JP2000101871A - Contour emphasis processing circuit - Google Patents

Abstract

PROBLEM TO BE SOLVED: To make an edge part steep by preventing insufficient edge emphasis and occurrence of ringing caused in the vicinity of the edge. SOLUTION: Class separation relating to an edge structure is applied to edge signals extracted from signals resulting from applying wavelet transformation to a received image signal. A high frequency component signal is added to an edge part depending on a class to which the edge part of the image signal receiving wavelet conversion belongs by referencing a class information table (class rule table) where cross reference between each class and each high frequency component signal to be added to the edge part of the image signal that is wavelet-transformed is decided in advance, and the resulting edge receives inverse wavelet transformation and the result is used for contour emphasis.

Description

DETAILED DESCRIPTION OF THE INVENTION

[0001]

[0001] 1. Field of the Invention [0002] The present invention relates to an apparatus for correcting image quality of an image, and more particularly to a circuit for performing edge restoration of an image.

[0002]

2. Description of the Related Art Conventionally, edge enhancement processing for contour correction has been performed mainly by adding a correction signal generated from an input image signal using a linear filter to an edge portion. "Digital Television Technology" edited by NHK Broadcasting Research Institute
(Japan Broadcasting Publishing Association) pp. FIG. 10A shows the operation and circuit configuration of the conventional edge enhancement processing shown in FIGS.
To (d). Here, by adding a correction signal (b) generated from the input image signal (a) to the edge portion (c), the inclination of the edge is steep, and the emphasis process is performed. An example of a typical edge emphasis circuit for realizing this is shown in FIG. This method is a high-frequency emphasis process in which a component created from an input image signal by performing a filter process having a high-pass characteristic is added to the input image signal, and is not an edge restoration in a strict sense.

[0003]

In the conventional edge enhancement processing using the correction signal generated from the input image signal by the linear filter as described above, the sharpening of the edge is not sufficient or ringing occurs near the edge. There were problems such as.

That is, in the edge enhancement of the method shown in FIGS. 10A to 10D, since a correction signal is generated from an input image signal, a gentle edge cannot be sufficiently enhanced. Further, in this method, since the same emphasis is performed on the entire image, emphasis is performed so that a certain portion in the image becomes clear, and as a result, interference such as ringing may occur in other portions.

[0005] In an image subjected to enlargement processing, the highest frequency of the image before processing is insufficient for the highest frequency that the resolution after enlargement can have. That is, since the high-frequency component required for the edge enhancement processing does not exist in the image subjected to the enlargement processing, the high-frequency component at the resolution after the enlargement cannot be enhanced, and the conventional contour correction causes a problem. .

SUMMARY OF THE INVENTION An object of the present invention is to provide a correction signal generated by using a linear filter from an input image signal to an edge portion, as in a conventional edge enhancement process, which occurs in insufficient edge enhancement or near an edge. An object of the present invention is to provide a contour emphasis processing circuit which does not have a problem such as ringing and aims to sharpen an edge by a novel method.

[0007]

In order to achieve the above object, the present invention provides a contour emphasis processing circuit comprising: a wavelet transform circuit for performing a wavelet transform on an input image signal; and an edge extracted from a signal subjected to a wavelet transform by the wavelet transform circuit. Classifying means for classifying an edge structure with respect to a signal, each class classified by the classifying means, and a high-frequency component signal to be added to an edge portion of the wavelet-transformed image signal for edge enhancement Is stored in advance as a class information table and the class information table read from the memory is referred to, and the wavelet-transformed image signal is added according to the class to which the edge part belongs. High-frequency component signal addition cycle to add power high-frequency component signal , And is characterized in that at least comprises an inverse wavelet transform circuit for inverse wavelet transform output signal of the high frequency component signal adding circuit.

In the contour enhancement processing circuit according to the present invention, the class information table may perform wavelet transform on each of a low-resolution image created by band-limiting a high-resolution image for learning and a high-resolution image for learning. Edge signals are extracted from the obtained signals, edge structures are classified for the extracted edge signals, and the classification of the edges corresponding to the low-resolution image and the high-resolution image obtained by the classification is performed. The edge structure and the wavelet transform component in the edge portion composed of the low-frequency component and the high-frequency component are tabulated.

In the contour emphasizing circuit according to the present invention, the high-frequency component signal added by the high-frequency component signal adding circuit may be a high-frequency component signal selected from the class information table based on a type of the input image signal. It is a signal.

Further, the contour emphasis processing circuit of the present invention is further interposed between the high frequency component signal adding circuit and the inverse wavelet transform circuit, and performs the contour emphasis processing according to the edge structure of the input image signal. An adaptive emphasis processing means for performing adaptive switching is provided.

[0011]

DESCRIPTION OF THE PREFERRED EMBODIMENTS The present invention will be described in detail below based on embodiments of the present invention with reference to the accompanying drawings. The processing circuit of the present invention is a circuit for improving the sharpness of an image by enhancing the outline of an image signal. From the viewpoint of a signal processing method in the circuit, the processing circuit removes high-frequency components of edges extracted from a learning image. It is to make a table, and to adaptively perform edge restoration of an image using the obtained table.

As described above, the edge enhancement processing circuit according to the present invention adaptively performs edge restoration of an image.
In particular, conventional high-frequency emphasis processing that adds a high-frequency component within the band to the point that sharpening of the edge is performed by adding a high-frequency component outside the band to the edge that has become gentle due to band limitation or the like. Is different from The outline emphasis processing circuit according to the present invention is suitable for use in an image quality correction device used in a broadcasting station or an image quality correction circuit in a television receiver of a general household.

As described above, in order to tabulate the high-frequency components of the edges extracted from the learning image and to adaptively perform the edge restoration of the image using the obtained table, the present invention first employs the learning Is prepared and passed through a low-pass filter to produce a low-resolution image. In general, the correspondence between the edge structure when the same subject is photographed at high resolution and the edge structure when photographing at the low resolution differs depending on the image type. That is, depending on how the edges included in the image are formed, the low-resolution waveform structure may be the same, but the high-resolution waveform structure may be different. For this reason, in order to actually apply the processing, it is necessary to use a learning image similar in waveform structure to the processing target, but after learning as many edge structure changes as possible using as many images as possible, This can be generalized by performing a process of switching the application range according to the type.

Here, switching the application range according to the image type means that it is necessary to add a different high-frequency component according to the edge structure depending on the image. For example, a character image, an electronically created image, an animation, etc. In general, the edge is sharp, and the level changes stepwise before and after the edge. On the other hand, a sharp edge does not exist in a natural image. Also, even within the same image, areas are divided (segmented), and the type of each area is, for example, an area containing many characters,
If it is known before the processing that the area is the one into which the image of the camera has been fitted, the corresponding rule can be switched accordingly.

The determination of the image type can be made to some extent automatically by examining the statistical properties of the image. However, in the case of an image containing various properties, the automatic determination is difficult, and a corresponding rule that provides an appropriate sharpness It is realistic and effective for a human to switch while viewing the image to be processed. In the present invention,
The subsequent processing is performed on the assumption that the type of the image or the area is determined in advance. However, the determination of the type of the image or the region is not an essential matter of the present invention, and a detailed description is omitted.

The characteristics of the low-pass filter for producing a low-resolution image from a high-resolution image are adapted to the band-limiting characteristics of a low-resolution image (input image) to be actually processed. For example, this edge restoration processing is performed by NT
When the resolution of the SC signal is used as the process of converting the resolution of the HDTV signal, the high-resolution image corresponds to the HDTV image, and the low-resolution image corresponds to the NTSC image.
What imitates the band limiting characteristic of the NTSC signal will be used.

Further, the specific signal processing in the processing circuit of the present invention is as follows. First, a high-resolution image prepared for learning and a low-resolution image created therefrom are each subjected to wavelet transform to obtain a transform component at a corresponding edge portion. (A high-resolution scale component and a low-resolution scale component) are obtained. Also, based on the structure of the edge portion of each image, the learning image is patterned in a plurality of ways to classify how the edge has deteriorated, and a pair of the class information and the corresponding conversion component is classified. Are converted into a table, and the correspondence between them (class information and conversion components) is obtained in advance as a class information table.

Here, the wavelet transform will be described. Wavelet transform is a “wavelet (wavelet; sazanami)” that converts video and audio signals to a finite width.
Is a method of expressing by superimposing. The Fourier transform, which expresses a signal as a superposition of sine waves that continues indefinitely, was not suitable for knowing where in time there was a sudden change in the signal,
The wavelet transform allows for both frequency and time analysis.

In general, a wavelet is an operation (scale conversion) of multiplying a function Φ (t), which is called a basic wavelet, which exists only in a limited range in terms of time and frequency, on the time axis (scale conversion) and time. Can be shifted by b (shift conversion)

(Equation 1) A set of functions like With this function, the frequency and time components of the signal corresponding to the parameters a and b can be extracted, and this operation is called wavelet transform. As will be described later, edges can be easily extracted by selecting a basic wavelet.

Returning to the present invention, the description will be continued. A high-frequency component (high-resolution scale component) is added to the edge of the input image so that the edge of the input image is applied to the table and a sharp edge before deterioration is restored. further,
In human visual characteristics, ringing that occurs along isolated edge structures is easy to see, while ringing at edges adjacent to complex uneven structures is difficult to see,
Considering that it is effective to switch the edge enhancement process according to the waveform structure around the edge to be emphasized in the high frequency range, whether the edge to be emphasized in high frequency is in contact with the flat waveform structure, etc. Ito, Yagi et al .: "Investigation of Image Enhancement Processing Using Wavelet Transform", which investigated whether the edge was close to the edge of the image and realized adaptive enhancement processing, Proceedings of the 23rd Annual Conference of the Institute of Image Information and Television Engineers 23- 2,
P. 280 (1997), the restoration process may be adaptively performed. That is, as in the technique described in the above-mentioned Ito and Yagi et al. Documents, a wavelet transform component corresponding to each edge is individually subjected to correction (emphasis processing) according to an edge structure and reconstructed (inverse transformation). Thereby, adaptive enhancement processing may be performed.

Here, even though the waveform structure of the edge of the input image is the same, the waveform structure of the edge of the high-resolution image to be restored may be different. From the groups having the same edge waveform structure of the high-resolution image but different from each other, the range to be applied as the correspondence relationship between the edge structures before and after the restoration is performed using a rule table determined in advance by the image type.

The wavelet transform used in the present invention is defined as the following equation (1) where the input waveform is f (x).

(Equation 2) Where Φ is the wavelet, s is the scale (intensity),
X represents a position on f (x). In the discrete wavelet transform, analysis is performed as s = 2 ^j , X = n (j, n: integer), and the low-frequency component, the component at j = 2, and the component at j = 1 are band-analyzed as shown in FIG. Has been split.

According to the present invention, a wavelet transform is performed on the high-resolution image and the low-resolution image in the horizontal and vertical directions of the image with a hardware configuration as shown in FIG. 2 by using a filter for realizing a one-dimensional wavelet transform. Do. Here, when frequency components are divided as shown in FIG.
The cutoff frequency of each filter in FIG.
1, HH1, horizontal sampling frequency / 4, VL1, V
H1 has a vertical sampling frequency of / 4, HL2 and HH2 have a horizontal sampling frequency of / 8, and VL2 and VH2 have a vertical sampling frequency of / 8. Thereby, each image can be divided into horizontal and vertical frequency components. In FIG. 2, the LL component refers to the horizontal low-frequency component and the vertical low-frequency component, and the LH component refers to the horizontal low-frequency component and the vertical high-frequency component.

By performing the wavelet transform in this manner, the horizontal edge and the vertical edge can be analyzed on the one-dimensional axis in each of the horizontal and vertical directions. In the case of the horizontal edge, since there is no change in the horizontal direction and there is a sharp change in the vertical direction, the high frequency component in the vertical direction, that is, the LH component corresponds to the high-resolution scale component (j = 1 component) of the horizontal edge. The middle band component in the direction, that is, the LLLH component corresponds to the low-resolution scale component (j = 2 component) of the horizontal edge.

In the case of a vertical edge, there is no change in the vertical direction and there is a sharp change in the horizontal direction. Therefore, the high-frequency component in the horizontal direction, that is, the HL component, becomes the high-resolution scale component (j = 1 component) of the vertical edge. The LLHL component in the horizontal direction corresponds to the low-resolution scale component (j = 2 component) of the vertical edge.

Similarly, the HH component is a high-resolution scale component (j = 1 component) of the oblique edge, and the LLHH component is a low-resolution scale component (j = 2 component) of the oblique edge. Here, an example in which the following processing is performed in one dimension each in the horizontal and vertical directions by using such components for the horizontal edge and the vertical edge will be described with one-dimensional processing in mind.

FIG. 3 shows an example of the above-described wavelet transform, and shows a one-dimensional horizontal multi-scale component. In FIG. 3, the horizontal axis represents the horizontal position of the image, and the vertical axis represents the signal level.

The wavelet transform of each of the high-resolution image and the low-resolution image is a horizontal or vertical one.
It is obtained by multiplying the band characteristics of the high-resolution image and the low-resolution image by the divided band characteristics of FIG. That is, the high-resolution image has components up to the highest frequency in FIG. 1, and the low-resolution image is band-limited. If the low-resolution image has half the bandwidth of the high-resolution image, the component of j = 1 is attenuated due to the band limitation, and the low-resolution image has a component of j = 2 or less. On the other hand, a high-resolution image has components of j = 1 and j = 2 in addition to low-frequency components.

In the above, a Laplacian type (secondary differential type) as shown in FIG. 4 is used as a basic wavelet. The coefficients are (-0.25, 0.5, -0.25). Wavelet transform using this is equivalent to second-order differentiation of an image, and a zero-crossing point of an edge can be easily detected from a transform component.

Further, in the present invention, the edge structure is classified and emphasized by using the correspondence of the wavelet transform components between the high-resolution image and the low-resolution image. This will be described below. FIGS. 5 (a) and 5 (b)
9 shows an example of classification of an edge structure. FIG. 5 (a)
In (b) and (b), there are two types of edges to be emphasized in the high frequency band, in which the inclination is positive (Up) or negative (Down). The edge structure is classified based on a combination of an edge to be emphasized in the high frequency band and a waveform structure adjacent to the left and right sides. Here, the structure adjacent to the left and right has a positive slope /
There may be a flat structure (Flat) in addition to the negative type. The patterns of these waveform structures and Up, Down
The edge strength in the case of is quantized to form a class. As described later, the edge intensity uses the absolute value of the luminance level difference between the pixels adjacent to each other with the zero crossing of the edge interposed therebetween.

As shown in FIG. 5A, it is assumed that the slope of the edge to be emphasized in the high frequency range is positive (Up), the quantized intensity is 2, and the left and right sides have a flat structure. ,
This edge class is defined as FU2-F (F: Flat,
U: Up, D: Down).

By using such a class expression, a change in the edge structure can be represented. For example, when it is desired to transform the edge structure shown in FIG. 5A of the low-resolution image into the high-resolution edge structure shown in FIG. The correspondence of the edge class when the size is increased can be expressed by the following equation (2) (F-U2-F) → (F-U3-F) (2).

When the edge structure as shown in FIG. 5B is transformed into the edge structure as shown in FIG.
-The edge structure of the class of U2-D1 is changed to the edge structure of the class of F-U4-D2. In this way, by setting the correspondence between the classes, the deformation of the edge structure in the edge restoration can be described using the class information.

When the intensity of each edge is quantized to the Q level, the number of patterns of the edge structure is expressed by the following equation (3): 2Q (2Q + 1) ² (3) 18 when Q = 1, 100 when Q = 2,
There are 294 patterns for Q = 3. The edge strength is obtained by quantizing the absolute value of the luminance level difference between the pixels adjacent to each other across the zero crossing point of the edge (see the wavelet transform components shown in FIGS. 5A and 5B). For example, if the quantization step is 1
If it is set to 0, the intensity of the edge when the absolute value of the difference is 30 is 30/10 = 3.

As shown in FIGS. 5A and 5B, the zero crossing point can be detected as a position where the wavelet transform component takes a value across zero. Whether the adjacent structure is flat or an edge is determined based on whether the distance to the zero crossing point of the adjacent edge is larger or smaller than a set threshold. Here, the threshold varies depending on the type of the image and how the edge structure is to be deformed, but is about 5 pixels. Classification cannot be performed at the left and right edges of the screen, but it is assumed that there is a flat structure virtually outside the left and right edges.

FIGS. 7A, 7B, and 7C show how to create a codebook (class information table) using learning images. Here, the above-described classification processing is performed on the low-resolution image and the high-resolution image, respectively, and as shown in FIGS. 7A, 7B, and 7C, the edge width ( x1 to x2), the class change between the low-resolution image and the high-resolution image, and the j = 2 component of the low-resolution image (j = 2 in FIG. 7A)
FIG. 7 shows a high-resolution image component at j = 1 (shown as a bold line as the j = 1 component in FIG. 7B).
Register in the code book shown in (c).

In general, a high-resolution image can have a higher edge strength. For example, even if the above-mentioned Q = 3 and 294 types of edge structures in a high-resolution image exist,
The corresponding low resolution image has fewer edge structures. Therefore, in the registered code book, there can be a plurality of classes corresponding to the same low-resolution image edge structure and different high-resolution image edge structures.

Here, the edge width (x1 to x2) is defined as follows. The component of j = 2 takes a peak value once as it goes away from the zero-crossing point, but then approaches zero. The edge width (x1 to x2) is defined based on the j = 2 component on the low-resolution image side as a section in which the component approaching 0 is equal to or greater than a predetermined threshold.

The high-frequency component addition based on the codebook created in this way is performed as follows. That is,
The low-resolution input image signal whose high-frequency component is degraded by passing through a transmission path or the like (or the high-frequency component is insufficient due to enlargement processing) is subjected to wavelet transform, and detected as in the case of the high-resolution input image. Classification of each edge structure is performed. Then, in the code book, j = 2 components (low-resolution components) were compared among the codes in which the low-resolution classes were the same and the high-resolution classes were in the range given in the rule table described later. The code with the smallest error is selected, and the component of the high-resolution code (j = 1) at that time is added to the corresponding edge.

Here, the rule table will be described. When a low-resolution image is given, based on the image type, how to change the edge in the image is defined and defined in advance within the range of the edge class correspondence described above. deep. A table defining the limited range of the correspondence is called a rule table. There are cases where the low-resolution class is the same, but the high-resolution class is different, when there are originally gentle edges from the high-resolution class, and when the high-resolution steep edge structure is This is because in some cases it has become moderate. The rule table is provided to limit cases that can be taken out of the plurality of cases.

When the above low resolution image is given,
Based on the image type, how to change the edge in the image type also considers how the band is limited, but basically, the edge structure that humans feel visually sharp according to the image type A rule table is created in advance so as to transform the rule into a rule table. For example, since a character image has a step-like waveform structure, it is appropriate to change to a step-like steep structure, and if it is a natural image, there should not be such a steep structure. Set rules so that intensity does not saturate. Here, the determination as to whether the image is a character image or a natural image can be automated to some extent by examining the statistical properties of the target image, but basically, a human judges the image to be processed.

In the above, when the code including the high-frequency component signal to be added is selected, the error evaluation based on the j = 2 components shows that the value of the following equation (4) is minimum over the edge width (x1 to x2). Look for

(Equation 3) Here, W _s2 (x) is j = 2 components in the codebook, W
_i2 (x) is the j = 2 component of the edge of the input image.

When there are a plurality of minimum values, the arithmetic average of these j = 1 components is calculated and used. That is, _assuming that there are N minimums, the j = 1 components in the codebook for the N are expressed as W _{s1, n} (x), and W _N (x) calculated as the average of these is: The equation (5) is as follows.

(Equation 4) The arithmetic mean of the j = 1 component is added as a high-resolution component signal.

FIG. 8 shows a first embodiment of the contour emphasis processing circuit of the present invention. By referring to the figure, it will be understood that the processing circuit of the present invention operates as follows. That is, at the time of learning, a high-resolution image as a learning image and a low-resolution image created by passing the high-resolution image through a low-pass filter are respectively subjected to wavelet transform to obtain a high-resolution scale (j = 1) component and a low-resolution scale at an edge portion. (J = 2) Find the correspondence between components. At the same time, the waveform structure of the edge portion is analyzed to perform a classification process. The correspondence between the wavelet transform component and the class is stored as a codebook. On the other hand, the class correspondence between the high-resolution image and the low-resolution image according to the image type is defined as a rule table.

At the time of processing, after the input image is subjected to wavelet transform, classification processing is performed. Then, for each edge, select a high-resolution scale component to be added from the classes associated in the rule table,
The signal processing of adding (high-frequency addition) to the edge of the input image is performed.

The adaptive emphasis shown in FIG. 8 means that the emphasis coefficient is changed according to the surrounding structure for each detected edge, as described in the aforementioned Ito and Yagi et al. Or changing the processing method. After this adaptive emphasizing process, an inverse wavelet transform is performed to reconstruct an image whose edges are adaptively emphasized in the high frequency range.

FIG. 9 shows a second embodiment of the contour emphasis processing circuit of the present invention. This embodiment is a case where the present invention is applied to the enlargement processing. The so-called basic parts other than the part related to the enlargement processing indicated by the dashed line are the same as those of the first embodiment.

In the present embodiment, first, at the time of learning, the high-resolution image is band-limited (this is included in the basic part of the present invention), and the enlarged size is matched with the original high-resolution image. For this reason, a low-resolution image that has been once reduced (by sampling point thinning or reduction by filter processing) at the reciprocal of the magnification is treated as a low-resolution image, and this is interpolated and expanded. The low-resolution image and the high-resolution image created in this way are each subjected to wavelet transform, and the correspondence between the high-resolution scale (j = 1) component and the low-resolution scale (j = 2) component at the edge is obtained. At the same time, the waveform structure of the edge portion is analyzed to perform a classification process. The correspondence between the wavelet transform component and the class is stored as a codebook. On the other hand, the correspondence between the high-resolution image and the low-resolution image according to the image type is defined as a rule table. In the present embodiment, when the enlargement ratios are different, it is necessary to create a codebook according to the enlargement ratio.

At the time of processing, the input image is subjected to enlargement processing using a filter having the same characteristics as the interpolation filter at the time of learning, and then the same processing as in the first embodiment is applied. That is, after the enlarging process, a wavelet transform is performed to perform a classifying process, and for each edge, a high-resolution scale component to be added is selected from the classes associated in the rule table, and the edge of the input image is selected. (High-frequency addition). The other signal processing is the same as in the first embodiment, and a description thereof will not be repeated.

[0050]

According to the present invention, a high-frequency component obtained in advance using a learning image is added to an edge portion of the image to perform an emphasizing process. Edge restoration is possible. In particular, in an image on which enlargement processing has been performed, it is effective as enhancement processing when high frequency components are insufficient. In addition, since the edge structure is classified, the adaptive emphasis processing according to the surrounding waveform structure described in Ito and Yagi et al. Mentioned above can be performed together, realizing flexible edge emphasis. can do. Furthermore, since the change range of the waveform structure before and after the edge restoration is defined by the rule table, flexible deformation of the edge structure can be handled in a manner of defining the rule.

As described above, the present invention provides: 1. Even if the emphasizing process is performed, interference is unlikely to occur, so that the emphasizing process can be performed effectively. It has features such as being able to increase the resolution of the image by appropriately adding a high-frequency component to an image that has a blurred feeling due to enlargement processing, etc., compared to the image quality correction device using the conventional linear processing. A high-quality edge-emphasized image can be obtained.

[Brief description of the drawings]

FIG. 1 shows a state of band division in a discrete wavelet transform.

FIG. 2 shows an example of a hardware configuration for performing a wavelet transform.

FIG. 3 shows an example of a wavelet transform.

FIG. 4 shows a Laplacian basic wavelet.

FIG. 5 shows an example of classifying an edge structure.

FIG. 6 shows a description of deformation of an edge structure using class information.

FIG. 7 shows a procedure for creating a codebook using learning images.

FIG. 8 shows a first embodiment of the contour emphasis processing circuit of the present invention.

FIG. 9 shows a second embodiment of the contour emphasis processing circuit of the present invention.

FIG. 10 shows an operation and a circuit configuration of a conventional edge enhancement process.

 ──────────────────────────────────────────────────続 き Continuing on the front page (72) Inventor Masahide Naemura 1-10-11 Kinuta, Setagaya-ku, Tokyo Japan Broadcasting Corporation Research Institute (72) Inventor Hitoshi Ogawa 1-110-11 Kinuta, Setagaya-ku, Tokyo No. Japan Broadcasting Corporation Broadcasting Research Institute (72) Inventor Hajinobu Mizutani 1-10-11 Kinuta, Setagaya-ku, Tokyo Japan Broadcasting Corporation Broadcasting Research Institute (72) Inventor Jun Fukuda 1, Kinuta, Setagaya-ku, Tokyo No. 10-11 Japan Broadcasting Corporation Broadcasting Research Institute F term (reference) 5C021 PA33 PA34 PA35 PA38 PA74 PA75 PA80 RA02 RB05 XB03

Claims (4)

[Claims] 1. A wavelet transform circuit for performing a wavelet transform on an input image signal, class classifying means for classifying an edge structure from an edge signal extracted from a signal subjected to wavelet transform by the wavelet transform circuit, A memory in which a mutual correspondence between each of the divided classes and a high-frequency component signal to be added to an edge portion of the wavelet-transformed image signal for edge enhancement is stored in advance as a class information table; A high-frequency component signal adding circuit for adding the high-frequency component signal to be added according to the class to which the edge portion of the image signal subjected to the wavelet transformation belongs, by referring to the class information table read from Performs an inverse wavelet transform on the output signal of the signal addition circuit A contour emphasis processing circuit comprising at least an inverse wavelet transform circuit. 2. The contour enhancement processing circuit according to claim 1, wherein the class information table includes a wavelet for each of a low-resolution image created by band-limiting a high-resolution image for learning and a high-resolution image for learning. Extract the edge signal from the signal obtained by the conversion,
The extracted edge signals are subjected to edge structure classification, the classified edge structures of the corresponding edges of the low-resolution image and the high-resolution image obtained by the classification, and the low-frequency component and the high-frequency component, respectively. A contour emphasis processing circuit characterized in that a wavelet transform component in an edge portion composed of components is tabulated. 3. The contour emphasis processing circuit according to claim 1, wherein the high-frequency component signal added by the high-frequency component signal addition circuit is selected from the class information table based on a type of the input image signal. An edge enhancement processing circuit characterized in that the signal is a processed high-frequency component signal. 4. The outline emphasis processing circuit according to claim 1, wherein said outline emphasis processing circuit further includes an intermediate circuit between said high frequency component signal adding circuit and said inverse wavelet transform circuit. An edge emphasis processing circuit comprising adaptive emphasis processing means for adaptively switching the edge emphasis processing according to the edge structure of the input image signal.