KR101412369B1 - Method and apparatus for decting object in image - Google Patents

Abstract

The present invention relates to a method and an apparatus for detecting an object from an image regardless of the rotational change thereof. A method for detecting an object from an image according to an embodiment of the present invention comprises the steps of: maintaining a database in which a first characteristic value of a first object image and a second characteristic value of a second object image extracted by using a histogram extracted from a sinogram of an image are divided and stored; receiving an input image; performing the Radon transform of the input image, and extracting a characteristic value of the input image by extracting a histogram from the Radon transform result; and detecting a specific object from the input image by comparing the characteristic value of the input image with the first and second characteristic values stored in the database.

Description

TECHNICAL FIELD The present invention relates to a method and an apparatus for detecting an object of an image,

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method and apparatus for detecting an object of an image, and more particularly, to a method and apparatus for detecting an object of an image regardless of a change in rotation.

Among the object detection methods of images, the face detection algorithm is based on a knowledge-based method that detects and creates human knowledge of faces by using rules, a feature-based method that detects faces by finding unchanging features of faces, A method of detecting compared with an image, an appearance-based method of extracting features from a training image set and then training the detector to detect it.

In the conventional appearance-based method training algorithm, when a face rotated in various directions is input, a detector for each direction must be separately trained. In the detection process, a plurality of detectors are constituted to constitute a plurality of detectors, A problem arises.

Korean Patent No. 10-0729268 discloses a method of detecting an inclined face in real time using an ADOBOSTING algorithm with subwindow rotation.

However, the above-mentioned Patent No. 10-0729268 does not disclose a method of detecting a face irrespective of the rotation of the image.

Therefore, it is necessary to study a technique for precisely detecting a specific object regardless of the rotation of the image.

SUMMARY OF THE INVENTION It is an object of the present invention to provide a method and apparatus for detecting an object of an image that can detect a specific object irrespective of the rotation of the image.

According to an aspect of the present invention, there is provided a histogram analyzing method for analyzing a histogram of a first feature value of a first object image extracted using a histogram extracted from a sinogram of an image, Maintaining the separately stored database; Receiving an input image; Extracting a feature value of the input image by radon transforming the input image, extracting a histogram from the result of the radon transformation, And comparing the feature value of the input image with the first feature value and the second feature value stored in the database to detect a specific object in the input image.

According to an aspect of the present invention, there is provided a histogram analyzing method for analyzing a histogram of a first feature value of a first object image extracted using a histogram extracted from a sinogram of an image, This separately stored database; A receiving unit for receiving an input image; A feature value extracting unit for radon-transforming the input image, extracting a histogram from the result of the radon transformation, and extracting a feature value of the input image; A detector for comparing a feature value of the input image with the first feature value and the second feature value stored in the database and detecting a specific object in the input image; And a control unit for controlling the database, the receiving unit, the feature value extracting unit, and the detecting unit.

A method and apparatus for detecting an image object according to an embodiment of the present invention can detect a specific object in an input image irrespective of rotation of the image.

According to an embodiment of the present invention, since features of an image are extracted using a radon conversion result of an image, there is an advantage that training images are not provided for each rotation angle.

1 is a block diagram of an apparatus for detecting an object of an image according to an embodiment of the present invention.
2 is a flowchart illustrating an object detection method of an image according to an embodiment of the present invention.
3 is a diagram illustrating a radon conversion process of an image according to an embodiment of the present invention.
4 is a diagram showing a result of radon transformation of a rotated image according to an embodiment of the present invention.
5 is a diagram illustrating a reverse projection method of radon transformation of a rotated image according to an embodiment of the present invention.
FIG. 6 is a diagram illustrating a method for extracting a histogram in a sinogram that is a result of radon transformation according to an embodiment of the present invention.
FIG. 7 is a diagram for comparing histograms extracted from a sinogram associated with an embodiment of the present invention. FIG.
8 is a diagram illustrating an example of an image collected in a database to perform an object detection method according to an embodiment of the present invention.
9 is a diagram illustrating a performance evaluation of an object detection method of an image according to an embodiment of the present invention.

Hereinafter, a method and an apparatus for detecting an object of an image according to an embodiment of the present invention will be described with reference to the drawings.

As used herein, the singular forms "a", "an" and "the" include plural referents unless the context clearly dictates otherwise. In this specification, the terms "comprising ", or" comprising "and the like should not be construed as necessarily including the various elements or steps described in the specification, Or may be further comprised of additional components or steps. Also, the terms "module "," module "and the like described in the specification mean units for processing at least one function or operation, which may be implemented by hardware or software or by a combination of hardware and software.

1 is a block diagram of an apparatus for detecting an object of an image according to an embodiment of the present invention.

The object detecting apparatus 100 may include a database 110, a receiving unit 120, a feature value extracting unit 130, a detecting unit 140, and a controller 150. [

The database 110 may store various images and feature values for the images. The database 110 may store characteristic values of an image extracted through a specific training algorithm. In addition, the database 110 may classify feature values grouped by object for specific object detection.

For example, a feature value (referred to as 'first feature value' for the sake of simplicity) and a feature value (referred to as 'second feature value' for convenience) ). ≪ / RTI > The first object image may be a human face image, and the second object image may be a background image.

The feature values stored in the database 110 may be extracted using a radon histogram of the training image. A radon histogram is a histogram extracted from a sinogram that is the result of radon transformation.

The first feature value and the second feature value may be distinguished using a trained detector. For example, a plurality of (e.g., 20,000) face images and background images collected via the MIT-CMU DB or the Internet may be stored in a training database (not shown). A binary detector (not shown) is constructed by extracting the features of the face image and the background image stored in the training database, and a criterion for distinguishing the characteristics of the face image and the features of the background image is provided through training of the binary detector can do. That is, the first feature value and the second feature value may be values separated by the reference of the trained binary detector.

Methods for extracting histograms from radon transforms, sinograms, and sinograms will be described later.

The receiving unit 120 may receive an input image. The input image may include a plurality of objects.

The feature value extracting unit 130 may extract the feature value of the input image using the radon conversion result of the input image. In this case, the feature value extracting unit 130 can divide the input image into a plurality of regions, and extract feature values for each divided region. The feature value extracting unit 130 may perform radon transformation for each divided region and extract the histogram from the result of the radon transformation. The feature value extracting unit 130 may extract a feature value of the input image for each divided region using the histogram.

The detection unit 140 may detect a specific object in the input image by comparing the feature value of the extracted input image with the feature value of the training image stored in the database 110. The specific object may be a face region of a person included in the input image.

The controller 150 may control the database 110, the receiver 120, the feature value extractor 130, and the detector 140 as a whole.

2 is a flowchart illustrating an object detection method of an image according to an embodiment of the present invention. Hereinafter, the case where the object to be extracted from the input image is a human face will be described. However, the present invention is not limited to the detection of the face.

The database 110 may store the first feature value for the first object image and the second feature value for the second object image (S210). Hereinafter, an example will be described in which the first object image is a human face image and the second object image is a background image.

The database 110 may store a first feature value for a plurality of face images (e.g., thousands) and a second feature value for a plurality of background images (e.g., thousands). The first feature value and the second feature value may be values extracted using a radon histogram.

To understand the present invention, the radon conversion will be described first.

The radon conversion is used in a process of analyzing a CT (computational tomography) image, and can be expressed by a mathematical expression that integrates physical properties of a transmitted x-ray object. The result of the radon transformation appears as a sinogram, and a tomogram can be obtained by inverse transformation of the sinogram. Before describing the features applying the radon inversion process, we define the radon conversion process and the inverse conversion process and describe the characteristics of the conversion process.

The radon transformation is a process in which the image f (x, y) of the rectangular coordinate system is linearly divided with respect to the projection angle θ and accumulated in the radon space. This can be expressed by the following equation (1).

Where P (ρ, θ) is the result of line segmentation of the image f (x, y), ρ is the length of the normal line through the origin and θ is the angle between the normal and the x axis. And [delta] denotes the Dirac delta function. Since the image f (x, y) at the angle? Is expressed as follows in the rotated coordinate system, the image coordinates x and y can be expressed by the following equation (2).

Finally, the radon transformation can be expressed as shown in Equation (3).

3 is a diagram illustrating a radon conversion process of an image according to an embodiment of the present invention.

As shown, the radon transformation means that the image f (x, y) is linearly distributed in the two-dimensional rectangular coordinate system and is accumulated in the radon transform space (sinogram). An image can be represented as a two-dimensional function expressed by two coordinate values x and y as two-dimensional data, and a linear function is expressed as a linear function of the image. In the right sinogram, the red line is one-dimensional data of the line segment in the oblique direction of θ ₀ , which is a cumulative value of pixel values existing on each dotted line in the image.

4 is a diagram showing a result of radon transformation of a rotated image according to an embodiment of the present invention.

The resultant period of the radon transformation has a period of 2?, And when the image is rotated by? ₀ , it has a property of translational movement. For example, if a sinogram is generated by linearly interpolating the frontal image from 0 to 180 degrees at an interval of 1 degree, when the sinogram is generated by linearly interpolating at 30 degrees to 1 degree intervals from 30 degrees to 30 degrees, It has the same sinogram as the image. That is, the parallel movement of the sinogram by the image rotation is performed in the horizontal direction.

The inverse transformation process of the radon transformation can be inversely transformed by inverse Fourier transform after obtaining the two-dimensional Fourier transform using the central slice theorem, and can be expressed as shown in Equation (4).

However, since there is a coordinate transformation process and there are many errors, a back-projection method can be used. This can be expressed by the following equation (5).

5 is a diagram illustrating a reverse projection method of radon transformation of a rotated image according to an embodiment of the present invention.

The information of the radon domain that reconstructs one pixel of an image using a back projection method can be shown as an oblique line in the sinogram. The value obtained by accumulating the data of the red dotted line in the sinogram of FIG. 5 corresponds to the red pixel value of the image reconstructed through the inverse transparent method, and the dotted green line parallel to the y axis corresponds to the center pixel value of the reconstructed image .

As described above, in order to extract characteristic values robust to rotation, a sinogram that is a result of radon transformation is used in the example of the present invention.

Hereinafter, a method of extracting feature values of an input image using a sinogram will be described.

The receiving unit 120 may receive the input image (S220). The input image may include a plurality of objects. For example, the input image may include a human face and a background.

The feature value extracting unit 130 may divide the image into a plurality of regions for extracting feature values of the received input image and extract feature values using a sinogram that is a result of radon transformation for each divided region (S230, S240). The feature value extracting unit 130 extracts a radon histogram from the sinogram and extracts feature values of the input image using the extracted radon histogram.

FIG. 6 is a diagram illustrating a method for extracting a histogram in a sinogram that is a result of radon transformation according to an embodiment of the present invention.

The rotation of θ in the image appears as a parallel movement in the θ-axis direction in the radon space. The histogram ψρ ₀ of the projection data for each direction corresponding to an arbitrary ρ ₀ value has the same value irrespective of the rotation.

The histogram of each projection data represents the information of the outer region of the circle centered at the image. Therefore, the difference of each histogram preserves ring-shaped pixel information in the image area. More specifically, it is as follows.

Considering the values of the cutoffs for each angle, a data set on the horizontal axis of the sinogram (right) appears as the sum of the dotted lines in the image (left figure). In this case, considering the coordinates of the data of the line segments for each angle, the data on the sinograms constituting the histogram can be seen to contain the information of the pixels located in the outer circle of the red circle in the image. Therefore, the histogram difference generated by using a data set on the horizontal axis of the sineogram (right figure) contains the data of the ring-shaped section as follows.

That is, according to an embodiment of the present invention, the feature value of the input image can be extracted using at least two histograms extracted from the sinogram. In this embodiment, an example of extracting a feature value using the difference of two extracted histograms is shown.

If a radon conversion is performed by rotating an image of M x M size by 180 degrees in units of 1 degree, a monogram of M√2 x 180 size can be generated. That is, if the image size is 24 x 24, a total of 35 histograms can be generated. If a 24x24 image is linearized, a 24x1 vector is generated. In the process of generating a sonogram, the angles are calculated by angles. Therefore, the maximum size of a vector having a stroke of an image rotated 45 degrees, such as 45 degrees and 135 degrees, is 24√2x1, ie, a vector of 35x1 is calculated Thus, when a sonogram is generated by accumulating the histogram, 35 histograms can be generated.

FIG. 7 is a diagram for comparing histograms extracted from a sinogram associated with an embodiment of the present invention. FIG.

7 (a), the histogram is extracted for each image. As a result, there is no change in the histogram set according to the rotation. In addition, FIG. 7B shows that a histogram of a face image and a background image can be compared with each other to distinguish a face and a background. That is, we can see that the same histogram is displayed even if the facial image is rotated through FIG. 7, and it can be seen that the facial image and the background image can be distinguished through the histogram.

Finally, two histograms extracted from the radon space can be selected and the feature values for each bin (rank) can be defined as shown in Equation (6).

F is the feature value, ψ _r denotes the rth histogram, b is the bin of the histogram, B is the number of bin (rank) of the histogram, ε is an arbitrary value to prevent the denominator of the equation from becoming zero It is a small value.

Where B is the number of bin in each histogram. Using Equation (6), feature values can be extracted for all possible cases in a radon histogram set. If we calculate the histogram of M √ 2 bin B by Radon transformation process on M × M size images, we can extract total ( _{Mv 2} C ₂ × _{B 2} ) feature values.

Hereinafter, the performance of a method of detecting a face image using an image object detecting apparatus according to an embodiment of the present invention will be described with reference to experimental examples.

8 is a diagram illustrating an example of an image collected in a database to perform an object detection method according to an embodiment of the present invention.

In order to measure the performance of the face detection algorithm, 5000 front face / background images collected through the MIT-CMU DB and the Internet were rotated at an arbitrary angle. Detectors used in the experiment were trained using 20,000 frontal facial images at each step and 20,000 non - facial images at arbitrary positions in background images not including facial images.

Table 1 shows the detection rate (DR) and false detection rate (FP) for each rotation angle.

Rotation angle Test DB set Test DB set DR (%) FP (%) DR (%) FP (%) 0 degrees 95.2 9.0 90.3 12.9 30 degrees 95.2 9.0 90.3 12.9 45 degrees 95.2 9.0 90.3 12.9 90 degrees 95.2 9.0 90.3 12.9 135 degrees 95.2 9.0 90.3 12.9 180 degrees 95.2 9.0 90.3 12.9

FIG. 9 is a diagram illustrating performance evaluation of face detection of a rotated image using the results of Table 1. FIG.

Experimental results show that the detection rate is more than 95% and less than 10% for the training image, and since the performance is not changed according to the rotation angle, the detector has the same performance irrespective of the rotation conversion.

As described above, the method and apparatus for detecting an object of an image related to an embodiment of the present invention extracts characteristics of an image using the radon conversion result of the image, can do.

The above-described method of detecting an object of an image may be implemented in the form of a program command that can be executed through various computer means and recorded in a computer-readable recording medium. At this time, the computer-readable recording medium may include program commands, data files, data structures, and the like, alone or in combination. On the other hand, the program instructions recorded on the recording medium may be those specially designed and configured for the present invention or may be available to those skilled in the art of computer software.

The computer-readable recording medium includes a magnetic recording medium such as a magnetic medium such as a hard disk, a floppy disk and a magnetic tape, an optical medium such as a CD-ROM and a DVD, a magnetic disk such as a floppy disk, A magneto-optical media, and a hardware device specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like.

The recording medium may be a transmission medium, such as a light or metal line, a wave guide, or the like, including a carrier wave for transmitting a signal designating a program command, a data structure, and the like.

The program instructions also include machine language code, such as those generated by the compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware devices described above may be configured to operate as one or more software modules to perform the operations of the present invention, and vice versa.

The method and apparatus for detecting an object of an image as described above can be applied to a case where the configuration and method of the embodiments described above are not limitedly applied. As shown in FIG.

100: object detection device of an image
110: Database
120: Receiver
130: Feature value extraction unit
140:
150:

Claims (10)

Maintaining a database in which a first feature value of the first object image extracted using the histogram extracted from the sinogram of the image and a second feature value of the second object image are distinguished from each other;
Receiving an input image;
Extracting a feature value of the input image by radon transforming the input image, extracting a histogram from the result of the radon transformation, And
And comparing the feature value of the input image with the first feature value and the second feature value stored in the database to detect a specific object in the input image. The method of claim 1, wherein the step of extracting feature values of the input image
Dividing the received input image into a plurality of regions; And
And extracting a histogram from the result of the radon transformation and extracting a feature value of the input image for each of the plurality of regions by performing a radon transformation for each of the plurality of divided regions, . The method according to claim 1,
Wherein the first object image is a human face image,
The second object image is a background image,
Wherein the specific object to be detected is a human face. The method of claim 1, wherein the step of extracting feature values of the input image comprises:
And extracting feature values of the input image using at least two histograms extracted from the result of the radon transformation. The method of claim 1, wherein the step of extracting feature values of the input image comprises:
And extracting characteristic values of the input image using two histograms extracted from the result of the radon transformation. A database storing a first feature value of a first object image extracted using a histogram extracted from a sinogram of the image and a second feature value of the second object image;
A receiving unit for receiving an input image;
A feature value extracting unit for radon-transforming the input image, extracting a histogram from the result of the radon transformation, and extracting a feature value of the input image;
A detector for comparing a feature value of the input image with the first feature value and the second feature value stored in the database and detecting a specific object in the input image; And
And a control unit for controlling the database, the receiving unit, the feature value extracting unit, and the detecting unit. 7. The apparatus of claim 6, wherein the feature value extractor
Extracting a histogram from the result of the radon transformation, and extracting a feature value of the input image for each of the plurality of regions by performing a radon transformation for each of the plurality of regions, The object detecting apparatus comprising: The method according to claim 6,
Wherein the first object image is a human face image,
The second object image is a background image,
Wherein the specific object to be detected is a human face. 7. The apparatus of claim 6, wherein the feature value extractor
And extracts feature values of the input image using at least two histograms extracted from the result of the radon transformation. 7. The apparatus of claim 6, wherein the feature value extractor
And extracts feature values of the input image using two histograms extracted from the result of the radon transformation.

Images