Research and Application of Image Stitching Technology Based on

SURF Feature Points

Qihang Zheng and Dong Huang

Shantou Polytechnic, Shantou, Guangdong Province,515041, China

Keywords: Image Theory, SURF Feature Extraction Algorithm, Image Acquisition, Image Preprocessing, Image

Registration, Image Fusion.

Abstract: The role of image stitching technology in the field of image research is very important, but there is a problem

of inaccurate image fusion. The classic SIFT detection algorithm cannot solve the image stitching problem in

the field of image research, and the evaluation is unreasonable. Therefore, a SURF feature extraction

algorithm is proposed for fast image stitching analysis. Firstly, the image theory is used to evaluate the pattern,

and the indicators are divided according to the image stitching requirements to reduce it Distractors in image

stitching. Then, image theory optimizes the image stitching technology to form an image stitching scheme

and performs the image stitching results Comprehensive analysis. MATLAB simulation shows that under

certain evaluation criteria, the acquisition, registration and fusion speed of image stitching by SURF feature

extraction algorithm are improved All are better than the SIFT detection algorithm.

1 INTRODUCTION

Image stitching is widely used in the field of

computer vision, and its main purpose is to stitch

multiple pictures into an infinitely extended large

picture (Bouchekara, Sadiq, et al. 2023). The SURF

(Speeded Up Robust Feature) feature is a feature

point used for image matching and object recognition.

This paper will focus on the influence of SURF

features on image stitching (Bui, Nguyen, et al.

2023), analyze the advantages and disadvantages of

SURF features in image stitching, and propose

improvement measures (Cai, Du, et al. 2023).

1.1 The Rationale for the SURF

Feature

A SURF feature is a feature descriptor that describes

local features in an image. The SURF feature point

detection algorithm mainly includes three steps:

Scale spatial extreme value detection: at different

scales and directions, the Hesen matrix is used to

detect local extreme points (Chatterjee, and Issac,

2023).

Key point positioning: precise positioning of the

position of extreme value points, and removal of low-

contrast extreme value points and edge response

points (Chen, Wang, et al. 2023).

Feature point description: The SURF descriptor

based on the gradient direction and local scale

information of the entire image is used to describe the

feature point (Chen, Zhou, et al. 2023).

1.2 Advantages and Disadvantages of

SURF Features in Image Stitching

1.2.1 Pros

(1) SURF feature algorithm is a fast feature point

detection and matching algorithm, which has the

characteristics of fast calculation speed (Ciortea,

Chen, et al. 2023).

(2) The SURF feature algorithm can extract and

match the features of images at different scales, so it

can adapt to image stitching of different sizes and

angles (Cong, Wang, et al. 2023).

(3) The SURF feature algorithm has good

invariance for rotation, scaling and translation, and

can match and stitch pictures more accurately (Cui,

Shao, et al. 2023).

150

Zheng, Q. and Huang, D.

Research and Application of Image Stitching Technology Based on SURF Feature Points.

DOI: 10.5220/0013537200004664

Paper published under CC license (CC BY-NC-ND 4.0)

In Proceedings of the 3rd International Conference on Futuristic Technology (INCOFT 2025) - Volume 1, pages 150-156

ISBN: 978-989-758-763-4

1.2.2 Cons

(1) The SURF feature algorithm is sensitive to

some noise and changes in special cases, such as

lighting changes and occlusion.

(2) The number of feature points extracted by the

SURF feature algorithm is large, resulting in

mismatching and duplicate matching in the matching

process (Dai, Degenhardt, et al. 2023).

(3) The SURF feature algorithm is not suitable for

some occasions that require high-precision matching,

such as medical images, three-dimensional vision,

precise positioning, etc (Deng, Song, et al. 2023)].

1.3 Application of SURF Feature in

Image Stitching

SURF features are widely used in image stitching,

mainly used in feature point detection, feature

matching and image transformation. In the process of

image stitching, SURF features can find similar

feature points between multiple images to achieve

stitching of different pictures (Gao, Huang, et al.

2023).

1.3.1 Feature Point Detection

SURF features can detect multi-scale and multi-

directional feature points on images, and can detect

feature points in some detail areas (Hong, and Kim,

2023). These feature points are the basic units in the

image stitching process and are the key to subsequent

matching and transformation.

1.3.2 Feature Matching

SURF features can match feature points between

different images, find similar feature points and

establish correspondence. In the process of feature

point matching, the matching of feature points can be

achieved by calculating the distance of the SURF

descriptor (Jiang, Sun, et al. 2023).

1.3.3 Image Transformation

The SURF feature can transform images by

calculating the transformation matrix between two

images. Among them, the transformation matrix can

be estimated by the RANSAC algorithm to achieve

high-precision image transformation.

1.4 Optimization Measures of SURF

Features in Image Stitching

1.4.1 Noise Rejection

The SURF feature is sensitive to environmental

changes such as lighting, and some noise feature

points are introduced. Therefore, before surf feature

extraction, the image should be preprocessed and

denoised, and non-valid feature points should be

removed (Kang, Wu, et al. 2023).

1.4.2 Feature Point Filtering

The number of feature points extracted by SURF

features is large, so a suitable feature point screening

method should be adopted to remove some duplicate

and useless feature points. In general, the screening

method based on the SURF descriptor distance can be

used to remove the feature points with too large

distance (Kim, Lee, et al. 2023).

1.4.3 Fine Matching

SURF feature matching has certain mismatches and

missing matches, so some fine matching methods

need to be adopted to improve the accuracy and

robustness of matching. In general, techniques such

as optical flow estimation and panoramic stitching

can be used for fine matching and optimization.

1.4.4 Image Transformation

The SURF feature has accuracy errors in the image

transformation process, so some high-precision

image transformation methods need to be adopted to

improve the quality and accuracy of image stitching.

In general, high-precision image transformation can

be achieved by using technologies such as corner

point detection and image registration.

This paper mainly discusses the application and

influence analysis of SURF features in image

stitching, which has the advantages of fast calculation

speed and strong scale invariance, while its

shortcomings such as strong sensitivity, large number

of feature points, and unsuitability for high-precision

occasions also need to be improved. Through the

optimization measures of SURF features, such as

noise rejection, feature point screening, fine matching

and image transformation, the quality and accuracy of

image stitching can be improved, which has a wide

application prospect in the field of computer vision.

Image fusion is one of the important contents of

image stitching technology, which is of great

significance for image research. However, in the

Research and Application of Image Stitching Technology Based on SURF Feature Points

151

process of image stitching, the image stitching

scheme has the problem of inaccurate acquisition and

registration, which brings certain image distortion

and seam problems to image stitching. Some scholars

believe that the application of SURF feature

extraction algorithm to the analysis of image research

can effectively analyze the image stitching scheme

and provide corresponding support for image

stitching. On this basis, this paper proposes a SURF

feature extraction algorithm to optimize the image

stitching scheme and verify the effectiveness of the

model.

2 RELATED CONCEPTS

2.1 Mathematical Description of the

SURF Feature Extraction

Algorithm

The SURF feature extraction algorithm uses image

theory to optimize the image stitching scheme is

and finds the unqualified values in image fusion

according to the indicators in image stitching is

and integrates the image stitching scheme. Finally is

(w )

iij

acrd q⋅

, the feasibility of image fusion is

judged, and the calculation is shown in Equation (1).

() ) w(

iij ij ij X

aqctrd ex ec ps

⋅=≥

)

Among them, the judgment of outliers is shown in

Equation (2).

()( 4) ( )

ij x

epxsec gradd

=÷



(2)

The SURF feature extraction algorithm combines

the advantages of image theory and uses image fusion

for quantification, which can improve the quality of

image stitching.

Suppose I. The image stitching requirement is

, the image stitching scheme is

, the satisfaction of

the image stitching scheme is

set

, and the image

stitching scheme judgment function is

(0)

STe−≈

,As shown in Equation (3).

( ) sin

ii i

STe XY k

−

−= ÷ ⋅



)

2.2 Selection of Entrepreneurial

Quality Programs

Hypothesis II The image fusion function is

and the weight coefficient is

()

, then the image

stitching requires unqualified image fusion as shown

in Equation (4).

()=

qws e m

−

⋅⋅−

)

Based on hypotheses I and II, a comprehensive

function of entrepreneurship education can be

obtained, and the result is shown in Equation (5).

() () ( )

iijX

S exsecTe qws p

−+⋅≤

)

In order to improve the effectiveness of quality

assessment, all data needs to be standardized and the

results are shown in Equation (6).



() () ( )

SgraTe qw dsx−+⋅↔



)

2.3 Analysis of Image Stitching Scheme

Before the SURF feature extraction algorithm, the

image stitching scheme should be analyzed in

multiple dimensions, and the image stitching

requirements should be mapped to the image fusion

library to eliminate the unqualified image stitching

scheme is

()

. According to Equation (6), the

anomaly evaluation scheme can be proposed, and the

results are shown in Equation (7).



)

(()

()

STe q

gr d

−+⋅



（7）

Among them,



() ()

( )

Te qws

ad xg

−+⋅

≤



it is

stated that the scheme needs to be proposed,

INCOFT 2025 - International Conference on Futuristic Technology

152

otherwise the scheme integration is required,

()

RMS

and the result is shown in Equation (8).



() min[ () ()]

iji

xSTReqwMS s=−+⋅

(8)

The image fusion is comprehensively analyzed,

and the threshold and index weight of the image

stitching scheme are set to ensure the accuracy of the

SURF feature extraction algorithm. Image fusion is a

systematic test image stitching scheme that requires

innovative analysis. If the image fusion is in a

nonnormal distribution, the image stitching scheme

()

unno x

will be affected, reducing the accuracy of

the overall image stitching, and the calculation result

is as shown in the formula (9

()

accur x

) shown.



min[ ( ) ( )]

( ) 100%

() ()

STe qws

accur x

STe qws

−+⋅

=×

−+⋅

(9)

The survey image stitching scheme shows that the

entrepreneurial quality scheme presents a multi-

dimensional distribution, which is in line with the

objective facts. The image fusion is not directional,

indicating that the entrepreneurial quality scheme has

strong randomness, so it is regarded as a high

analytical study. If the random function of image

fusion is

()

randon x

, then the calculation of

equation (9) can be expressed as formula (10).



min[ ( ) ( )]

( ) 100% (

() ()

STe qws

accur x randon x

STe qws

−+⋅

=×+

−+⋅

(10)

2.4 Analysis of Image Stitching Scheme

Before the SURF feature extraction algorithm, the

image stitching scheme should be analyzed in

multiple dimensions, and the image stitching

requirements should be mapped to the image research

field library, and the unqualified image stitching

scheme should be eliminated. First, the image

research field is comprehensively analyzed, and the

threshold and index weight of the image stitching

scheme are set to ensure the accuracy of the SURF

feature extraction algorithm. The field of image

research is to systematically test image stitching

schemes, which requires innovative analysis. If the

image research field is in a nonnormal distribution,

the image stitching scheme will be affected, reducing

the accuracy of the overall image stitching. In order

to improve the accuracy of the SURF feature

extraction algorithm and improve the level of image

stitching, the image stitching scheme should be

selected, and the specific scheme selection is shown

in Figure I.

Image

acquisition

Image

preprocessin

Image fusionImage registration

Image

distortion

Establish a

transformati

on model

Search speed

SURF Feature

Extraction

Algorithm

Hessian Matrix

Figure 1: Results of selection of image fusion scheme

The survey image stitching scheme shows that the

image stitching scheme presents a multi-dimensional

distribution, which is in line with the objective facts.

The image research field is not directional, indicating

that the image stitching scheme has strong

randomness, so it is regarded as a high analytical

study. The image research field meets the normal

requirements, mainly image theory adjusts the image

research field, removes duplicate and irrelevant

schemes, and supplements the default scheme to

make the entire image stitching The dynamic

correlation of scenarios is strong.

3 OPTIMIZATION STRATEGIES

IN THE FIELD OF IMAGE

RESEARCH

The SURF feature extraction algorithm adopts a

random optimization strategy for image research and

adjusts image parameters to realize the scheme

optimization in image research field. The SURF

feature extraction algorithm divides the image

research field into different image stitching levels,

and randomly selects different schemes. In the

iterative process, the image stitching scheme with

different image stitching levels is optimized and

analyzed. After the optimization analysis is

completed, compare the image stitching levels of

different schemes to record the best image research

area.

Research and Application of Image Stitching Technology Based on SURF Feature Points

153

4 PRACTICAL EXAMPLES IN

THE FIELD OF IMAGE

RESEARCH

4.1 Introduction to Image Stitching

In order to facilitate image stitching, this paper takes

the image research field in complex cases as the

research object, with 12 paths and a test time of 12h,

and the image stitching in the specific image research

field The scheme is shown in Table 1.

The image stitching process in Table 1. is shown

in Figure 2.

Table 1: Technical requirements for image stitching

Scope of

lication

grade Innovative

effect

Image

fusion

Image

uisition

I 92.26 90.53

II 88.56 90.77

Image

istration

I 90.82 89.34

II 87.96 88.07

Image

rocessin

I 89.12 88.78

II 92.99 87.58

Figure 2: The analysis process in the field of image research

Compared with the SIFT detection algorithm, the

image stitching scheme of the SURF feature

extraction algorithm is closer to the actual image

stitching requirements. In terms of rationality and

fluctuation amplitude in the field of image research,

the SURF feature extraction algorithm is better than

the SIFT detection method. It can be seen from the

changes of the image stitching scheme in Figure II

that the SURF feature extraction algorithm has better

stability and faster judgment speed. Therefore, the

image stitching scheme of SURF feature extraction

algorithm has better effect on image registration and

image fusion.

4.2 Situation in the Field of Image

Research

Image stitching schemes in the field of image

research include image acquisition, image

preprocessing, image registration and image fusion.

After the preselection of SURF feature extraction

algorithm, the image stitching scheme of the

preliminary image research field is obtained, and the

image research field is obtained The feasibility of

image stitching scheme is analyzed. In order to more

accurately verify the innovative effect of image

research field, the image stitching scheme of different

image stitching levels is selected as shown in Table 2.

Table 2: The overall picture of the image fusion scheme

cate

Precision Anal

sis rate

Image

acquisition

88.43 86.33

Image

registration

93.44 91.34

Ima

e fusion 93.11 89.73

mean 95.54 88.96

90.59 90.24

P=2. 849

4.3 Image Fusion and Stabilization for

Image Stitching

In order to verify the accuracy of the SURF feature

extraction algorithm, the image stitching scheme is

compared with the SIFT detection algorithm, and the

image stitching scheme is shown in Figure 3.

Figure 3: Image fusion of different algorithms

It can be seen from Figure 3 that the image fusion

of the SURF feature extraction algorithm is higher

than that of the SIFT detection algorithm, but the

error rate is lower, indicating that the image stitching

of the SURF feature extraction algorithm is indicated

It is relatively stable, while the image stitching of the

SIFT detection algorithm is uneven. The average

INCOFT 2025 - International Conference on Futuristic Technology

154

image stitching scheme of the above three algorithms

is shown in Table 3.

Table 3: Comparison of image stitching accuracy of

different methods

Algorithm Image

fusion

Magnitude

of change

Error

SURF feature

extraction

algorith

87.72 94.17 3.42

SIFT detection

algorith

85.76 86.86 9.49

P 39.536 34.410 30.025

By Table 3 it can be seen that the SIFT detection

algorithm has shortcomings in image fusion and

image distortion in the field of image research, and

the image research field has undergone great changes

, the error rate is high. The image fusion degree of the

SURF feature extraction algorithm is high, which is

better than the SIFT detection algorithm. At the same

time, the image fusion of the SURF feature extraction

algorithm is greater than 87%, and the accuracy does

not change significantly. In order to further verify the

superiority of the SURF feature extraction algorithm.

In order to further verify the effectiveness of the

proposed method, the SURF feature extraction

algorithm is generally analyzed by different methods.

5 CONCLUSIONS

Aiming at the problem of unsatisfactory image fusion

in the field of image research, this paper proposes a

SURF feature extraction algorithm, and combines

image theory to optimize the image research field. At

the same time, the image stitching innovation and

threshold innovation are analyzed in depth to

construct an image collection. The research shows

that the SURF feature extraction algorithm can

improve the distortion and stability of the image

research field, which can improve the image research

field Optimized image stitching technology.

However, in the process of SURF feature extraction

algorithm, too much attention is paid to the analysis

of image stitching, resulting in irrationality in the

selection of image stitching indicators.

ACKNOWLEDGEMENTS

2019 Science and Technology Special Fund Project

of Guangdong Province (a research on photovoltaic

panel image splicing technology based on improved

SURF-BRISK);

2018 Key scientific research platform and

scientific research projects of General universities in

Guangdong Province (research on photovoltaic

module image recognition technology based on

improved ASIFT algorithm).

REFERENCES

Bouchekara, H., Sadiq, B. O., Zakariyya, S., Sha'aban, Y.

A., Shahriar, M. S., & Isah, M. M.(2023) SIFT-CNN

Pipeline in Livestock Management: A Drone Image

Stitching Algorithm. Drones, 7(1).

Bui, M., Nguyen, T., Ninh, H., Nguyen, T., & Tran, T.

H.(2023) Global suppression heuristic: fast GraphCut

in GPU for image stitching. Signal Image and Video

Processing, 17(6): 2671-2678.

Cai, W. X., Du, S. L., & Yang, W. K.(2023) UAV image

stitching by estimating orthograph with RGB cameras.

Journal of Visual Communication and Image

Representation, 94.

Chatterjee, S., & Issac, K. K.(2023) Viewpoint planning

and 3D image stitching algorithms for inspection of

panels. Ndt & E International, 137.

Chen, D., Wang, Y. Q., & Zhang, R. Z.(2023) Determining

the size of the overlapping area for image stitching in

dark-field detection. Journal of Modern Optics, 70(3):

181-188.

Chen, G. L., Zhou, H., Huang, G., Song, G. H., & Zhang, J.

J.(2023) A deep image segmentation-based method for

stitching ancient-book images without an overlapping

region. Iet Image Processing.

Ciortea, L. I., Chen, D. Q., & Xiao, P. R. Y.(2023) Skin

Capacitive Image Stitching and Occlusion

Measurements. Cosmetics, 10(1).

Cong, Y. Z., Wang, Y., Hou, W. J., & Pang, W.(2023)

Feature Correspondences Increase and Hybrid Terms

Optimization Warp for Image Stitching. Entropy, 25(1).

Cui, Z. Y., Tang, R. X., & Wei, J. B.(2023) UAV Image

Stitching With Transformer and Small Grid

Reformation. Ieee Geoscience and Remote Sensing

Letters, 20.

Dai, G. L., Degenhardt, J., Hu, X. K., Wolff, H., Tutsch, R.,

& Manske, E.(2023) A feasibility study towards

traceable calibration of size and form of microspheres

by stitching AFM images using ICP point-to-plane

algorithm. Measurement Science and Technology,

34(5).

Deng, Z. P., Song, S. Z., Han, S. Y., Liu, Z. Q., Wang, Q.,

& Jiang, L. Y.(2023) Geological Borehole Video Image

Stitching Method Based on Local Homography Matrix

Offset Optimization. Sensors, 23(2).

Gao, H., Huang, Z. Q., Yang, H. P., Zhang, X. B., & Cen,

C.(2023) Research on Improved Multi-Channel Image

Stitching Technology Based on Fast Algorithms.

Electronics, 12(7).

Research and Application of Image Stitching Technology Based on SURF Feature Points

155

Hong, S. H., & Kim, J.(2023) Three-dimensional visual

mapping of underwater ship hull surface using image

stitching geometry. Ocean Engineering, 269.

Jiang, H. H., Sun, X. M., Fang, W. T., Fu, L. S., Li, R.,

Cheein, F. A., & Majeed, Y.(2023) Thin wire

segmentation and reconstruction based on a novel

image overlap-partitioning and stitching algorithm in

apple fruiting wall architecture for robotic picking.

Computers and Electronics in Agriculture, 209.

Kang, Y., Wu, R., Wu, S., Li, P. Z., Li, Q. P., Cao, K., Tan,

T. T., Li, Y. R., & Zha, G. Q.(2023) A novel multi-view

X-ray digital imaging stitching algorithm. Journal of X-

Ray Science and Technology, 31(1): 153-166.

Kim, H., Lee, H., Ahn, S., Jung, W. K., & Ahn, S. H.(2023)

Broken stitch detection system for industrial sewing

machines using HSV color space and image processing

techniques. Journal of Computational Design and

Engineering, 10(4): 1602-1614.

INCOFT 2025 - International Conference on Futuristic Technology

156