. 7 in order to suppress artifacts and noise generated in edge enhancement / noise suppression processing. Correction obtained in accordance with (difference image signal between non-sharp image signals in adjacent frequency bands before density-dependent correction processing (for original image signal Sin, difference image signal from unsharp image signal g1 in the highest frequency band)) This is a process of subtracting the component value from the image signal value after the noise smoothing process and / or the edge smoothing process. The correction function shown in FIG. 7 is defined for each frequency band.

  In the edge enhancement / noise suppression processing, if the processing is applied strongly, the image quality may be adversely affected due to artifacts or noise amplification. In edge emphasis processing, amplification of noise components becomes a problem, and it tends to stand out in low density areas. Artifacts and noise are conspicuous in the low density portion.

  Here, in the edge enhancement / noise suppression processing, as described above, the original image signal Sin and the unsharp image signal in each frequency band are subjected to pixel value conversion processing to generate a converted image signal, and then the converted image signal and By adding a difference image signal (a difference from the unsharp image signal g1 in the highest frequency band for the converted image signal of the original image signal Sin) to the original image signal Sin from an adjacent non-sharp image signal in the adjacent frequency band, an edge is obtained. A processed image Sout that is enhanced and noise-suppressed is obtained. That is, the contrast value shown in FIG. 7 is added to the original image signal Sin. The difference image signal (contrast) is an image signal mainly composed of a noise component if it is subjected to edge smoothing, and is an image signal mainly composed of an edge component if it is subjected to noise smoothing. . As the correction amount in the correction function shown in FIG. 7 is increased so that the value of the difference image signal (contrast) is close to 0, the signal component added to the original image signal Sin decreases, that is, edge enhancement / Noise suppression processing is weakened.

FIG. 8 is a diagram showing the response of the correction function of FIG. 7 to the contrast. As shown in FIG. 8, it can be seen that the correction function shown in FIG. 7 performs stronger correction in the high contrast portion than in the low contrast portion in any frequency band. FIG. 9 is a diagram showing the response to the density of the correction function of FIG. As shown in FIG. 9, it can be seen that the correction function shown in FIG. 7 performs stronger correction in the low density part than in the high density part in any frequency band.
In this way, by performing correction that suppresses edge enhancement / noise suppression processing in a low density portion by density-dependent correction processing, it is possible to suppress artifacts and noise generated in edge enhancement and noise suppression processing. it can.

[Difference processing]
When pixel value conversion processing is performed on the original image signal Sin and the unsharp image signal gk (k = 1 to k = L−1) in each frequency band, first, the converted original image is processed by the subtractor 124 in FIG. By subtracting the unsharp image signal g1 'from the image signal S' (converted image signal), a difference image signal b0 is obtained. This subtraction is performed between corresponding pixels of the converted original image signal S ′ and the unsharp image signal g1 ′.

  Further, the subtractor 124 subtracts the unsharp image signal g2 'from the unprocessed unsharp image signal G1 (converted image signal) to obtain a difference image signal b1.

  By sequentially repeating such processing, a difference image signal bk-1 (k = 1 to k = L) is obtained. Then, the obtained difference image signal bk-1 is multiplied by an edge enhancement / noise suppression coefficient βk, which is a predetermined adjustment coefficient, to adjust the image signal component to be added to the original image signal Sin.

  FIG. 10 shows a difference signal obtained when subtracting the non-sharp image signal in the next lower frequency band from the non-sharp image signal that has been subjected to edge smoothing and / or noise smoothing and then subjected to density-dependent correction processing. FIG. 6 is a diagram showing a relationship between processing response and density when an edge enhancement / noise suppression coefficient βk (k = 1 to k = L) is applied to the image. As shown in FIG. 10, the processing response when the edge enhancement / noise suppression coefficient βk is applied to the differential image signal bk-1 (k = 1 to k = L) is lower in the low density portion where artifacts and noise are conspicuous. Thus, the strength of processing is suppressed as the density is low.

[Restore processing]
FIG. 11 shows a difference obtained by subjecting a non-sharp image signal of a plurality of frequency bands obtained by multi-resolution decomposition processing to pixel value conversion processing and subtracting the non-sharp image signal of a frequency band one level below. It is a block diagram which shows an example of a structure of the decompression | restoration process part 13 which adds the image signal bk-1 (k = 1-L) to the original image signal Sin, and carries out reverse conversion. In this embodiment, it is assumed that inverse transformation (restoration) is performed from a plurality of detail images by the Laplacian pyramid method. Although this Laplacian pyramid method enables rapid processing, other methods can also be used.

  First, the difference image signal bL-1 is interpolated between the pixels by the interpolation means 131 to obtain an image signal bL-1 'having a quadruple magnitude. Next, in the adder 132, the interpolated image signal bL-1 ′ and the unsharp image signal bL-2 having a high one-stage frequency band are added together in corresponding pixels, and the added image signal (bL-1 ′ + BL-2) is obtained.

  Next, the added image signal (bL-1 ′ + bL-2) is interpolated between the pixels by the interpolation unit 131 to obtain an image bL-2 ′ having a size four times as large. Next, the adder 132 adds the interpolated image bL-2 ′ and the unsharp image signal bL-3 having a high one-stage frequency band at the corresponding pixels, and adds the added image signal (bL-2 ′ + bL). -3) is obtained.

  The above processing is repeated. Then, this process is sequentially performed on the higher-frequency difference image signal, and finally the image signal b1 'interpolated by the adder 132 and the difference image signal b0 having the highest resolution are added to the original image by the adder 133. The processed image signal Sout is obtained by adding the signal Sin.

[Image output processing]
As described above, the processed image signal Sout subjected to the decomposition / conversion processing and the restoration processing is printed or displayed as a visible image in the image output unit 14. Further, it is stored as digital data in various disk devices.

  As described above, according to the image processing apparatus 10, a plurality of frequency bands are obtained based on edge information and / or noise information acquired in a reference frequency band obtained by performing multi-resolution decomposition processing on an image signal. The edge component and / or noise component is obtained by applying at least one of noise smoothing processing and edge smoothing processing to the unsharp image signal and the original image signal of the image, and the density depends on the obtained edge component and / or noise component. Correction image processing is performed to obtain a converted image signal. Then, a difference image signal is obtained from each of the obtained converted image signals and a non-sharp image signal in an adjacent frequency band (the highest frequency band for the original image signal), and predetermined adjustment is performed for each difference image signal. Multiply the coefficients and add each to the original image signal to obtain a processed image signal.

  Therefore, it is possible to enhance edge components such as the outline of a structure in an image signal and / or reduce noise components while suppressing the occurrence of noise and artifacts due to edge enhancement / noise suppression processing.

<Second Embodiment>
Hereinafter, a second embodiment of the present invention will be described.
In the second embodiment, the processing in the decomposition / conversion processing unit 12 is different from that in the first embodiment.

  FIG. 12 is a block diagram illustrating an example of the configuration of the decomposition / conversion processing unit 12 (12B) according to the second embodiment. In the present embodiment, the image (original image signal Sin) input via the image input unit 11 is subjected to multi-resolution conversion by the Laplacian pyramid method to be decomposed into image signal components of a plurality of different frequency bands. Shall. The Laplacian pyramid method enables rapid processing, but other methods such as wavelet transform can also be used.

[Multi-resolution decomposition processing]
As shown in FIG. 12, when a digital original image signal Sin representing an input image (original image) is input to the decomposition / conversion processing unit 12, first, it is filtered by a filtering means 121 constituting a low-pass filter. .

  The original image signal Sin filtered by such a low-pass filter is sampled (down-sampled) every other pixel, thereby generating a non-sharp image signal g1. The unsharp image signal g1 has a size that is ¼ that of the original image.

  Next, the interpolation unit 122 interpolates pixels having a value of 0 at the sampled interval of the unsharp image signal g1. This interpolation is performed by inserting a row and a column having a value of 0 for each column and each row of the unsharp image signal g1. Note that the non-sharp image signal interpolated in this way is in a state in which the change in signal value is not smooth because pixels having a value of 0 are inserted every other pixel. Then, after such interpolation is performed, the low-pass filter included in the interpolation unit 122 performs filtering again to obtain an unsharp image signal g1 ′. The unsharp image signal g1 ′ has a smooth change in signal value as compared with the unsharp image signal just after the interpolation.

  This non-sharp image signal g1 'becomes the same size as the original image signal Sin and the original image signal by performing interpolation and filtering of 0 every other pixel after making the image 1/4, and the original image signal Sin. The frequency higher than half of the spatial frequency is disappeared.

  Further, the above-described unsharp image signal g 1 is subjected to filtering processing by the filtering means 121. As a result, the unsharp image signal g1 is further sampled every other pixel and converted into a 1/4 (original 1/16) unsharp image signal g2. Then, the non-sharp image signal g2 is subjected to the same processing by the interpolation means 122 to generate a non-sharp image signal g2 '.

  By sequentially repeating such processing, the unsharp image signal gk (here, k = 1 to L) and the unsharp image signal gk ′ can be obtained. The unsharp image signals g1 ', g2', g3 ', ... are unsharp image signals in a plurality of frequency bands having different frequency characteristics.

[Edge / noise information acquisition processing]
When the non-sharp image signal gk ′ (where k = 1 to L) in a plurality of frequency bands is obtained by the multi-resolution decomposition process, any one of the non-sharp image signals gk ′ in the plurality of frequency bands is obtained. Edge noise information acquisition processing is performed on the unsharp image signal gn ′ in the nth frequency band. Although the functional configuration of the edge / noise information acquisition unit is not shown in FIG. 12, edge component information, edge direction information, and noise information are obtained by processing similar to the edge information / noise information acquisition processing described in the first embodiment. Get.

  When the edge component information, edge direction information, and noise information are acquired, based on the acquired information, the original image signal Sin and the unsharp image signal gk (k = 1 to k = L-1) in each frequency band are obtained. Pixel value conversion processing is performed. In the second embodiment, edge smoothing processing, density-dependent correction processing P1, noise, and the like are performed on the original image signal Sin and the unsharp image signal gk (k = 1 to k = L-1) in each frequency band. Smoothing processing and density-dependent correction processing P2 are performed separately.

  The edge smoothing process and the noise smoothing process are performed by the same process as described in the first embodiment. In the edge smoothing process, an image signal mainly including a noise component and a frequency component in a frequency band one level below is obtained. In the noise smoothing process, only an image signal mainly composed of an edge component and a frequency component in a frequency band one level below is obtained.

  Edge smoothing processing and noise smoothing processing are separately performed on the original image signal Sin and the non-sharp image signal gk (k = 1 to k = L-1) of each frequency band, and edge components and noise components are extracted separately. As a result, it is possible to individually perform density-dependent correction processing or the like on each of the edge component and the noise component.

  The density-dependent correction process is substantially the same as that described in the first embodiment, but when the density correction-dependent process is performed on a non-sharp image signal mainly composed of edge components after the noise smoothing process. Prepare a correction function to be used and a correction function to be used when density-dependent correction processing is applied to a non-sharp image signal mainly composed of noise components after noise smoothing. Different corrections can be made.

FIG. 13 is a diagram showing the response to the density of the correction function used for the non-sharp image signal mainly composed of edge components after the noise smoothing process. As shown in FIG. 13, it can be seen that stronger correction is applied in the low density part than in the high density part, and the edge enhancement processing is applied more strongly in the high density part.
FIG. 14 is a diagram showing the response to the density of the correction function used for the non-sharp image signal mainly composed of noise components after the edge smoothing process. As shown in FIG. 14, it can be seen that stronger correction is applied in the high density part than in the low density part, and that the noise suppression process is applied more strongly in the low density part.

As described above, in the low density portion where noise is conspicuous, the noise can be made inconspicuous by performing correction so as to weaken the edge emphasis processing by applying the noise suppression processing stronger. On the other hand, noise and artifacts due to the noise suppression process can be reduced by performing a correction that weakens the noise suppression process and strengthens the edge enhancement process in the high density portion where the noise is not noticeable.
[Difference processing]
When edge smoothing processing and density-dependent correction processing are performed on the original image signal Sin and the unsharp image signal gk (k = 1 to k = L-1) in each frequency band, first, a subtracter 128 shown in FIG. , The non-sharp image signal g1 ′ is subtracted from the original image signal SN ′ (first converted image signal) that has been subjected to the edge smoothing process and the density-dependent correction process. This subtraction is performed between corresponding pixels of the original image signal SN 'and the unsharp image signal g1'. Then, the difference image signal N0 obtained by multiplying the difference signal mainly composed of noise components obtained by subtraction by a predetermined noise adjustment coefficient (noise control coefficient) βN1 (0>βNk> −1) is adjusted to attenuate the noise components. Is obtained.

  Similarly, the subtractor 129 subtracts the unsharp image signal g1 ′ from the original image signal SE ′ (second converted image signal) that has been subjected to the noise smoothing process and the density-dependent correction process. This subtraction is performed between the corresponding pixels of the original image signal SE ′ and the unsharp image signal g1 ′ that have been subjected to the noise smoothing process and the density-dependent correction process. Then, a difference signal mainly composed of edge components obtained by subtraction is multiplied by a predetermined edge adjustment coefficient (edge control coefficient) βE1 (βEk> 1) to obtain a difference image signal E0 in which the edge components are emphasized. It is done. By adding the difference image signal N0 and the difference image image signal E0 by the adder 130, the difference image signal B0 is obtained.

  Further, the subtracter 128 subtracts the unsharp image signal g2 ′ from the unsharp image signal gN1 ′ (first converted image signal) that has been subjected to the edge smoothing process and the density-dependent correction process. This subtraction is performed between the corresponding pixels of the unsharp image signal gN1 'and the unsharp image signal g2'. Then, a difference signal mainly composed of noise components obtained by subtraction is multiplied by a predetermined noise control coefficient βN2, and a difference image signal N1 whose noise components are attenuated is obtained.

  Similarly, the subtracter 129 subtracts the unsharp image signal g2 ′ from the unsharp image signal gE1 ′ (second converted image signal) that has undergone the noise smoothing process and the density-dependent correction process. This subtraction is performed between the corresponding pixels of the unsharp image signal gE1 'and the unsharp image signal g2'. Then, a difference signal mainly composed of edge components obtained by subtraction is multiplied by a predetermined edge control coefficient βE1 to obtain a difference image signal E1 in which the edge components are emphasized and adjusted. The difference image signal B1 is obtained by adding the difference image signal N1 and the difference image signal E1 by the adder 130.

  By sequentially repeating such processing, a difference image signal Bk-1 (k = 1 to k = L) is obtained. The obtained difference image signal Bk-1 is stored in an image memory (not shown).

  FIG. 15 shows a difference signal mainly composed of edge components obtained by subtracting a non-sharp image signal in the next lower frequency band from a non-sharp image signal that has been subjected to noise smoothing processing and density-dependent correction processing. The figure which shows the relationship between the processing response (edge component processing response in FIG. 15) and the density when the control coefficient βEk is applied, and the frequency one level lower than the non-sharp image signal which has been subjected to the edge smoothing process and the density dependent correction process 15 shows the relationship between the processing response (noise component processing response in FIG. 15) and density when the noise control βNk is applied to the difference signal mainly composed of noise components obtained when subtracting the non-sharp image signal in the band. FIG. As shown in FIG. 15, as a result of performing the density-dependent correction processing separately for the noise component and the edge component, the noise suppression processing is strengthened in the low density portion, and the edge enhancement processing is suppressed. Thereby, noise can be made inconspicuous in the low density portion. In a high density portion where noise is not noticeable, the noise suppression processing is weakened and the edge enhancement processing is strong. Thereby, in the high density part, it is possible to weaken the noise suppression process and suppress the artifacts caused by the noise process, and to apply the edge enhancement strongly.

  Although the edge component exists in any frequency band, the noise component is mainly a high-frequency component, so that if the noise suppression process is strongly applied from the low-frequency band, the image is blurred. Therefore, the noise control coefficient βNk to be applied to the image signal acquired by the edge smoothing process and the edge control coefficient βEk to be applied to the image signal acquired by the noise smoothing process are individually set, and the edge control coefficient βEk is set to the low frequency band. 16 is performed by performing edge enhancement from the low frequency band to the high frequency band from the low frequency band to the high frequency band and performing noise suppression only on the high frequency band by applying the noise control coefficient βNk to the noise component of only the high frequency band. Thus, different frequency responses can be obtained for the edge component and the noise component, and it is possible to suppress blurring of the image by suppressing noise suppression processing in the low frequency band.

[Restore processing]
In the restoration processing unit 13, the difference image signal Bk−1 (k = 1 to k = L) acquired by the decomposition / conversion processing unit 12 is added to the original image signal Sin for inverse conversion, and the image signal Sout is output. To do. The configuration of the restoration processing unit 13 is the same as that described with reference to FIG.

  The restored image signal Sout is printed or displayed as a visible image in the image output unit 14. Further, it is stored as digital data in various disk devices.

  As described above, according to the image processing apparatus 10, a plurality of frequency bands are obtained based on edge information and / or noise information acquired in a reference frequency band obtained by performing multi-resolution decomposition processing on an image signal. The non-sharp image signal and the original image signal are subjected to noise smoothing processing and density-dependent correction processing to obtain a first converted image signal mainly composed of edge components, and adjacent to each of the first converted image signals. A difference image signal with a non-sharp image signal in a matching frequency band (the highest frequency band for the original image signal) is acquired, and an edge control coefficient is applied to each difference image signal to enhance edge components. Further, based on edge information and / or noise information acquired in a reference frequency band obtained by multiresolution decomposition processing of an image signal, edge smoothing is performed on unsharp image signals and original image signals in a plurality of frequency bands. To obtain a second converted image signal mainly composed of noise components, and a frequency band adjacent to each of the second converted image signals (maximum for the original image signal). A difference image signal with a non-sharp image signal in a frequency band is acquired, and noise components are suppressed by multiplying each difference image signal by a noise control coefficient. Then, the obtained edge-enhanced difference image signal and noise suppression difference image signal are added to the original image signal to obtain a processed image.

  Therefore, it is possible to enhance edge components such as contours of structures in the image signal and attenuate noise components while suppressing the generation of noise and artifacts due to edge enhancement / noise suppression processing.

  In the above embodiment, the Laplacian pyramid method is used to perform the conversion to the multi-resolution space and the inverse conversion from the multi-resolution space. By performing this transformation and inverse transformation by wavelet transformation, it is possible to perform enhancement processing in any direction (vertical direction, horizontal direction, diagonal direction).

  In the first and second embodiments, the difference image signal is acquired in all frequency bands and added to the original image signal. However, the present invention is not limited to this, and at least one frequency is used. Edge enhancement and noise suppression can be performed by obtaining a differential image signal in the band and performing edge enhancement and / or noise suppression processing.

  In addition, the image processing method and the image processing apparatus according to the above-described embodiments are suitable for performing enhancement processing on a radiation image, an MRI image, an CT image, and other various images.

  In addition, the detailed configuration and detailed operation of each device constituting the image processing apparatus 10 can be changed as appropriate without departing from the spirit of the present invention.

It is a block diagram which shows the functional structure of the image processing apparatus 10 which concerns on this invention. 4 is a flowchart showing a processing procedure for obtaining a processed image signal Sout from an original image signal Sin by a decomposition / conversion processing unit 12 and a restoration processing unit 13. It is a block diagram which shows the structural example of the decomposition | disassembly / conversion process part 12 (12A) in 1st Embodiment. 6 is a flowchart illustrating an example of edge component information acquisition processing executed in a decomposition / conversion processing unit 12. (A) is a figure which shows typically the image signal for a certain row | line | column of a difference image signal, (b) is the edge component at the time of performing an edge component information acquisition process to the difference image signal of (a), and non- It is a figure which shows the classification | category of an edge component typically. (A) is a figure which shows an example of the two-dimensional smoothing filter coefficient (5tap) used when downsampling of multiresolution decomposition | disassembly is performed by the Laplacian pyramid method which performs binomial filtering 8 times, (b), It is a figure which shows an example of the one-dimensional smoothing filter coefficient (5tap) used when the downsampling of multiresolution decomposition | disassembly is performed by the Laplacian pyramid method which performs binomial filtering 8 times. It is a figure which shows an example of the correction function used by a density | concentration dependence correction process. It is a figure showing the response with respect to the contrast of the correction function of FIG. It is a figure showing the response with respect to the density | concentration of the correction function of FIG. Edge enhancement is performed on the difference signal obtained by subtracting the unsharp image signal in the next lower frequency band from the unsharp image signal that has been subjected to edge smoothing and / or noise smoothing and then subjected to density-dependent correction. It is a figure which shows the relationship between the processing response when applying the noise suppression coefficient βk (k = 1 to k = L) and the density. 3 is a block diagram illustrating a configuration example of a restoration processing unit 13. FIG. It is a block diagram which shows the structural example of the decomposition | disassembly / conversion process part 12 (12B) in 2nd Embodiment. It is a figure showing the response with respect to the density | concentration of the correction function used for the non-sharp image signal mainly consisting of an edge component after a noise smoothing process. It is a figure showing the response with respect to the density | concentration of the correction function used for the non-sharp image signal mainly consisting of a noise component after an edge smoothing process. The edge control coefficient βEk is added to the difference signal consisting mainly of edge components obtained when subtracting the unsharp image signal in the next lower frequency band from the unsharp image signal that has been subjected to noise smoothing processing and density-dependent correction processing. Figure showing the relationship between processing response and density when applied, and subtracting the unsharp image signal in the next lower frequency band from the non-sharp image signal that has undergone edge smoothing processing and density-dependent correction processing It is a figure which shows the relationship between a process response when a noise control (beta) Nk is applied to the difference signal which mainly consists of a noise component, and a density | concentration. The edge control coefficient βEk is applied to each edge component in the low frequency band to the high frequency band, edge enhancement is performed from the low frequency band to the high frequency band, and the noise control coefficient βNk is applied to the noise component in the high frequency band only to suppress noise only in the high frequency band. It is a figure which shows an example of the frequency response of the edge component obtained by performing, and the frequency response of a noise component.

Explanation of symbols

DESCRIPTION OF SYMBOLS 10 Image processing apparatus 11 Image input part 12 Disassembly / conversion process part 13 Restoration process part 14 Image output part

Claims (14)

A resolution processing step of obtaining multi-resolution conversion of the input image signal to obtain non-sharp image signals of a plurality of different frequency bands; and at least one of the input image signal and / or the non-sharp image signals of the plurality of frequency bands A conversion processing step for obtaining a difference image signal by taking a difference between the converted image signal obtained by performing the conversion process and an image signal in the frequency band adjacent to the converted image signal or an image signal in the highest frequency band; and the difference image signal And a restoration processing step of obtaining a processed image by adding to the input image signal,
The conversion process step includes
An information acquisition step of acquiring edge information and / or noise information of the input image signal;
Based on the acquired edge information and / or noise information, at least one of an edge smoothing process or a noise smoothing process is performed on at least one of the input image signal and / or the unsharp image signal of the plurality of frequency bands. After performing two processes, performing a density-dependent correction process to obtain the converted image signal;
An image processing method comprising: The conversion step includes
Based on the acquired edge information and / or noise information, at least one of the input image signal and / or the non-sharp image signal of the plurality of frequency bands is subjected to a noise smoothing process, and the noise smoothing is performed. A step of performing density-dependent correction processing on the image signal to obtain the first converted image signal; the first converted image signal and an image signal in a frequency band adjacent to the converted image signal or an image signal in the highest frequency band; And obtaining a differential image signal in which the edge component is adjusted by applying a predetermined edge adjustment coefficient to the differential signal after obtaining a differential signal mainly composed of edge components by taking the difference of When,
Based on the acquired edge information and / or noise information, at least one of the input image signal and / or the non-sharp image signal of the plurality of frequency bands is subjected to an edge smoothing process, and the edge smoothing is performed. A step of performing density-dependent correction processing on the image signal to obtain the second converted image signal; the second converted image signal and an image signal in a frequency band adjacent to the converted image signal or an image signal in the highest frequency band; And obtaining a differential image signal in which the noise component is adjusted by applying a predetermined noise adjustment coefficient to the differential signal after obtaining a differential signal mainly composed of a noise component by taking the difference of When,
The image processing method according to claim 1, further comprising:   The edge information includes at least one of information indicating a pixel position of an edge component, a sign of a signal value, a direction of the edge component, and a pixel position of an edge inflection point. The noise information is a noise component. The image processing method according to claim 1, further comprising: at least one of information indicating a local variance value, an entropy value, and a pixel position.   In the noise smoothing process, two-dimensional smoothing is performed for pixels at positions that are not edge components, and one-dimensional only for directions other than the edge gradient direction or directions perpendicular to the edge gradient direction for pixels of edge components. The image processing method according to claim 1, wherein a smoothing process is performed.   5. The image processing method according to claim 1, wherein the edge smoothing process is a one-dimensional smoothing process performed on edge-element pixels only in an edge gradient direction. 6.   The image processing method according to claim 1, wherein the multi-resolution conversion uses a Laplacian pyramid method.   6. The image processing method according to claim 1, wherein the multi-resolution conversion uses wavelet conversion. Decomposition processing means for acquiring non-sharp image signals of a plurality of different frequency bands by performing multi-resolution conversion of the input image signal; and at least one of the input image signals and / or the non-sharp image signals of the plurality of frequency bands A conversion processing means for obtaining a difference image signal by taking a difference between the converted image signal obtained by performing the conversion process and an image signal in the frequency band adjacent to the converted image signal or an image signal in the highest frequency band; and the difference image signal In an image processing apparatus comprising restoration processing means for obtaining a processed image by adding to the input image signal,
The conversion processing means includes
Information acquisition means for acquiring edge information and / or noise information of the input image signal;
Based on the acquired edge information and / or noise information, at least one of an edge smoothing process or a noise smoothing process is performed on at least one of the input image signal and / or the unsharp image signal of the plurality of frequency bands. Means for obtaining the converted image signal by performing density-dependent correction processing after performing two processes;
An image processing apparatus comprising: The converting means includes
Based on the acquired edge information and / or noise information, at least one of the input image signal and / or the non-sharp image signal of the plurality of frequency bands is subjected to a noise smoothing process, and the noise smoothing is performed. Means for performing a density-dependent correction process on the image signal to obtain the first converted image signal; the first converted image signal and an image signal in a frequency band adjacent to the converted image signal or an image signal in the highest frequency band; And obtaining a difference image signal in which the edge component is adjusted by applying a predetermined edge adjustment coefficient to the difference signal after obtaining a difference signal mainly composed of edge components by taking the difference of When,
Based on the acquired edge information and / or noise information, at least one of the input image signal and / or the non-sharp image signal of the plurality of frequency bands is subjected to an edge smoothing process, and the edge smoothing is performed. Means for performing a density-dependent correction process on the image signal to obtain the second converted image signal; the second converted image signal and an image signal in a frequency band adjacent to the converted image signal or an image signal in the highest frequency band; A noise component adjusting means comprising: obtaining a difference signal mainly composed of noise components by taking a difference between the difference signals and obtaining the difference image signal in which the noise components are adjusted by applying a predetermined noise adjustment coefficient to the difference signal When,
The image processing apparatus according to claim 8, further comprising:   The edge information includes at least one of information indicating a pixel position of an edge component, a sign of a signal value, a direction of the edge component, and a pixel position of an edge inflection point. The noise information is a noise component. The image processing apparatus according to claim 8, further comprising at least one of information indicating a local variance value, an entropy value, and a pixel position.   In the noise smoothing process, two-dimensional smoothing is performed for pixels at positions that are not edge components, and one-dimensional only for directions other than the edge gradient direction or directions perpendicular to the edge gradient direction for pixels of edge components. The image processing apparatus according to claim 8, wherein a smoothing process is performed.   The image processing apparatus according to any one of claims 8 to 11, wherein the edge smoothing process performs a one-dimensional smoothing process only in an edge gradient direction on an edge component pixel.   The image processing apparatus according to claim 8, wherein the multi-resolution conversion uses a Laplacian pyramid method.   The image processing apparatus according to claim 8, wherein the multi-resolution conversion uses wavelet conversion.

Images