KR102550864B1 - Method and System for Object Identification by Patterns of Object Image - Google Patents

Abstract

A method and system for recognizing an object by a pattern of an object image are presented. The object recognition identification method according to the pattern of the object image proposed in the present invention is a first object image (I _A ) and a second object image (I _B ) having the smallest size among a plurality of ROIs having different sizes. Finding an affine transformation (T _f ) for the first ROI by performing a RANSAC alignment algorithm with 1 ROI (v ₁ ), using an affine transformation (T _f ) for the first ROI without performing the RANSAC alignment algorithm Finding an affine transformation (T f,2 ) corresponding to performing the RANSAC alignment algorithm with a second ROI (v ₂ ) having a size larger than the first ROI among the plurality of ROIs, affine transformation (T _{f,2 )} and A second ROI (v ₂ ) for the second object image (I _B ) using four vertices of the ROI image (R _A,1 ) having the size of the second ROI (v ₂ ) for the first object image (I _A ) ) generating an ROI image (R ^* _B,2 ) of size, and an ROI image (R _A,2 ) for the first object image (I _A ) and an ROI image (R for the second object image (I _B )) ^* _B,2 ) and calculating the Hamming distance of each template generated from

Description

Method and system for object identification by pattern of object image {Method and System for Object Identification by Patterns of Object Image}

The present invention relates to a method and system for identifying an object by object recognition using a pattern of an object image.

Recently, with the rapid increase in the number of companion animals nationwide, various problems are also increasing. The government is also implementing policies such as the companion animal registration system and the operation of shelters for abandoned animals, but it is true that they are not enough.

In particular, in the case of the current companion animal registration system, there is the hassle of registering companion animals at the veterinary hospital and separately reporting at the city, county, or district office. There is an inconvenience of attaching a wireless identification device or attaching an identification tag.

In addition, many problems regarding companion animals can be solved through identification registration, but the identification registration rate for companion animals is low due to the side effects and inconvenience of the existing identification registration method, and the problem of abandoned animals is not improved, and the existing artificial wearing or insertion It is necessary to avoid methods that can cause side effects on companion animals, such as In addition, the United States also has serious problems with abandoned animals, but because of the strong negative perception of existing registration methods such as the anti-microchip movement, the need for a registration method without inconvenience or side effects in registration is emerging.

Korean Patent Registration No. 10-1539944 (2015.07.22) Korean Patent Registration No. 10-1706345 (2017.02.07)

The technical problem to be achieved by the present invention is to calculate a dissimilarity score through a new object recognition method based on the pattern of an object image, and to reduce the time required for object recognition while maintaining the accuracy of object recognition. An object of the present invention is to provide a method and system using a plurality of ROIs having different sizes.

The present invention relates to a method and system for identifying an object by object recognition using a pattern of an object image. In order to reduce the time required for object recognition while maintaining the accuracy of object recognition, a plurality of images having different sizes are used in image recognition. We propose a method and system using the ROI of

The larger the size of the ROI, the higher the accuracy of object recognition, but has the disadvantage of taking a long calculation time. In other words, during the object recognition calculation process, the RANSAC alignment algorithm according to various changes (eg, translation, rotation, zoom, etc.) takes the most time, and the calculation time of this step increases as the size of the ROI increases. It takes a long time.

Therefore, in the present invention, in order to reduce the time required for object image pattern recognition while maintaining the accuracy of object image pattern recognition, the RANSAC alignment algorithm, which takes the most time, calculates using a small ROI, and the optimal A large-sized ROI is created using the affine transformation of , and matching is performed using it.

In one aspect, the inscription recognition identification method of a companion animal proposed in the present invention uses RANSAC (Random Sample Consensus) as a region of interest (ROI) for the first inscription image (I _A ) and the second inscription image (I _B ) Finding an affine transformation (T _f ) by performing a sorting algorithm, using the affine transformation (T _f ) and four vertices of the ROI image (R _A ) for the first inscription image (I _A ) Step 2 of generating an ROI image (R _B ) for the inscription image (I _B ), and an ROI image (R _A ) for the first inscription image (I _A ) and an ROI image for the second inscription image (I _B ) ( and calculating a hamming distance of each template generated from R _B ).

In another aspect, the object recognition identification method according to the pattern of the object image proposed in the present invention includes a plurality of ROIs having different sizes for the first object image (I _A ) and the second object image (I _B ). Finding an affine transformation (T _f ) for the first ROI by performing a Random Sample Consensus (RANSAC) alignment algorithm with the first ROI (v ₁ ) having the smallest size among (Region of Interest) An ROI image to which the first ROI (v ₁ ) is applied to one object image (I _A ) is denoted by R _A,1 , and an ROI image to which the first ROI (v ₁ ) is applied to the second object image (I _B ) is Represented by R _B,1- , without performing the RANSAC alignment algorithm, using an affine transformation (T _f ) for the first ROI to a second ROI (v ₂ ) having a size larger than the first ROI among the plurality of ROIs Finding an affine transformation (T _f,2 ) corresponding to that performed by the RANSAC alignment algorithm, an ROI of the size of the second ROI (v ₂ ) for the affine transformation (T _f,2 ) and the first object image (I _A ) generating an ROI image (R _{* B,2} ) having a size of a second ROI (v ₂ ) for a second object image (I _B ) by using four vertices of the image (R ^A _,2 ); and Calculate the hamming distance of each template generated from the ROI image (R _A,2 ) of the image (I _A ) and the ROI image (R ^* _B,2 ) of the second object image (I _B ) It includes steps to

Corresponding to performing the RANSAC alignment algorithm with a second ROI (v ₂ ) having a larger size than the first ROI among a plurality of ROIs using an affine transformation (T _f ) for the first ROI without performing the RANSAC alignment algorithm The step of finding the affine transformation (T _f,2 ) to calculate the affine transformation (T _f,2 ) using the following equation,

,

,

이므로, here

,

,

Because of,

일 때,

이다.

when,

am.

In order to reduce the time required for recognition using the pattern of an object image, the object recognition identification method according to the pattern of the object image proposed in the present invention uses the RANSAC alignment algorithm to select the first ROI (v ₁ ) having the smallest size among the plurality of ROIs To obtain an affine transformation (T _f ) for the first ROI, and to increase the accuracy of recognition using the pattern of the object image, the affine transformation (T _f ) for the first ROI is used to obtain a first one of the plurality of ROIs. The affine transformation (T _f,2 ) corresponding to the RANSAC alignment algorithm was performed with the second ROI (v ₂ ) having a size larger than 1 ROI, and the Hamming distance was calculated with the generated template, and the Hamming distance calculation of the template In order to increase accuracy, the RANSAC alignment algorithm is used to find the optimal affine transformation, and the change in coordinate information of each ROI of the first object image (I _A ) and the second object image (I _B ) and the misalignment of the ROI image are excluded use a technique that

In another aspect, the object recognition identification system according to the pattern of the object image proposed in the present invention is a plurality of ROIs having different sizes for the first object image I _A and the second object image I _B . A RANSAC alignment algorithm for finding an affine transformation (T _f ) for the first ROI by performing a RANSAC (Random Sample Consensus) alignment algorithm with the first ROI (v ₁ ) having the smallest size among (Region of Interest) Execution Unit - An ROI image to which the first ROI (v ₁ ) is applied to the first object image (I _A ) is represented by R _A,1 , and the first ROI (v ₁ ) is applied to the second object image (I _B ). The applied ROI image is represented by R _B,1- , a second ROI having a larger size than the first ROI among a plurality of ROIs by using an affine transformation (T _f ) for the first ROI without performing the RANSAC alignment algorithm ( v ₂ ), an affine transformation performer that finds an affine transformation (T _f,2 ) corresponding to the RANSAC alignment algorithm, and a second ROI for the affine transformation (T _f,2 ) and the first object image (I _A ) An ROI image (R _{* B,2} ) of the size of the second ROI (v ₂ ) is obtained for the second object image (I _B ) using four vertices of the ROI image (R ^A _{, 2} ) of size (v 2 ). ROI area generation unit to generate and each generated from the ROI image (R _A,2 ) for the first object image (I _A ) and the ROI image (R ^* _B,2 ) for the second object image (I _B ) A Hamming distance calculation unit for calculating a Hamming distance of the template is included.

According to embodiments of the present invention, a dissimilarity score is calculated through a new object recognition method based on a pattern of an object image, and object recognition accuracy is maintained by using a plurality of ROIs having different sizes in image recognition. while reducing the time required for object recognition.

에 대한 비교와 ROC 곡선을 나타내는 도면이다.
도 4는 본 발명의 일 실시예에 따른 개체 이미지의 패턴에 의한 개체 인식 식별 방법을 설명하기 위한 흐름도이다.
도 5는 본 발명의 일 실시예에 따른 개체 이미지의 패턴에 의한 개체 인식 식별 시스템의 구성을 나타내는 도면이다. 1 is a diagram showing a sample of a nose image for object recognition according to an embodiment of the present invention.
2 is a diagram illustrating phase codes for ROIs of a nose image for object recognition according to an embodiment of the present invention.
3 is a dissimilarity score according to an embodiment of the present invention

It is a diagram showing the comparison and the ROC curve for .
4 is a flowchart illustrating a method for recognizing and identifying an object based on a pattern of an object image according to an embodiment of the present invention.
5 is a diagram showing the configuration of an object recognition identification system based on a pattern of an object image according to an embodiment of the present invention.

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

1 is a diagram showing a sample of a nose image for object recognition according to an embodiment of the present invention.

In the present invention, a nose image of a companion animal will be described as an example of an object recognition and identification method based on a pattern of an object image. Here, the nose image of a companion animal is only an example and is not limited thereto, and an object recognition and identification method based on a pattern of an object image proposed for various other objects may be applied.

Figure 1 (a) is a view showing a nose image captured from the front according to an embodiment of the present invention, Figure 1 (b) is a segmentation result by the deep learning method in the nose image shown in Figure 1 (a) FIG. 1(c) is a diagram showing guide lines for obtaining an ROI image from the nose image shown in FIG. 1(a), and FIG. 1(d) shows an ROI area cut by these guide lines. it is a drawing

Biometric recognition of a companion animal (eg, a puppy) by a muzzle pattern is performed by comparing a muzzle pattern of a corresponding companion animal. Therefore, the first step in muzzle pattern identification is to extract the muzzle pattern from the nose image. To extract the muzzle pattern, we first need to identify a Region of Interest (ROI) in a given nose image. Since the nose is a three-dimensional object, the captured muzzle pattern may be greatly distorted depending on the viewing angle. Therefore, it is important to keep the capture direction somewhat consistent in order to increase the accuracy of identification. Taking a nose image from the front is natural and intuitive, so it is assumed that the nose image is taken from the front in the present invention. Figure 1 (a) shows a nose image sample captured from the front. The most prominent features in the captured nose image are the nostrils and philtrum. For this reason, a partial area between the two nostrils is used as an ROI.

를 통해 좌표계를 제공한다. 여기서

는 선들의 두 교차점의 중간 점이고,

는

에서 오른쪽 교차점까지의 벡터이다. 이후,

는 이미지 평면의 직교 기준(orthogonal basis)이다. 여기서

는

를 90도 회전에서 얻은 벡터를 나타낸다. 직교 기준

로 정의된 좌표계를 사용하여, 직사각형 영역을 ROI로 사용한다. 도 1(d)는 이러한 구성을 사용하여 도 1(a)의 코 이미지에서 잘린 ROI를 나타낸다. Therefore, to extract ROIs, we first need to identify the nostrils and filtrum in a given nose image. In the present invention, the nostril and filtrum regions are segmented using a deep learning method. Figure 1 (b) shows the result of segmentation by the deep learning method in the image shown in Figure 1 (a). Once these areas are detected, it is easy to find two perpendicular lines that touch each nostril area between the two nostrils, perpendicular to the tangent to the upper tangent of the two nostril areas. Figure 1(c) shows these lines on top of the nose image shown in Figure 1(a). These lines naturally coordinate information

Provides a coordinate system through here

is the midpoint of the two intersections of the lines,

Is

is the vector from to the right intersection. after,

is the orthogonal basis of the image plane. here

Is

represents the vector obtained from a 90 degree rotation. Cartesian standard

Using the coordinate system defined by , a rectangular region is used as the ROI. Figure 1(d) shows an ROI cropped from the nose image of Figure 1(a) using this configuration.

2 is a diagram illustrating phase codes for ROIs of a nose image for object recognition according to an embodiment of the present invention.

When a rectangular ROI is acquired, a muzzle pattern is encoded into a phase code by applying various frequency transforms or wavelet transforms to the rectangular region. For simplicity, a phase code can be thought of as a two-dimensional bit array. In the present invention, we use the same phase code as Daugman [2] used for human iris recognition defined using the complex 2D Gabor wavelet as follows:

)와 허수부(

)는 이진수(0 또는 1)이다. 자세한 내용은 Daugman [2]를 참조한다. 도 1(d)에서 잘린 ROI에 대한 위상 코드

은 도 2에 나타내었다. In particular, the real part (

) and the imaginary part (

) is a binary number (0 or 1). See Daugman [2] for details. Phase code for cropped ROI in Fig. 1(d)

is shown in Figure 2.

Muzzles are usually wet and may be obscured by some object, such as hair, so parts of the muzzle pattern are damaged and can be ignored in comparison. For this purpose, a masking bit vector indicating which parts of the phase code should be ignored as artifacts must also be calculated. In the present invention, it is assumed that the size of the masking bit vector is equal to the size of the phase code, and a value of 0 in the masking bit vector indicates that the corresponding bit value in the phase code is somehow corrupted and ignored.

) 및 마스크(

)로 템플릿

를 나타내기 위해

표기법을 사용한다.Thus, each muzzle pattern is encoded with a pair of bit vectors: a phase code and a masking bit vector called a mask. For simplicity, when referring to muzzle patterns, phase code and mask pairs are simply referred to as muzzle templates or templates. In the following, the phase code (

) and mask (

) as template

to indicate

use notation.

A dissimilarity score or distance score between two muzzle patterns is measured by comparing corresponding templates. The most well-known basic distance between two templates is the fractional Hamming distance as in [2]. However, if most of the muzzle phase codes are occluded, the partial Hamming distance may misrepresent the comparison result. Therefore, in the present invention, the dissimilarity of muzzle templates is measured using the modified partial Hamming distance proposed in [22].

는 다음과 같이 정의된다: Modified partial Hamming distance or simply Modified Hamming distance between template A and template B

is defined as:

와

는 각각 머즐 템플릿 A와 B의 위상 코드를 나타낸다;

및

는 각각 머즐 템플릿 A 및 B의 마스크를 나타낸다. 거리

의 정의에서,

는 비트 부울 XOR 연산자,

는 비트 AND 연산자,

은 비트 NOT 연산자,

는 비트 벡터의 표준이고

은 위상 코드의 비트 수를 나타낸다. 이러한 거리의 기본 아이디어는 각각의 유효하지 않은 비트에 대해 비 유사도 점수 0.5를 제공하는 것이다. here

and

denote the phase codes of muzzle templates A and B, respectively;

and

denotes the masks of muzzle templates A and B, respectively. distance

In the definition of

is the bitwise Boolean XOR operator,

is the bitwise AND operator,

is the bitwise NOT operator,

is the standard of the bit vector and

represents the number of bits of the phase code. The basic idea of these distances is to give a dissimilarity score of 0.5 for each invalid bit.

According to an embodiment of the present invention, it is assumed that the nose image is taken from the front. However, it is impossible to always take nose images from the same direction in 3D space. There may also be segmentation errors in identifying ROIs. Therefore, it should not be assumed that the same muzzle pattern is encoded in the same part of the muzzle phase code of the same object. In order to compensate for the difference between the two cropped ROIs of the two nose images for comparison, codes shifting is applied during comparison. More specifically, sub-blocks are independently performed without moving phase codes as a whole. This is because what actually works to capture a 3D object into a 2D image is a projection transformation, so what needs to be applied to compare two phase codes is a non-linear transformation. With this idea, we first divide the phase code into several sub-blocks. Masks are also divided into sub-blocks in the same way. The segmented phase code pair and the corresponding segmentation mask are simply referred to as sub-blocks of the muzzle template. Then, the two sub-blocks are compared in the following way:

와

는 각각 템플릿 A와 B의 k 번째 하위 블록을 나타내고

는

에서 수평으로

픽셀, 수직으로

픽셀만큼 이동한 하위 블록을 나타낸다.

는 거리 점수 계산에 적용할 이동 세트

를 나타낸다. 블록 거리 점수

가 계산되면 최종 거리 점수

이 블록 거리의 기하학적 평균으로 주어진다: here

and

denotes the kth subblock of templates A and B, respectively.

Is

horizontally from

pixels vertically

Indicates a sub-block moved by a pixel.

is the set of moves to be applied in calculating the distance score

indicates block distance score

is calculated, the final distance score

The geometric mean of this block distance is given by:

where A and B represent muzzle templates to be compared and m is the number of subblocks in the template.

에 대한 비교와 ROC 곡선을 나타내는 도면이다. 3 is a dissimilarity score according to an embodiment of the present invention

It is a diagram showing the comparison and the ROC curve for .

In order to measure the performance of the object identification algorithm proposed in the present invention, a comparative test was performed with a nose image dataset for object recognition.

1080 해상도의 CCD 카메라를 사용하여 수집되었다. 정반사를 최소화하기 위해 코에 덮개를 사용했다. In this dataset the image is 1920

It was collected using a CCD camera with 1080 resolution. A cover was used on the nose to minimize specular reflection.

에 대한 비교 결과를 나타낸다. ROC 곡선에서 알 수 있듯이, 알고리즘의 정확도는 실제 목적에 충분하지 않다. 주요 문제점 중 하나는 동일한 개체를 비교한 경우(genuine)의 비교 점수의 분포가 오른쪽에 상대적으로 헤비 테일(heavy tail)을 갖는 다는 것이다. 이러한 동일한 개체를 비교한 경우(genuine)의 높은 비교 점수는 주로 이미지가 비교되는 두 비 유사도 ROI가 다른 방향에서 촬영될 때 서로 정렬되지 않았기 때문이다. 이러한 결과는 다른 투영 변환으로 얻은 두 이미지를 정렬하는 방법이 필요함을 의미한다. 다음에서 두 이미지를 정렬하는 방법에 대해 설명한다.3 is a dissimilarity score

Indicates the comparison result for . As can be seen from the ROC curve, the accuracy of the algorithm is not good enough for practical purposes. One of the major problems is that the distribution of comparison scores when comparing identical individuals (genuine) has a relatively heavy tail on the right. The high comparison score in the case of comparing the same object (genuine) is mainly because the two dissimilarity ROIs in which images are compared are not aligned with each other when taken from different directions. These results imply that we need a way to align the two images obtained with different projection transformations. The following describes how to align the two images.

와

가 동일한 피사체에서 찍은 두 개의 다른 코 이미지이고

와

가 각각 코 이미지

와

에서 잘라낸 ROI라고 가정한다.

를

와 정렬하기 위해,

의 어느 부분이

의 어느 부분에 해당하는지에 대한 정보가 필요하다. 이 정보를 얻기 위해

의 포인트

을 샘플링하고 각각의 포인트

에 대하여

의 가장 가까운 포인트

를 찾는다. 이를 위해, 중심이

인

의 하위 직사각형

는 크기가

와 동일한

의 가능한 모든 하위 직사각형과 비교된다. 두 개의 직사각형을 비교하여 두 개의 하위 직사각형의 두 개의 해당 하위 템플릿의 비 유사도 점수

를 측정한다. 그러면, 가장 가까운 포인트

는

에서

에 가장 가까운 하위 직사각형의 중심이다.

and

are two different nose images taken from the same subject, and

and

are the respective nose images

and

Assume that the ROI is cropped from .

cast

to align with

which part of

Information on which part of the to get this information

point of

and for each point

about

nearest point of

look for For this purpose, the center

person

sub-rectangle of

is the size

same as

All possible sub-rectangles of are compared. The dissimilarity score of the two corresponding sub-templates of the two sub-rectangles by comparing the two rectangles

to measure Then, the nearest point

Is

at

is the center of the sub-rectangle closest to .

의

가

의

에 해당한다는 정보 바탕으로 RANSAC(Random Sample Consensus) 패러다임 [21]을 적용하여 최적의 변환 T를 피팅한다.

of

go

of

Based on the information that corresponds to , the RANSAC (Random Sample Consensus) paradigm [21] is applied to fit the optimal transformation T.

The Random Sample Consensus (RANSAC) sorting algorithm is as follows:

에서 m 개의 포인트의 부분 집합

을 랜덤으로 선택하고, 하기식의 제곱 오차를 최소화하는 최상의 아핀 변환(affine transform) T를 찾는다: One.

subset of m points in

is randomly selected and finds the best affine transform T that minimizes the squared error of the equation:

의 포인트

는

의 포인트

에 해당한다. here

point of

Is

point of

corresponds to

내에 있는, 즉 하기식을 만족하는, 포인트

의 부분 집합

을 결정한다: 2. Some error tolerance in L using the fitted affine transformation T

Points that are within, that is, that satisfy the following equation

subset of

determines:

은 T에 의한 합의 세트(consensus set) 라고 한다.subset

is called the consensus set by T.

3. Steps 1 and 2 are repeated according to a predetermined number of trials, and T having the largest set of sums is the final output.

에서 정렬된 ROI

는 역 4 포인트

로 정의된다. 여기서

는

의 경계 직사각형의 4 개의 정점이다. RANSAC 정렬 알고리즘을 적용하면, 새로운 비 유사도 점수

이 다음과 같이 계산된다: Using the fitted affine transform T, the nose image

ROIs sorted from

is inverse 4 points

is defined as here

Is

are the four vertices of the bounding rectangle of Applying the RANSAC sorting algorithm, the new dissimilarity score

is calculated as:

여기서

는 머즐 템플릿 A, 잘린 ROI

에서 생성된 머즐 템플릿 B, 및 RANSAC 정렬 알고리즘에 의해 정렬 된 ROI

에서 생성된 머즐 템플릿

와 비교할 머즐 이미지이다. 점수

은 정렬된 ROI를 얻기 위해 원래 코 이미지가 필요하기 때문에 두 머즐 템플릿 사이가 아니라 머즐 템플릿 A와 코 이미지 사이의 비 유사도를 측정한다.

here

is muzzle template A, cropped ROI

Muzzle template B generated from , and ROIs aligned by the RANSAC alignment algorithm

Muzzle Templates Generated from

This is the muzzle image to compare with. score

measures the dissimilarity between the muzzle template A and the nose image, rather than between the two muzzle templates, since the original nose image is needed to obtain an aligned ROI.

4 is a flowchart illustrating a method for recognizing and identifying an object based on a pattern of an object image according to an embodiment of the present invention.

The object recognition identification method according to the pattern of the object image proposed uses the smallest size among a plurality of regions of interest (ROIs) having different sizes for the first object image (I _A ) and the second object image (I _B ). Finding an affine transformation (T _f ) for the first ROI by performing a RANSAC (Random Sample Consensus) alignment algorithm with the first ROI (v ₁ ) having (410), without performing the RANSAC alignment algorithm, An affine transformation (T _{f , 2} ) is found (420), using the affine transformation (T _f,2 ) and the four vertices of the ROI image (R _A,2 ) having the size of the second ROI (v ₂ ) for the first object image (I _A ). generating an ROI image (R ^* _{B ,} 2) having a size of a second ROI (v ₂ ) for the second object image (I B ) (430) and an ROI image (for the first object image (I _A )) A step 440 of calculating a hamming distance of each template generated from R _A,2 ) and the ROI image (R ^* _B,2 ) of the second object image (I _B ).

Here, the ROI image to which the first ROI (v ₁ ) is applied to the first object image (I _A ) is represented by R _A,1 , and the first ROI (v ₁ ) to the second object image (I _B ) is applied. ROI images are denoted by R _B,1 .

Parameter R _A,k used in an embodiment of the present invention represents a v _k type ROI image generated from the first object image I _A (k = 1, 2). T _A,k : I _A plane→ R _A,k plane represents conversion from the first object image I _A to R _A,k .

A parameter R _B,k used in an embodiment of the present invention represents an ROI image in the form of v _k generated from the second object image I _B (k = 1, 2). T _B,k : I _B plane→ R _B,k plane represents conversion from the second object image I _B to R _B,k .

Hereinafter, a method for recognizing and identifying an object based on a pattern of an object image according to an embodiment of the present invention will be described in detail.

In the present invention, a nose image of a companion animal will be described as an example of an object recognition and identification method based on a pattern of an object image. Here, the nose image of a companion animal is only an example and is not limited thereto, and an object recognition and identification method based on a pattern of an object image proposed for various other objects may be applied.

According to an embodiment of the present invention, when creating a rectangular region-of-interest ROI from a muzzle image, the following two types can be used depending on the horizontal and vertical sizes of the ROI:

v ₁ : 64x96 (ROI size width x height, unit is pixels)

_v2 : 256x384

These two ROI types are examples only, and other types and larger numbers of ROIs may be used.

The larger the size of the ROI, the higher the accuracy of muzzle pattern recognition, but has the disadvantage of taking a long calculation time. In other words, during the muzzle pattern recognition calculation process, the RANSAC alignment algorithm according to various changes (e.g., translation, rotation, zoom, etc.) takes the most time, and the calculation time for this step is the size of the ROI. The bigger it is, the longer it takes.

Comparing the time required for full matching including RANSAC alignment, v ₁ can match about 4 times faster than v ₂ .

Therefore, in the present invention, in order to reduce the time required for object image pattern recognition while maintaining the accuracy of object image pattern recognition, the RANSAC alignment algorithm, which takes the most time, is calculated using a small ROI, and the optimal A large-sized ROI is created using the affine transformation of , and matching is performed using it.

First, in step 410, RANSAC alignment with a first ROI (v ₁ ) having the smallest size among a plurality of ROIs having different sizes for the first object image I _A and the second object image I _B An algorithm is performed to find the affine transformation (T _f ) for the first ROI.

인 행렬 A와 b를 계산한다. The RANSAC alignment algorithm is applied to R _A,1 and R _B,1 using the first ROI (v ₁ ), and an optimal affine transformation T _f : R _B,1 → R _A,1 is calculated. in other words,

Calculate matrices A and b that are

In step 420, RANSAC alignment with a second ROI (v ₂ ) having a larger size than the first ROI among the plurality of ROIs using an affine transformation (T _f ) for the first ROI without performing the RANSAC alignment algorithm Find the affine transformation (T _f,2 ) that corresponds to the algorithm performed.

Using the second ROI (v ₂ ), the affine transformation T _f, 2 between R _A,2 and R _B, 2 : R _B,2 → R _A,2 is calculated as follows:

,

,

이므로, here

,

,

Because of,

일 때,

이다. 여기서

는 제1 ROI 사이즈와 제2 ROI 사이즈의 가로,세로의 비율을 나타내고,

는 일 실시예일뿐 제1 ROI 사이즈와 제2 ROI 사이즈에 대한 다양한 비율이 적용될 수 있다.

when,

am. here

Represents the horizontal and vertical ratio of the first ROI size and the second ROI size,

is only an example, and various ratios for the first ROI size and the second ROI size may be applied.

In step 430, the affine transformation (T _f,2 ) and the four vertexes of the ROI image (R _A, 2 ) having the size of the second ROI (v ₂ ) for the first object image (I _A ) are used to generate the first object image (I A ). An ROI image (R ^* _B, 2 ) having a size of the second ROI (v ₂ ) is generated for the 2-object image (I _B ).

를 구하고(단,

는 ROI 이미지(R_A,2)의 4개의 꼭짓점), 제2 개체 이미지(I_B)에서 꼭짓점이

인 영역으로부터 v₂형태의 ROI 이미지 R^* _B,2을 생성한다. Using the affine transformation (T _f,2 ) calculated in step 420

Find (however,

is the four vertices of the ROI image (R _A,2 ), and the vertices in the second object image (I _B )

An ROI image R ^* _B,2 in the form of v ₂ is created from the area of .

In step 440, each template generated from the ROI image (R _A,2 ) for the first object image (I _A ) and the ROI image (R ^* _B,2 ) for the second object image (I _B ) Calculate the Hamming distance of In other words, the dissimilarity score is calculated by calculating the Hamming distance between template A generated from R _A,2 and template B ^* generated from R ^* _B,2 .

Through this, it is possible to perform matching that maintains the accuracy of v ₂ while the speed is similar to that of using the ROI image in the form of v ₁ .

In order to reduce the time required for recognition using the pattern of the object image, the RANSAC alignment algorithm is performed with the first ROI (v ₁ ) having the smallest size among the plurality of ROIs to obtain an affine transformation (T _f ) for the first ROI and RANSAC with a second ROI (v ₂ ) having a larger size than the first ROI among a plurality of ROIs by using an affine transformation (T _f ) for the first ROI to increase the accuracy of recognition using the pattern of the object image. Calculate the Hamming distance with the generated template by finding the affine transformation (T _f,2 ) corresponding to the one performed by the alignment algorithm. In addition, in order to increase the accuracy of calculating the Hamming distance of the template, the optimal affine transformation is found using the RANSAC alignment algorithm as described above, and the first object image (I _A ) and the second object image (I _B ) of each ROI A technique that excludes changes in coordinate information and misalignment of ROI images is used.

5 is a diagram showing the configuration of an object recognition identification system based on a pattern of an object image according to an embodiment of the present invention.

The object recognition identification system 500 according to this embodiment may include a processor 510, a bus 520, a network interface 530, a memory 540, and a database 550. The memory 540 may include an operating system 541 and an object recognition identification routine 542 based on a pattern of an object image. The processor 510 may include a RANSAC alignment algorithm performer 511 , an affine transform performer 512 , an ROI area generator 513 and a Hamming distance calculator 514 . In other embodiments, entity recognition identification system 500 may include more components than those of FIG. 5 . However, there is no need to clearly show most of the prior art components. For example, entity recognition identification system 500 may include other components such as a display or a transceiver.

The memory 540 is a computer-readable recording medium and may include a random access memory (RAM), a read only memory (ROM), and a permanent mass storage device such as a disk drive. In addition, the memory 540 may store program codes for the operating system 541 and the object recognition identification routine 542 based on the pattern of the object image. These software components may be loaded from a computer-readable recording medium separate from the memory 540 using a drive mechanism (not shown). The separate computer-readable recording medium may include a computer-readable recording medium (not shown) such as a floppy drive, a disk, a tape, a DVD/CD-ROM drive, and a memory card. In another embodiment, software components may be loaded into the memory 540 through the network interface 530 rather than a computer-readable recording medium.

Bus 520 may enable communication and data transfer between components of object recognition identification system 500 . Bus 520 may be configured using a high-speed serial bus, a parallel bus, a storage area network (SAN), and/or other suitable communication technology.

The network interface 530 may be a computer hardware component for connecting the entity recognition identification system 500 to a computer network. The network interface 530 may connect the entity recognition identification system 500 to a computer network through a wireless or wired connection.

The database 550 may play a role of storing and maintaining all information necessary for object recognition and identification based on a pattern of an object image. 5 shows that the database 550 is built and included inside the object recognition identification system 500, but is not limited thereto, and may be omitted depending on the system implementation method or environment, or all or part of the database It is also possible to exist as an external database built on a separate and different system.

The processor 510 may be configured to process commands of a computer program by performing basic arithmetic, logic, and input/output operations of the entity recognition and identification system 500 . Instructions may be provided to processor 510 by memory 540 or network interface 530 and via bus 520 . The processor 510 may be configured to execute program codes for the RANSAC alignment algorithm performer 511 , the affine transform performer 512 , the ROI area generator 513 and the Hamming distance calculator 514 . These program codes may be stored in a recording device such as memory 540 .

The RANSAC alignment algorithm performer 511, the affine transform performer 512, the ROI region generator 513, and the Hamming distance calculator 514 may be configured to perform steps 410 to 440 of FIG. there is.

The object recognition identification system 500 may include a RANSAC alignment algorithm performer 511, an affine transform performer 512, an ROI area generator 513, and a Hamming distance calculator 514.

In the present invention, in order to reduce the time required for object image pattern recognition while maintaining the accuracy of object image pattern recognition, the RANSAC alignment algorithm, which takes the most time, is calculated using a small ROI, and the optimal affine obtained from this is calculated. A large-sized ROI is created using transformation and matching is performed using it.

The RANSAC alignment algorithm execution unit 511 sets the first ROI (v ₁ ) having the smallest size among a plurality of ROIs having different sizes for the first object image I _A and the second object image I _B . The RANSAC alignment algorithm is performed to find the affine transformation (T _f ) for the first ROI.

인 행렬 A와 b를 계산한다. The RANSAC alignment algorithm is applied to R _A,1 and R _B,1 using the first ROI (v ₁ ), and an optimal affine transformation T _f : R _B,1 → R _A,1 is calculated. in other words,

Calculate matrices A and b that are

The affine transformation performer 512 does not perform the RANSAC alignment algorithm, but uses the affine transformation (T _f ) for the first ROI to convert it into a second ROI (v ₂ ) having a larger size than the first ROI among a plurality of ROIs. Find an affine transformation (T _f,2 ) that corresponds to performing the RANSAC sorting algorithm.

Using the second ROI (v ₁ ), the affine transformation T _f, 2 between R _A,2 and R _B, 2 : R _B,2 → R _A,2 is calculated as follows:

,

,

이므로, here

,

,

Because of,

일 때,

이다.

when,

am.

The ROI area generating unit 513 uses the affine transformation (T _f,2 ) and the four vertices of the ROI image (R _A,2 ) having the size of the second ROI (v ₂ ) for the first object image (I _A ). to generate an ROI image (R ^* _B,2 ) having a size of the second ROI (v ₂ ) for the second object image (I _B ).

를 구하고(단,

는 ROI 이미지(R_A,2)의 4개의 꼭짓점), 제2 개체 이미지(I_B)에서 꼭짓점이

인 영역으로부터 v₂형태의 ROI 이미지 R^* _B2을 생성한다. Using the affine transformation (T _f,2 ) calculated by the affine transformation performer 512

Find (however,

is the four vertices of the ROI image (R _A,2 ), and the vertices in the second object image (I _B )

An ROI image R ^* _B2 in the form of v ₂ is created from the in region.

The hamming distance calculation unit 514 generates ROI images (R _A,2 ) for the first object image (I _A ) and ROI images (R ^* _B,2 ) for the second object image (I _B ), respectively. Calculate the Hamming distance of the template of In other words, the dissimilarity score is calculated by calculating the Hamming distance between template A generated from R _A,2 and template B ^* generated from R ^* _B,2 .

Through this, it is possible to perform matching that maintains the accuracy of v ₂ while the speed is similar to that of using the ROI image in the form of v ₁ .

In order to reduce the time required for recognition using the pattern of the object image, the RANSAC alignment algorithm is performed with the first ROI (v ₁ ) having the smallest size among the plurality of ROIs to obtain an affine transformation (T _f ) for the first ROI and RANSAC with a second ROI (v ₂ ) having a larger size than the first ROI among a plurality of ROIs by using an affine transformation (T _f ) for the first ROI to increase the accuracy of recognition using the pattern of the object image. Calculate the Hamming distance with the generated template by finding the affine transformation (T _f,2 ) corresponding to the one performed by the alignment algorithm. In addition, in order to increase the accuracy of calculating the Hamming distance of the template, the optimal affine transformation is found using the RANSAC alignment algorithm as described above, and the first object image (I _A ) and the second object image (I _B ) of each ROI A technique that excludes changes in coordinate information and misalignment of ROI images is used.

The devices described above may be implemented as hardware components, software components, and/or a combination of hardware components and software components. For example, devices and components described in the embodiments may include, for example, a processor, a controller, an arithmetic logic unit (ALU), a digital signal processor, a microcomputer, a field programmable array (FPA), It may be implemented using one or more general purpose or special purpose computers, such as a programmable logic unit (PLU), microprocessor, or any other device capable of executing and responding to instructions. A processing device may run an operating system (OS) and one or more software applications running on the operating system. A processing device may also access, store, manipulate, process, and generate data in response to execution of software. For convenience of understanding, there are cases in which one processing device is used, but those skilled in the art will understand that the processing device includes a plurality of processing elements and/or a plurality of types of processing elements. It can be seen that it can include. For example, a processing device may include a plurality of processors or a processor and a controller. Other processing configurations are also possible, such as parallel processors.

Software may include a computer program, code, instructions, or a combination of one or more of the foregoing, which configures a processing device to operate as desired or processes independently or collectively. You can command the device. Software and/or data may be any tangible machine, component, physical device, virtual equipment, computer storage medium or device, intended to be interpreted by or provide instructions or data to a processing device. can be embodied in Software may be distributed on networked computer systems and stored or executed in a distributed manner. Software and data may be stored on one or more computer readable media.

The method according to the embodiment may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer readable medium. The computer readable medium may include program instructions, data files, data structures, etc. alone or in combination. Program commands recorded on the medium may be specially designed and configured for the embodiment or may be known and usable to those skilled in computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tapes, optical media such as CD-ROMs and DVDs, and magnetic media such as floptical disks. - includes hardware devices specially configured to store and execute program instructions, such as magneto-optical media, and ROM, RAM, flash memory, and the like. Examples of program instructions include high-level language codes that can be executed by a computer using an interpreter, as well as machine language codes such as those produced by a compiler.

As described above, although the embodiments have been described with limited examples and drawings, those skilled in the art can make various modifications and variations from the above description. For example, the described techniques may be performed in an order different from the method described, and/or components of the described system, structure, device, circuit, etc. may be combined or combined in a different form than the method described, or other components may be used. Or even if it is replaced or substituted by equivalents, appropriate results can be achieved.

Therefore, other implementations, other embodiments, and equivalents of the claims are within the scope of the following claims.

<References>

[1] J. G. Daugman, "High confidence visual recognition of persons by a test of statistical independence," IEEE Transactions on Pattern Analysis and Machine Intelligence, vol. 15, no. 11, p. 1148-1161, Nov 1993.

[2] J. Daugman, "How iris recognition works," Circuits and Systems for Video Technology, IEEE Transactions on, vol. 14, no. 1, p. 21-30, Jan 2004.

[3] ――, "Probing the uniqueness and randomness of iriscodes: Results from 200 billion iris pair comparisons," Proceedings of the IEEE, vol. 94, no. 11, p. 1927-1935, Nov 2006.

[4] J. G. Daugman, "Complete discrete 2-D gabor transforms by neural networks for image analysis and compression," IEEE Transactions on Acoustics, Speech, and Signal Processing, vol. 36, no. 7, p. 1169-1179, July 1988.

[5] S. Kumar, A. Pandey, K. S. R. Satwik, S. Kumar, S. K. Singh, A. K. Singh, and A. Mohan, "Deep learning framework for recognition of cattle using muzzle point image pattern," Measurement, vol. 116, p. 1 - 17, 2018. [Online]. Available: http://www.sciencedirect.com/science/article/pii/S0263224117306991

[6] A. Noviyanto and A. M. Arymurthy, "Beef cattle identification based on muzzle pattern using a matching refinement technique in the sift method," Computers and Electronics in Agriculture, vol. 99, p. 77 - 84, 2013. [Online]. Available: http://www.sciencedirect.com/science/article/pii/S0168169913002093

[7] A. I. Awad, "From classical methods to animal biometrics: A review on cattle identification and tracking," Computers and Electronics in Agriculture, vol. 123, p. 423 - 435, 2016. [Online]. Available: http://www.sciencedirect.com/science/article/pii/S0168169916300837

[8] T. Gaber, A. Tharwat, A. E. Hassanien, and V. Snasel, "Biometric cattle identification approach based on weber's local descriptor and adaboost classifier," Computers and Electronics in Agriculture, vol. 122, p. 55 - 66, 2016. [Online]. Available: http://www.sciencedirect.com/science/article/pii/S016816991600003X

[9] A. Noviyanto and A. Arymurthy, "Automatic cattle identification based on muzzle photo using speed-up robust features approach," Proceedings of the 3rd European Conference of Computer Science, pp. 110-114, 01 2012.

[10] S. Kumar, S. Chandrakar, A. Panigrahi, and S. K. Singh, "Muzzle point pattern recognition system using image pre-processing techniques," in 2017 Fourth International Conference on Image Information Processing (ICIIP), Dec 2017, pp. . 1-6.

[11] H. M. E. Hadad, H. A. Mahmoud, and F. A. Mousa, "Bovines muzzle classification based on machine learning techniques," Procedia Computer Science, vol. 65, pp. 65; 864 - 871, 2015, international Conference on Communications, management, and Information technology (ICCMIT'2015). [Online]. Available: http://www.sciencedirect.com/science/article/pii/S1877050915028744

[12] "Muzzle classification using neural networks," The International Arab Journal of Information Technology, vol. 14, no. 4.

[13] A. EDWIN and M. GEORGE, "A fuzzy mathematical approach to eigen muzzle print recognition," International J. of Math. Sci. and Engg. Appls. (IJMSEA), vol. 11, no. 1, p. 57-65, April 2017.

[14] A. I. Awad, A. E. Hassanien, and H. M. Zawbaa, "A cattle identification approach using live captured muzzle print images," in Advances in Security of Information and Communication Networks, A. I. Awad, A. E. Hassanien, and K. Baba, Eds. Berlin, Heidelberg: Springer Berlin Heidelberg, 2013, pp. 143-152.

[15] A. Tharwat, T. Gaber, A. E. Hassanien, H. A. Hassanien, and M. F. Tolba, "Cattle identification using muzzle print images based on texture features approach," in Proceedings of the Fifth International Conference on Innovations in Bio-Inspired Computing and Applications IBICA 2014, P. Komer, A. Abraham, and V. Sn¨ a´sel, Eds. Cham: ¡Springer International Publishing, 2014, pp. 217-227.

[16] B. Barry, U. Gonzales-Barron, K. Mcdonnell, F. Butler, and S. Ward, "Using muzzle pattern recognition as a biometric approach for cattle identification," Transactions of the ASABE, vol. 50, pp. 50; 1073- 1080, 05 2007.

[17] A. I. Awad, H. M. Zawbaa, H. A. Mahmoud, E. H. H. A. Nabi, R. H. Fayed, and A. E. Hassanien, "A robust cattle identification scheme using muzzle print images," in 2013 Federated Conference on Computer Science and Information Systems, Sep. 2013, p. 529-534.

[18] A. Tharwat, T. Gaber, and A. E. Hassanien, "Cattle identification based on muzzle images using gabor features and svm classifier," in Advanced Machine Learning Technologies and Applications, A. E. Hassanien, M. F. Tolba, and A. Taher Azar, Eds. Cham: Springer International Publishing, 2014, pp. 236-247.

[19] T. G. Tharwat, Alaa and A. E. Hassanien, "Two biometric approaches for cattle identification based on features and classifiers fusion," International Journal of Image Mining, vol. 1, no. 4, p. 342-365, 2015.

[20] S. Kumar and S. Singh, "Biometric recognition for pet animals," Journal of Software Engineering and Applications, vol. 7, p. 470-482, 2014.

[21] M. A. Fischler and R. C. Bolles, "Random sample consensus: A paradigm for model fitting with applications to image analysis and automated cartography," Commun. ACM, vol. 24, no. 6, p. 381-395, Jun. 1981. [Online]. Available: http://doi.acm.org/10.1145/358669.358692

[22] C. Y. Han, S.-H. Kwon, H. I. Choi, H. P. Lee, Sung Jin Moon, and N.-S. Wee, "Distributed iris recognition system," 2019, IEEE Transactions on Information Forensic and Security (submitted).

Claims (9)

In the object recognition identification method,
_{RANSAC ( Random} Sample Finding an affine transformation (T _f ) for the first ROI by performing a Consensus alignment algorithm - The ROI image obtained by applying the first ROI (v ₁ ) to the first object image (I _A ) is R _{A ,1} , and an ROI image obtained by applying the first ROI (v ₁ ) to the second object image (I _B ) is represented by R _B,1- ;
Corresponding to performing the RANSAC alignment algorithm with a second ROI (v ₂ ) having a larger size than the first ROI among a plurality of ROIs using an affine transformation (T _f ) for the first ROI without performing the RANSAC alignment algorithm Finding an affine transformation (T _f,2 )
The second object image (I _B ) is obtained by using four vertices of the ROI image (R _{A ,2} ) having the size of the second ROI (v ₂ ) for the affine transformation (T f,2 ) and the first object image (I _{A )} . ) generating an ROI image (R ^* _B,2 ) having a size of a second ROI (v ₂ ); and
The hamming distance of each template generated from the ROI image (R _A,2 ) for the first object image (I _A ) and the ROI image (R ^* _B,2 ) for the second object image (I _B ). ) step of calculating
including,
Corresponding to performing the RANSAC alignment algorithm with a second ROI (v ₂ ) having a larger size than the first ROI among a plurality of ROIs using an affine transformation (T _f ) for the first ROI without performing the RANSAC alignment algorithm The step of finding an affine transformation (T _f,2 ) that
Calculate the affine transformation (T _f,2 ) using the following formula,

when,

ego,
here

Represents the horizontal and vertical ratio of the first ROI size and the second ROI size
Object recognition identification method. According to claim 1,
In order to reduce the time required for recognition using the pattern of the object image, the RANSAC alignment algorithm is performed with the first ROI (v ₁ ) having the smallest size among the plurality of ROIs to obtain an affine transformation (T _f ) for the first ROI and RANSAC with a second ROI (v ₂ ) having a larger size than the first ROI among a plurality of ROIs by using an affine transformation (T _f ) for the first ROI to increase the accuracy of recognition using the pattern of the object image. Calculate the Hamming distance with the generated template by finding an affine transformation (T _f,2 ) corresponding to that performed by the sorting algorithm,
In order to increase the accuracy of the Hamming distance calculation of the template, the RANSAC alignment algorithm is used to find the optimal affine transformation, and the change in coordinate information of the ROI of each of the first object image (I _A ) and the second object image (I _B ) and the ROI Using techniques to exclude misalignment of images
Object recognition identification method. In the object recognition identification system,
_{RANSAC (} Random _Sample Consensus) RANSAC alignment algorithm performer for finding an affine transformation (T _f ) for the first ROI by performing the alignment algorithm -ROI to which the first ROI (v ₁ ) is applied to the first object image (I _A ) The image is denoted by R _A,1 , and the ROI image obtained by applying the first ROI (v ₁ ) to the second object image (I _B ) is denoted by R _B,1- ;
Corresponding to performing the RANSAC alignment algorithm with a second ROI (v ₂ ) having a larger size than the first ROI among a plurality of ROIs using an affine transformation (T _f ) for the first ROI without performing the RANSAC alignment algorithm an affine transformation performer to find an affine transformation (T _f,2 ) to do;
The second object image (I _B ) is obtained by using four vertices of the ROI image (R _{A ,2} ) having the size of the second ROI (v ₂ ) for the affine transformation (T f,2 ) and the first object image (I _{A )} . ), an ROI area generator generating an ROI image (R ^* _B,2 ) having a size of a second ROI (v ₂ ); and
The hamming distance of each template generated from the ROI image (R _A,2 ) for the first object image (I _A ) and the ROI image (R ^* _B,2 ) for the second object image (I _B ). ) Hamming distance calculation unit that calculates
including,
Affine transformation performing unit,
Calculate the affine transformation (T _f,2 ) using the following formula,

when,

ego,
here

Represents the horizontal and vertical ratio of the first ROI size and the second ROI size
Object Recognition Identification System. According to claim 3,
In order to reduce the time required for recognition using the pattern of the object image, the RANSAC alignment algorithm is performed with the first ROI (v ₁ ) having the smallest size among the plurality of ROIs to obtain an affine transformation (T _f ) for the first ROI and RANSAC with a second ROI (v ₂ ) having a larger size than the first ROI among a plurality of ROIs by using an affine transformation (T _f ) for the first ROI to increase the accuracy of recognition using the pattern of the object image. Calculate the Hamming distance with the generated template by finding an affine transformation (T _f,2 ) corresponding to that performed by the sorting algorithm,
In order to increase the accuracy of the Hamming distance calculation of the template, the RANSAC alignment algorithm is used to find the optimal affine transformation, and the change in coordinate information of the ROI of each of the first object image (I _A ) and the second object image (I _B ) and the ROI Using techniques to exclude misalignment of images
Object Recognition Identification System. A program stored in a computer readable storage medium for executing an object recognition identification method performed by an object recognition identification system,
_{RANSAC (} Random _Sample Finding an affine transformation (T _f ) for the first ROI by performing a Consensus alignment algorithm - The ROI image obtained by applying the first ROI (v ₁ ) to the first object image (I _A ) is R _{A ,1} , and an ROI image obtained by applying the first ROI (v ₁ ) to the second object image (I _B ) is represented by R _B,1- ;
Corresponding to performing the RANSAC alignment algorithm with a second ROI (v ₂ ) having a larger size than the first ROI among a plurality of ROIs using an affine transformation (T _f ) for the first ROI without performing the RANSAC alignment algorithm Finding an affine transformation (T _f,2 )
The second object image (I _B ) is obtained by using four vertices of the ROI image (R _{A ,2} ) having the size of the second ROI (v ₂ ) for the affine transformation (T f,2 ) and the first object image (I _{A )} . ) generating an ROI image (R ^* _B,2 ) having a size of a second ROI (v ₂ ); and
The hamming distance of each template generated from the ROI image (R _A,2 ) for the first object image (I _A ) and the ROI image (R ^* _B,2 ) for the second object image (I _B ). ) step of calculating
including,
Corresponding to performing the RANSAC alignment algorithm with a second ROI (v ₂ ) having a larger size than the first ROI among a plurality of ROIs using an affine transformation (T _f ) for the first ROI without performing the RANSAC alignment algorithm The step of finding an affine transformation (T _f,2 ) that
Calculate the affine transformation (T _f,2 ) using the following formula,

when,

ego,
here

Represents the horizontal and vertical ratio of the first ROI size and the second ROI size,
In order to reduce the time required for recognition using the pattern of the object image, the RANSAC alignment algorithm is performed with the first ROI (v ₁ ) having the smallest size among the plurality of ROIs to obtain an affine transformation (T _f ) for the first ROI and RANSAC with a second ROI (v ₂ ) having a larger size than the first ROI among a plurality of ROIs by using an affine transformation (T _f ) for the first ROI to increase the accuracy of recognition using the pattern of the object image. Calculate the Hamming distance with the generated template by finding an affine transformation (T _f,2 ) corresponding to that performed by the sorting algorithm,
In order to increase the accuracy of the Hamming distance calculation of the template, the RANSAC alignment algorithm is used to find the optimal affine transformation, and the change in coordinate information of the ROI of each of the first object image (I _A ) and the second object image (I _B ) and the ROI Using techniques to exclude misalignment of images
A program stored on a computer readable storage medium. delete delete delete delete

Images