KR20170059232A

KR20170059232A - Method and device for extracting feature of image data

Info

Publication number: KR20170059232A
Application number: KR1020150163320A
Authority: KR
Inventors: 백준걸; 박승환; 홍재열
Original assignee: 고려대학교 산학협력단
Priority date: 2015-11-20
Filing date: 2015-11-20
Publication date: 2017-05-30
Also published as: KR101751871B1

Abstract

Disclosed are an image data feature extracting apparatus, which comprises a filter generation unit, a feature point generation unit, a feature point utilization unit, a storage unit, and a control unit, and an image data feature extracting method using the same. The image data feature extracting method comprises: a step of generating a modification box filter for generating a modification box filter in a linear combination form of n (n is an integer of two or more) box filters using linear regression; a step of applying the modification box filter to feature extracting algorithm; and a step of generating at least one feature point in the image data using the feature extracting algorithm. The feature point does not change with respect to the rotation, movement, or scaling of the image data. The image data feature extracting apparatus and method thereof according to the embodiment of the present invention can extract the feature of the image data more accurately than applying the conventional Gaussian filter by applying the modification box filter to the image data feature extracting algorithm.

Description

TECHNICAL FIELD [0001] The present invention relates to an image data extracting method,

The present invention relates to a method and apparatus for extracting features from image data, and more particularly to a method and apparatus for extracting features of image data that extract feature points accurately in image data using a correction box filter .

This patent proposes a feature extraction method that can more accurately express the characteristics of image data.

There are SIFT (Scale Invariant Feature Transform) and SURF (Speeded Up Robust Features) as a characteristic extraction technique of representative image data. After extracting characteristic parts of image data using SIFT and SURF, it can be used for image matching and object recognition in the field of computer vision, and can be used for image anomaly detection and classification in manufacturing field.

The existing SIFT and SURF use a Gaussian filter, which applies the Gaussian distribution, which is the most commonly used distribution in all scientific fields, to image processing. The Gaussian distribution used in the image processing is symmetric and has one protruding part. Since the detailed information about the image is erased and only the information about the entire contour is left, the optimum performance is not shown for the feature extraction of the image data . Particularly, there is a problem that the performance deteriorates in distinguishing the high frequency due to the feature of reducing or eliminating the high frequency part while leaving the low frequency part.

Therefore, it is necessary to develop new filters and extract feature points in order to extract discriminative features and increase the predictive power of the results.

SUMMARY OF THE INVENTION The present invention has been made in view of the above problems, and it is an object of the present invention to provide a method and apparatus for extracting features of image data with high discrimination by applying a correction box filter to an image data feature extraction algorithm.

The image data attribute extracting apparatus according to an embodiment of the present invention includes a filter generating unit for generating a correction box filter of a linear combination type of n (n is an integer of 2 or more) box filters using linear regression, A feature point generation unit for generating at least one feature point in the image data by applying the correction box filter to a feature extraction algorithm, a storage unit for storing the correction box filter and the feature points, And a control unit for controlling the operation of the storage unit.

The method of extracting feature of image data according to an embodiment of the present invention includes generating a modified box filter of linear combination type of n (n is an integer of 2 or more) box filters using linear regression Applying the modified box filter to a feature extraction algorithm and generating feature points for generating at least one feature point in the image data using the feature extraction algorithm, And does not change with respect to rotation, movement, or scaling.

The image data feature extracting method and apparatus according to the embodiment of the present invention can accurately extract the characteristics of the image data by applying the correction box filter to the image data feature extracting algorithm rather than applying the conventional Gaussian filter.

In addition, the method and apparatus for extracting feature of image data according to the embodiment of the present invention can extract feature points of high discrimination power by applying correction box filter to SIFT and SURF algorithm, and furthermore, , Or the performance of image classification can be improved.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS In order to more fully understand the drawings recited in the detailed description of the present invention, a detailed description of each drawing is provided.
1 is a functional block diagram of an image data attribute extracting apparatus according to an embodiment of the present invention.
2 is a diagram showing a filter for extracting features of image data.
3 is a diagram illustrating an image that is filtered using an input image source and a conventional filter.
FIG. 4 is a diagram showing a result image obtained by extracting features. FIG.
5 is a flowchart illustrating an image data attribute extraction method according to an embodiment of the present invention.

It is to be understood that the specific structural or functional description of embodiments of the present invention disclosed herein is for illustrative purposes only and is not intended to limit the scope of the inventive concept But may be embodied in many different forms and is not limited to the embodiments set forth herein.

The embodiments according to the concept of the present invention can make various changes and can take various forms, so that the embodiments are illustrated in the drawings and described in detail herein. It should be understood, however, that it is not intended to limit the embodiments according to the concepts of the present invention to the particular forms disclosed, but includes all modifications, equivalents, or alternatives falling within the spirit and scope of the invention.

The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In this specification, the terms "comprises" or "having" and the like are used to specify that there are features, numbers, steps, operations, elements, parts or combinations thereof described herein, But do not preclude the presence or addition of one or more other features, integers, steps, operations, components, parts, or combinations thereof.

Unless otherwise defined, all terms used herein, including technical or scientific terms, have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. Terms such as those defined in commonly used dictionaries are to be interpreted as having a meaning consistent with the meaning of the context in the relevant art and, unless explicitly defined herein, are to be interpreted as ideal or overly formal Do not.

Hereinafter, embodiments of the present invention will be described in detail with reference to the drawings attached hereto.

First, with reference to FIG. 1 to FIG. 4, an apparatus for extracting characteristics of image data according to an embodiment of the present invention will be described in detail.

1 is a functional block diagram of an image data attribute extracting apparatus 100 according to an embodiment of the present invention. FIG. 2 is a view showing a filter for extracting features of image data, FIG. 3 is a view illustrating an image filtered using an input image source and a conventional filter, FIG. Fig.

1, the image data feature extraction apparatus 100 includes a filter generation unit 110, a feature point generation unit 120, a feature point utilization unit 130, a storage unit 180, and a control unit 190 .

As used herein, the term "minus" may mean a functional and structural combination of hardware for carrying out the technical idea of the present invention and software for driving the hardware. For example, the '-section' may mean a logical unit of a predetermined code and a hardware resource to be executed by the predetermined code, and does not necessarily mean a physically connected code or a kind of hardware.

The filter generating unit 110 of the image data attribute extracting apparatus 100 generates a modified box filter for extracting the attribute of the image data under the control of the control unit 190. More specifically, the filter generation unit 110 generates n (n is an integer of 2 or more) box filters (see FIG. 2 (a)) through a linear regression, (See Fig. 2 (c)).

For example, the correction box filter can be generated using the following equation (1) using linear regression.

Is the transform of each box filter into a column vector and is used as an independent variable of the regression equation. y is used as a dependent variable as a Gaussian filter and finds a value that reduces the error between the linear combination of the Gaussian filter and the box filter.

The filter generation unit 110 according to an exemplary embodiment of the present invention generates a correction box filter that can enhance image feature extraction performance as compared with a conventional Gaussian filter or box filter. Referring to FIG. 3C, the box filter for calculating a result by adding the values of neighboring pixels and then averaging is characterized by leaving a fine portion of the image as compared with the Gaussian filter. However, the performance of a feature point using a box filter alone is lower than that of a feature point using a Gaussian filter. Therefore, in the present invention, a correction box filter (see Fig. 2 (c)) through a linear combination of box filters is proposed. We use a Gaussian filter as a dependent variable, a box filter of different size as an independent variable, and use a linear regression to approximate the Gaussian distribution to produce a filter having the properties of a box filter. The correction box filter has the advantage of extracting the characteristics of the image more sensitively than the Gaussian filter when extracting the characteristics of the image data.

The feature point generation unit 120 of the image data feature extraction apparatus 100 extracts the characteristics of the image data under the control of the control unit 190. At this time, a feature extracting algorithm that extracts the same region at the same position even if a change in size or rotation occurs in the image is used. For example, a scale invariant feature transforms (SIFT) algorithm can be used as an image processing technique for extracting feature points. The SIFT is an image processing technique for extracting prominent feature descriptions that do not change with respect to rotation, movement, and scaling. The SIFT algorithm is well known to those of ordinary skill in the art to which the present invention belongs. It is omitted.

In addition, various image data feature extracting techniques such as SURF (Speed Up Robust Feature) that can extract the same feature at the same position can be used even if a change in size or rotation occurs in the image.

4 is a result image obtained by extracting feature points of an original image using a Gaussian filter or a modified box filter in which a threshold and an octave are differently set to the SIFT algorithm. Referring to FIG. 4, compared with the case where the conventional Gaussian filter is applied to the SIFT algorithm to extract the feature points of the image data (see FIGS. 4A and 4B), the correction box shown in FIG. When the filter is applied to the SIFT algorithm to extract the feature points of the image data (see (c) of FIG. 4), the discrimination power is improved. Particularly, the edge of the image data can be recognized more accurately, and the feature points of the high frequency region can be extracted more than the Gaussian filter. In other words, by applying the correction box filter generated by the filter generation unit 110 instead of the Gaussian filter to the feature extraction algorithm of the image data, it is possible to extract more distinguishable feature points than when extracting the feature points of the image data by applying the Gaussian filter Do.

The feature point utilization unit 130 of the image data feature extraction apparatus 100 utilizes the feature points generated by the feature point generation unit 120 in various fields such as object recognition, image registration, and image classification under the control of the control unit 190 . The feature point utilization unit 130 may be provided in a separate apparatus other than the image data feature extraction apparatus 100, unlike the present embodiment.

The correction box filter generated by the filter generation unit 110 is used for the feature point extraction to more accurately recognize the object or the like. More specifically, the license plate recognition can be performed more accurately, and a printed circuit board Can be more accurately extracted as feature points and used for abnormal / normal discrimination and classification.

The storage unit 180 may include a program storage unit and a data storage unit. Programs for controlling the operation of the image data feature extraction apparatus 100 may be stored in the program storage unit. The data storage unit may store data generated during the execution of the programs.

The control unit 190 controls the overall operation of the image data attribute extracting apparatus 100. That is, the operation of the filter generating unit 110, the feature point generating unit 120, the feature point utilizing unit 130, and the storage unit 180 can be controlled.

Hereinafter, an image data feature extraction method using an image data feature extraction apparatus according to an embodiment of the present invention will be described in detail with reference to FIG.

5 is a flowchart illustrating an image data feature extraction method using the image data feature extraction apparatus shown in FIG.

First, a filter to be used in the feature extraction algorithm of the image data is generated (S100). Specifically, n (n is a natural number of 2 or more) box filters (see Fig. 2 (a)) are subjected to linear regression by using a box filter formed by a linear combination of box filters and a Gaussian filter (see Fig. 2 (b)).

We use a Gaussian filter as a dependent variable, a box filter of different size as an independent variable, and use a linear regression to approximate the Gaussian distribution to produce a filter having the properties of a box filter. The correction box filter has the advantage of extracting the characteristics of the image more sensitively than the Gaussian filter when extracting the characteristics of the image data.

Next, the modification box filter is applied to the SIFT algorithm (S200), and the characteristics of the image data are extracted (S300). That is, feature points that are always detected identically are generated irrespective of changes in scaling, rotation, observation angle, and the like, which can express characteristics of the image in the original image (S300).

In the case of the existing image data attribute extraction algorithm, a Gaussian filter (see Fig. 2 (b)) was used. The Gaussian filter is a filter for reducing or eliminating a low frequency in which a change in an image pixel is large and a low frequency in which a change in an image pixel is small is reduced. When a Gaussian filter is applied to an original image (FIG. b), there is a problem that the feature point extraction performance is poor.

The image data feature extraction method according to an embodiment of the present invention applies a correction box filter using a box filter (see (c) of FIG. 3) capable of finely dividing high frequencies and to apply it to an image data feature extraction algorithm, The characteristic extracting performance in the high frequency region can be improved while maintaining the performance of the filter.

4 (a) or 4 (b), in which image features are extracted using a conventional Gaussian filter, with reference to FIG. 4, which is a comparison of images obtained by extracting feature points, and FIG. 4 (c), it can be confirmed that the feature extraction performance and the edge extraction performance in the high frequency region of the image are improved when the correction box filter is used.

Finally, utilizing the generated feature point (S400) may be further included. For example, the feature points extracted by applying the correction box filter to an image feature extracting algorithm can be utilized in various fields such as object recognition, image registration, and image classification.

While the present invention has been particularly shown and described with reference to exemplary embodiments thereof, it is evident that many alternatives, modifications and variations will be apparent to those skilled in the art. Accordingly, the true scope of the present invention should be determined by the technical idea of the appended claims.

100: image data attribute extracting device
110: filter generation unit 120:
180: storage unit 190: control unit

Claims

A filter generating unit for generating a correction box filter of a linear combination type of n (n is an integer of 2 or more) box filters using linear regression,
A feature point generation unit for generating at least one feature point in the image data by applying the correction box filter to a feature extraction algorithm,
A storage for storing the correction box filter and the feature points, and
And a control unit for controlling operations of the filter generating unit, the minutia point generating unit, and the storage unit,

The method according to claim 1,
Further comprising a feature point utilizing unit that uses the feature point in any one of object recognition, image registration, and image classification.

The method according to claim 1,
Wherein the feature point does not change with respect to rotation, movement, or scaling of the image data.

The method according to claim 1,
Wherein the feature extracting algorithm is any one of a scale invariant feature transform (SIFT) and a speed up feature (SURF).

A method for extracting characteristics of image data in an image data attribute extracting apparatus,
A modified box filter generating step of generating a modified box filter in the form of a linear combination of n (n is an integer of 2 or more) box filters using linear regression;
Applying the modified box filter to a feature extraction algorithm; And
Generating feature points for generating at least one feature point in the image data using the feature extraction algorithm,
Wherein the feature points do not change with respect to rotation, movement, or scaling of the image data.

6. The method of claim 5,
Further comprising a feature point utilization step of utilizing the feature points in any one of object recognition, image matching, and image classification.

6. The method of claim 5,
Wherein the feature extraction algorithm is one of a scale invariant feature transform (SIFT) and a speed up robust feature (SURF).