KR101533925B1 - Method and apparatus for small target detection in IR image - Google Patents

Abstract

Disclosed are a method for detecting a small target in an infrared image, and an apparatus thereof. The method for detecting a small target in an infrared image comprises the steps of: detecting a small target candidate region from an infrared image by using a difference of Gaussian (DoG) filter; generating a saliency map by using residual spectrum of the infrared image; and applying the saliency map to the detected small target candidate region to detect a small target candidate.

Description

TECHNICAL FIELD The present invention relates to a method and an apparatus for detecting a small target in an infrared image,

The present invention relates to a method and an apparatus for detecting a small target in an infrared image.

Outside alert systems and infrared seek / track devices are used primarily to detect and alert threat targets that are approaching from a remote location, such as missiles, fighter jets, helicopters, and others. These threat targets are very small and do not have shape and texture information because they are located remotely from infrared image acquisition equipment at a distance of several kilometers to several tens of kilometers. In addition, infrared images including small targets are characterized by very low signal-to-noise ratio due to atmospheric turbulence, atmospheric radiation, features, bright clouds, and noise.

These small target and infrared image features make it more difficult to detect small targets. Accordingly, various methods have been proposed for suppressing the background in the infrared image and improving the small target to be prominent in order to more effectively detect a small target.

Typical methods include Top-hat transform, mean filter, median filter, max-mean filter and max-median filter. Such conventional methods are sensitive to the distribution of background noise, and thus have a disadvantage in that the performance is significantly lowered when the background is complex.

An object of the present invention is to provide a method and an apparatus for detecting a small target in an infrared image capable of effectively detecting a small target in an infrared image of various background environments.

According to an aspect of the present invention, a method for detecting a small target in an infrared image capable of effectively detecting a small target in an infrared image of various background environments is provided.

According to an embodiment of the present invention, there is provided a method for detecting a small target candidate region using a Difference of Gaussian (DoG) filter in an infrared image; Generating spectral maps using spectral residuals of the infrared image; And detecting a small target candidate by weighting the dark map on the detected small target candidate region. The method of detecting a small target in an infrared image can be provided.

Prior to the step of detecting the small target candidate region, modeling a small target in an infrared image; Generating a mask image using a feature of a small target according to the modeling of the small target; And performing a histogram smoothing on the infrared image and then weighting the mask image to adjust the regional contrast.

The characteristic of the small target is the standard deviation of the maximum size of the small model target.

The mask image may be generated as a difference between the infrared image and a Gaussian blurred image reflecting characteristics of the small target.

The spectral residual is a difference between a log amplitude spectrum of the infrared image and a log-amplitude spectrum of the image blurred by the infrared image.

According to another embodiment of the present invention, there is provided a method comprising: modeling a small target in an infrared image; Generating a mask image using a feature of a small target according to the modeling of the small target; Detecting a small target candidate region in the infrared image using the mask image; Generating spectral maps using spectral residuals of the infrared image; And detecting a small target candidate by weighting the dark map on the detected small target candidate region. The method of detecting a small target in an infrared image can be provided.

According to another aspect of the present invention, a small-sized target detection apparatus is provided in an infrared image capable of effectively detecting a small target in an infrared image of various background environments.

According to an embodiment of the present invention, there is provided a target candidate region detection unit for detecting a small target candidate region using a DoG (Difference of Gaussian) filter in an infrared image; A star map generation unit for generating a star map using the spectral residual of the infrared image; And a target detection unit for detecting a small target candidate by weighting the brightness map to the detected small target candidate region.

By providing a method and apparatus for detecting a small target in an infrared image according to an embodiment of the present invention, it is possible to effectively detect a small target in an infrared image of various background environments.

1 is a flowchart illustrating a method of detecting a small target in an infrared image according to an exemplary embodiment of the present invention.
FIG. 2 is a flowchart illustrating a method of improving a local contrast in consideration of a small target characteristic according to an exemplary embodiment of the present invention.
FIG. 3 is a diagram illustrating a result of an improvement of local contrast in accordance with an embodiment of the present invention. FIG.
4 is a diagram illustrating a result of detecting a small target candidate region using a DoG filter according to an embodiment of the present invention.
FIG. 5 is a diagram illustrating a dark map generated using spectral residuals according to an embodiment of the present invention; FIG.
FIG. 6 is a diagram illustrating results of respective steps in a process of detecting a small target in an infrared image according to an exemplary embodiment of the present invention. FIG.
FIG. 7 is a graph showing a result of comparing small-size target detection results according to an embodiment of the present invention. FIG.
FIG. 8 is a table that quantitatively evaluates performance according to a conventional method and a small-sized target detection according to an embodiment of the present invention. FIG.
9 is a block diagram schematically showing an internal configuration of a target detection apparatus for detecting a small target in an infrared image according to an embodiment of the present invention.

BRIEF DESCRIPTION OF THE DRAWINGS The present invention is capable of various modifications and various embodiments, and specific embodiments are illustrated in the drawings and described in detail in the detailed description. It is to be understood, however, that the invention is not to be limited to the specific embodiments, but includes all modifications, equivalents, and alternatives falling within the spirit and scope of the invention. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the accompanying drawings.

The terms first, second, etc. may be used to describe various components, but the components should not be limited by the terms. The terms are used only for the purpose of distinguishing one component from another.

The terminology used in this application is used only to describe a specific embodiment and is not intended to limit the invention. The singular expressions include plural expressions unless the context clearly dictates otherwise. In the present application, the terms "comprises" or "having" and the like are used to specify that there is a feature, a number, a step, an operation, an element, a component or a combination thereof described in the specification, But do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, or combinations thereof.

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

FIG. 1 is a flowchart illustrating a method of detecting a small target in an infrared image according to an exemplary embodiment of the present invention. FIG. 2 is a flowchart illustrating a method of improving local contrast in consideration of a small target characteristic according to an exemplary embodiment of the present invention FIG. 3 is a diagram illustrating a result of local contrast enhancement according to an exemplary embodiment of the present invention, and FIG. 4 illustrates a result of detecting a small target candidate region using a DoG filter according to an exemplary embodiment of the present invention FIG. 5 is a diagram illustrating a dark map generated using spectral residuals according to an embodiment of the present invention. FIG. 6 is a flowchart illustrating a method of detecting a small target in an infrared image according to an exemplary embodiment of the present invention. FIG. 7 is a diagram illustrating a result of comparing small-size target detection results according to an embodiment of the present invention, and FIG. And the performance according to the small-size target detection according to the embodiment of the present invention is quantitatively evaluated.

In step 110, the target detection apparatus 100 receives an infrared image.

For example, the target detection apparatus 100 can mount an infrared camera or the like, and acquires an infrared image through an image acquisition device such as an infrared camera.

In step 115, the target detection apparatus 100 models a small target in the input infrared image. The small target included in the infrared image captured by the target detection apparatus 100 has a small and round shape since it is photographed at a long distance.

These small targets can be modeled using Gaussian functions. For example, a small target can be modeled using the number one.

는 표적의 중심이(0,0)인 표적의 밝기를 나타내고, x,y는 공간 좌표를 나타내며,

,

는 각각 수평 방향의 표준편차와 수직 방향의 표준편차를 나타낸다.here,

Represents the brightness of the target whose center is (0, 0), x and y represent the spatial coordinates,

,

Represent the standard deviation in the horizontal direction and the standard deviation in the vertical direction, respectively.

)를 이용하여 소형 표적을 검출하는 것은 적합하지 않을 수 있다.Generally, a small target to be detected in an infrared image is defined as a target having a size of 0.15% or less of the whole image. For example, assuming that the size of the infrared image is 344 x 238, the size of the small target will be between 1 x 1 and 11 x 11. The larger the size of the infrared image, the larger the size range of the small target to be detected.

) May not be suitable for detecting small targets.

Accordingly, in an embodiment of the present invention, a small target can be modeled assuming a Gaussian shape with an average size of small targets of about 5 x 5 and a standard deviation of 1.5.

This is expressed by the following equation (2).

는 최대 크기 소평 표적의 표준편차를 나타내며,

는 소형 크기의 최대 크기의 비율을 나타내며,

는 소형 표적의 최대 크기를 나타낸다.here,

≪ / RTI > represents the standard deviation of the maximum size spot size target,

Represents the ratio of the smallest size to the largest size,

Represents the maximum size of a small target.

Thus, after modeling the small target in the infrared image, the target detection apparatus 100 improves the contrast in a localized manner by reflecting the characteristic of the small target in step 120. [

A method of locally improving contrast by reflecting characteristics of small targets is shown in Fig. Referring to FIG. 2, a description will be given of a method of improving contrast in a region by reflecting features of a small target.

In step 210, the target detection apparatus 100 generates a mask image using the difference between the original image and the Gaussian-blurred image, in order to improve the contrast of the target region locally by reflecting characteristics of the small target.

At this time, the target detection apparatus 100 may apply a value slightly larger than the standard deviation of the smallest-sized small target when applying the Gaussian blur in order to improve the contrast between the small target and the small target.

That is, when the normalized Gaussian kernel is set, the sum of the original image and the smallest target of the maximum size is set to be 1. Thus, the Gaussian blur can be used to improve the contrast of small targets and small target peripheral areas in the input image, by setting the standard deviation slightly larger than the maximum size small target, as defined in equation (3).

는 회선(convolution) 연산을 나타내고,

는 최대 크기 소평 표적의 표준 편차보다 조금 큰 값을 나타내며, 이는 소형 표적과 소평 표적 주변 영역의 명암을 개선하기 위해

이설정될 수 있다.here,

Represents a convolution operation,

Is slightly larger than the standard deviation of the maximal size criterion target to improve the contrast of the small target and the periphery of the target area

Can be determined.

For example, the target detection apparatus 100 can generate a mask image using Equation (4).

Here, N represents a normalization constant for normalizing the mask image to the [0, 1] range. The generated mask image can be used as a weight value of the image which is improved globally in contrast.

In step 215, the target detection apparatus 100 applies a histogram smoothing technique to improve the contrast of the original image globally.

In FIG. 2, the target detection apparatus 100 generates a mask image, and then performs a global contrast enhancement process on the original image. However, each step may be performed in parallel, It is a matter of course.

Further, since the histogram smoothing technique is obvious to those skilled in the art, a detailed description thereof will be omitted.

In step 220, the target detection apparatus 100 applies a mask image to the image to which the histogram smoothing technique is applied, thereby improving the contrast between small targets and small targets. Hereinafter, in order to facilitate understanding and explanation, a result obtained by applying a mask image to an image to which a histogram smoothing technique is applied will be referred to as a global contrast enhancement image.

This can be expressed by the following equation (5).

는 마스크 영상을 나타내고,

는 전역적으로 히스토그램 평활화가 수행된 영상을 나타낸다.here,

Represents a mask image,

Represents an image in which histogram smoothing is performed globally.

In step 225, the target detection apparatus 100 locally improves the contrast by applying a weight value set to a resultant image (i.e., a global contrast enhancement image) obtained by applying a mask image to an image to which the histogram smoothing technique is applied.

This can be expressed by the following equation (6).

는 0 내지 1사이의 실수값을 갖는 명암 개선 정도를 조정하기 위한 가중치이다. 본 발명의 일 실시예에서는

는 0.3으로 설정할 수 있다.here,

Is a weight for adjusting the degree of brightness and darkness having a real value between 0 and 1. In one embodiment of the present invention

Can be set to 0.3.

FIG. 3 shows the result of each process for improving the regional contrast that reflects characteristics of a small target according to an embodiment of the present invention.

In FIG. 3, reference numeral 310 denotes an original image, reference numeral 320 denotes a mask image, reference numeral 330 denotes a result image according to the histogram smoothing, and reference numeral 340 denotes a resultant image obtained by improving the local contrast.

Referring again to FIG. 1, in step 125, the target detection apparatus 100 detects a small target candidate region using a DoG (Difference of Gaussian) filter or a mask image in the resultant image with improved local contrast.

First Embodiment: A small target candidate region is detected using a DoG filter

The DoG filter utilizes the difference of different Gaussian functions. When the 2D DoG filter is expressed by the equation, it is defined as the number 7.

와

는 각 가우시안 함수의 표준편차를 나타낸다. 이때, (

)일 수 있다.Here, G (.) Represents a Gaussian function, x and y represent spatial coordinates,

Wow

Represents the standard deviation of each Gaussian function. At this time, (

).

In order to detect various small targets, a number 7 DoG filter can be extended to the sum of DoG filters passing through different bands.

는 n번째 DoG 필터를 나타낸다.Here, N represents the number of DoG filters,

Represents an n-th DoG filter.

이라고 하면, 수 8은 수 9와 같이 두 가우시안 함수의 차로 간략하게 표현될 수 있다.At this time,

, The number 8 can be briefly expressed as the difference between the two Gaussian functions as in the case of the number 9.

FIG. 4 shows an example of detecting a small target candidate region using a DoG filter according to an embodiment of the present invention.

In FIG. 4, 410a and 410b represent images obtained by applying a mask image to an image to which a histogram smoothing technique is applied, respectively, and applying weight values to the images, and 415a and 415b respectively denote small target candidates An area is detected.

Second Embodiment: A small target candidate region is detected using a mask image

Referring to 415a and 415b in FIG. 4, it can be seen that the mask image generated using the difference between the original image and the Gaussian-blurred image described with reference to FIG. 2 is similar to the mask image.

Both the process of detecting a small target candidate region using the DoG filter and the process of generating a mask image are similar because the standard deviation of the defined small target model is used.

When the mask image of the number 4 is rearranged, it can be summarized as the number 10.

], [

~

])을 통과시키는 것을 알 수 있다.Comparing the number 9 and the number 10, it can be seen that the DoG filter used in the mask image has a slightly wider band ([0 -

], [

~

]) Is passed.

Accordingly, it is possible to detect a small target candidate region using a mask image in order to prevent duplication and increase the speed of execution.

In step 130, the target detection apparatus 100 generates a conspicuous map using spectral residuals in the resultant image with improved local contrast.

That is, the target detection apparatus 100 can generate a bright spot map by using a visual attention model to effectively detect a small target in an infrared image including various clusters, and then apply it as a weight of a small target candidate region.

For example, the target detecting apparatus 100 can generate a salient map using the following equation (11).

는 가우시안 필터를 나타내고,

와

는 퓨리에 변환과 퓨리에 역변환을 나타내며,

이고,

이고,

는 크기가 n인 평균 필터를 나타내며,

는 위상 스펙트럼을 나타내며,

이다.here,

Represents a Gaussian filter,

Wow

Represents a Fourier transform and an inverse Fourier transform,

ego,

ego,

Represents an average filter of size n,

Lt; / RTI > represents the phase spectrum,

to be.

은 퓨리에 변환된 결과의 진폭 스펙트럼을 나타내는 연산자이고,

는 퓨리에 변환된 결과의 위상 스펙트럼을 나타내는 연산자이며, R은 로그-진폭 스펙트럼과 블러링된 로그-진폭 스펙트럼의 차이(Residual)를 나타낸다.

Is an operator representing the amplitude spectrum of the Fourier transformed result,

Is an operator representing the phase spectrum of the Fourier transformed result, and R represents the difference between the log-amplitude spectrum and the blurred log-amplitude spectrum.

That is, the target detection apparatus 100 can generate a clearance map by using the difference between the log amplitude spectrum of the original image and the blurred log-amplitude spectrum as shown in FIG.

In one embodiment of the present invention, since a small target to be detected corresponds to a high frequency component, information of a high frequency component may be lost in down sampling. Accordingly, the target detection apparatus 100 can derive spectral residuals using the original image as it is without down-sampling to prevent information loss occurring in the down-sampling process, and use it to generate a clear map.

FIG. 5 illustrates a dark map generated using spectral residuals.

In FIG. 5, reference numeral 510 denotes an input image, reference numeral 520 denotes a saturated map generated using spectral residual, and reference numeral 530 denotes spectral residual.

In step 135, the target detection apparatus 100 detects a small target by applying a weight map to the detected small target candidate region as a weight.

That is, the target detecting apparatus 100 can apply the generated clearance map using the spectral residual to the small target candidate region, thereby highlighting the visually concentrated region in the candidate region having the Gaussian form.

Applying a salient map to a small target candidate region as a weighting result, a small target region is emphasized and repetitive patterns existing in the background noise such as the inside of the cloud, the boundary and the feature are suppressed.

For example, the target detection apparatus 100 can detect a small target using the number 12.

는 소형 표적 후보 영역을 나타내고,

는 현저 지도를 나타낸다.here,

Represents a small target candidate region,

Represents a standing map.

Then, in step 140, the target detection apparatus 100 binarizes the detected small target to detect the final small target. In one embodiment of the present invention, a small target detected using an adaptive binarization technique such as the Otsu method can be binarized. The adaptive binarization technique is already known to those skilled in the art, so a detailed description thereof will be omitted.

FIG. 6 is a diagram illustrating a result of each step according to the process of detecting a small target in an infrared image according to an embodiment of the present invention.

FIG. 6A shows an input image, FIG. 6B shows a result of improving local contrast, FIG. 6C shows a result of detecting a small target candidate region using a DoG filter or mask image, , (d) represents the residual image generated by using the spectral residual, (e) represents the small target candidate region using the weighted map as the weight, and (f) represents the result of detection of the final small target.

FIG. 7 shows a result of comparing small-sized target detection results according to an embodiment of the present invention.

The images A, B, and C in FIG. 7 include small targets of 9 x 9, 7 x 7, and 7 x 7, respectively, and D images include small targets of 5 x 5 and 7 x 7, This is the result of comparing the small target detection results of the invention.

Referring to FIG. 7, it can be seen that the background removal performance of the average difference filter and the top-hat transformation is drastically degraded when the background is complex. Also, the new top-hat transform has better background removal ability than the top-hat transform. However, in the case of the experimental image C, a lot of background noise remains in a very complicated background environment and in the case of the experimental image D, the target existing in the cloud is removed together with the background Able to know.

On the other hand, it can be seen that the small target detection result according to an embodiment of the present invention improves the target area and effectively removes the background noise in various background environments.

FIG. 8 is a result of quantitatively evaluating the performance according to the conventional method and the small-sized target detection according to the embodiment of the present invention. As shown in FIG. 8, the conventional methods are classified according to the size of the target and the size of the filter The target detection performance is influenced.

However, it can be seen that the small-sized target detection method according to an embodiment of the present invention is excellent in suppressing background noise and enhancing a signal of a small target even when the background is complicated.

9 is a block diagram schematically illustrating an internal configuration of a target detection apparatus for detecting a small target in an infrared image according to an embodiment of the present invention.

9, a target detection apparatus 100 according to an exemplary embodiment of the present invention includes an input unit 910, a target modeling unit 915, a contrast adjustment unit 920, a target candidate region detection unit 925, A map generation unit 930, a target detection unit 935, a memory 940, and a control unit 945.

The input unit 910 functions to acquire an infrared image.

The target modeling unit 915 functions to model a small target to be detected. Since this is the same as that described above with reference to FIG. 1, redundant description will be omitted.

The contrast controller 920 functions to improve the contrast of a region locally by reflecting characteristics of a small target in an infrared image.

For example, the contrast controller 920 may generate a mask image using the difference between the original image and the Gaussian-blurred image to improve the contrast of the region locally by reflecting characteristics of the small target, and perform histogram smoothing Multiply one image by a mask image to improve global contrast, and apply localized weights in advance to improve local contrast.

The target candidate region detection unit 925 performs a function for detecting a small target candidate region in an infrared image using a DoG filter or a mask image. This is the same as that described above with reference to FIG. 1, so duplicate descriptions will be omitted.

The salient map generator 930 generates a salient map using the difference between the log amplitude spectrum of the original image and the blurred log-amplitude spectrum.

The target detection unit 935 detects a small target by applying weighting to the detected small target candidate region and binarizes the detected small target to detect a final small target.

The memory 940 stores various algorithms for detecting a small target in the infrared image according to an embodiment of the present invention, various data generated in the small target detection process, and the like.

The control unit 945 controls the internal components (e.g., the input unit 910, the target modeling unit 915, and the image processing unit) of the target detection apparatus 100 for detecting a small target in the infrared image according to an embodiment of the present invention. A target candidate region detection unit 925, a salient map generation unit 930, a target detection unit 935, a memory 940, and the like).

Meanwhile, the method of detecting a small target in an infrared image according to an embodiment of the present invention may be implemented in a form of a program command that can be executed through a variety of means for processing information electronically and recorded in a storage medium. The storage medium may include program instructions, data files, data structures, and the like, alone or in combination.

Program instructions to be recorded on the storage medium may be those specially designed and constructed for the present invention or may be available to those skilled in the art of software. Examples of storage media include magnetic media such as hard disks, floppy disks and magnetic tape, optical media such as CD-ROMs and DVDs, magneto-optical media such as floptical disks, magneto-optical media and hardware devices specifically configured to store and execute program instructions such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those produced by a compiler, as well as devices for processing information electronically using an interpreter or the like, for example, a high-level language code that can be executed by a computer.

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.

It will be apparent to those skilled in the art that various modifications and variations can be made in the present invention without departing from the spirit or scope of the invention as defined in the appended claims. It will be understood that the invention may be varied and varied without departing from the scope of the invention.

100: target detection device
910:
915: Target modeling unit
920: Contrast adjustment section
925: Target candidate region detection unit
930:
935: Target detection unit
940: Memory
945:

Claims (10)

Modeling a small target in an infrared image;
Generating a mask image using a feature of a small target according to the modeling of the small target;
Performing histogram smoothing on the infrared image and then weighting the mask image to adjust regional contrast;
Detecting a small target candidate region using a Difference of Gaussian (DoG) filter in the infrared image;
Generating spectral maps using spectral residuals of the infrared image; And
And detecting the small target candidate by weighting the salient map to the detected small target candidate region. delete The method according to claim 1,
Wherein the feature of the small target is a standard deviation of a maximum size of the small model target. The method according to claim 1,
Wherein the mask image is generated by a difference between the infrared image and a Gaussian-blurred image reflecting characteristics of the small target. The method according to claim 1,
Wherein the spectral residual is a difference between a log amplitude spectrum of the infrared image and a log-amplitude spectrum of the image blurred by the infrared image. Modeling a small target in an infrared image;
Generating a mask image using a feature of a small target according to the modeling of the small target;
Detecting a small target candidate region in the infrared image using the mask image;
Generating spectral maps using spectral residuals of the infrared image; And
And detecting the small target candidate by weighting the salient map to the detected small target candidate region. The method according to claim 6,
Before the step of detecting the small target candidate region,
Further comprising the steps of: performing histogram smoothing on the infrared image and then weight-applying the mask image to adjust the regional contrast. 8. The method of claim 7,
Wherein adjusting the regional contrast comprises:
Adjusting the global contrast by performing histogram smoothing of the infrared image;
Weighting the mask image with the global contrast adjusted image; And
And adjusting the regional contrast by applying a weighted weighted global contrast adjusted image and a weight for improving the degree of local contrast enhancement to the infrared image. 9. A computer-readable recording medium having recorded thereon a program code for performing the method according to any one of claims 1 to 8. A target modeling unit for modeling a small target in an infrared image;
A contrast control unit for generating a mask image using features of a small target according to the modeling of the small target, performing histogram smoothing on the infrared image, and weighting the mask image to adjust the regional contrast;
A target candidate region detection unit for detecting a small target candidate region using a Difference of Gaussian (DoG) filter in the infrared image;
A star map generation unit for generating a star map using the spectral residual of the infrared image; And
And a target detecting unit for detecting a small target candidate by weighting the dark map on the detected small target candidate region.

Images