iBlurDetect: Image Blur Detection Techniques Assessment and

Evaluation Study

Roxanne A. Pagaduan, Ma. Christina R. Aragon and Ruji P. Medina

Technological Institute of the Philippines, Information Technology Department Quezon City, Philippines

Keywords: computer vision, image processing, blur detection, blur measure operators, blur identification.

Abstract: The quality of images is essential in computer vision, image processing, and other related fields. Image

restoration is one of the categories in image processing, where the quality of an image plays a vital role in

the process. Blur detection is a pre-processing stage in image restoration. Using different blur detection

techniques, the quality of an image can identify if blurry or not. This study aims to provide a comparative

performance of the available state-of-the-art blur measure operators or blur detection techniques. Python 6.3

was used for testing and evaluating the blur detection techniques. Providing the confusion matrix, precision,

recall, f-measure, accuracy, and execution time were used to compare blur detection techniques. In testing,

the Gaussian kernel and threshold value were set to measure the performance of each technique. Provided

on the evaluation results, in terms of accuracy rate, HWT leads the best result. Based on the computed

scores, FFT got the highest precision score, while LAP got the highest recall score, and HWT got the

highest f-measure score. In terms of the execution time, MLAP performs the fastest processing time among

them all. Likewise, results of this study can use as resources before performing the image restoration.

1 INTRODUCTION

Image perform a significant role in technology as

well as in the research domain. These images can

applied in computer vision and image processing

such as image representation (Cruz-Roa et al, 2013),

object recognition and matching (Yadav and Singh,

2016), 3D scene reconstruction (Yang, Zhou and

Bai, 2013; Fang, Tao and Jia-Lin, 2017), and motion

tracking (Chen and Liu, 2018; Ancheta et al, 2018)

to name a few. Images are produced to record or

display important information.

The quality of an image contributes to the

success of determining certain information that can

used in different fields of research. In feature

detection, for example, the recognition rate depends

on the image quality(Dharavath et al, 2014).

Image quality can be degraded due to distortion

during acquisition and processing. Some common

factors may affect the quality of an image are

contrast, noise, artifacts, and blurring (Su, Lu, and

Tan, 2018). To address this issue, image recognition

techniques are continuously being performed and

improved (Sprawls, 1995).

Image blurring is a form of bandwidth reduction

on an ideal image caused by an imperfect image

construction procedure(Bovik and Gibson, 2000).

Blur is the typical image downfall problem when

capturing the photos. Image blur occurs in most

cases of image deterioration resulting from

defocusing or handshaking (Yang, Lin and Chuang,

2017).The reasons behind the output of blurry

images are camera shaking due to dynamic

movement of the lens during the process of capture,

object movement, out-of-focus due to camera lens

could not set a proper angle and focus, out-of-focus,

and low-quality cameras (Dharavath et al, 2014) and

(Su, Lu, and Tan, 2018).

Since image blur is a common issue, and it is, at

times difficult to remove in many situations. Due to

this problem, many researchers are working on

finding the best way to de-blur the image and restore

the blurred image (Bansal et al, 2017; Huang et al,

2019). Study (Bansal et al, 2017) stated that to

maintain the quality of the image it is vital to detect

and eliminate the blur from images.

Image processing techniques can use in the

modificationof digital data for refining the image

qualities with the aid of a computer system (Bansal

et al, 2017).

286

Pagaduan, R., R. Aragon, M. and Medina, R.

iBlurDetect: Image Blur Detection Techniques Assessment and Evaluation Study.

DOI: 10.5220/0010307700003051

In Proceedings of the International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies (CESIT 2020), pages 286-291

ISBN: 978-989-758-501-2

Figure 1: Flowchart of Image Blur Detection Framework.

Figure 1 shows the flowchart of image blur

detection framework. The study of (Koik and

Ibrahim, 2014), narrow down the researches

available in public domain about blur images into

three major processes: a) image blur detection, the

initial process in improving the quality of the image

that suffers from blur, b) blur classification, the

second process of the research related to blur

images. The goal of this process is to classify the

blur areas according to their characteristics or types,

and 3) image restoration, the third process, perform

deblurring process based on their characteristics.

This paper concentrates only on the stage of image

blur detection that consider the blur/sharp estimation

and measure which is enclosed in red-dotted lines in

figure 1.

This paper focuses on the different blur detection

techniques and aims to compare the performance of

each one in terms of accuracy rate and execution

time. Also, compare and analyze the existing

techniques in identifying if the input image is

blurred or sharp to achieve the best possible results.

The project’s long term goal aims to maintain a

comparable number of extracted feature points with

a sharp image and to increase the number of

correctly matched feature points of inputted blur

image. The project groundwork lays on computer

vision, and image processing notably features point

detector. Blur detection techniques are useful in

image blur detection because it is used as the

preliminary process to detect specific regions that

need for image restoration or deblurring process. As

the primary step towards the goal of the project, we

conducted a review and analysis of the different blur

measure operators or state-of-the-art of image blur

detector techniques.

This paper has been organized as follows.

Section 2. Blur detection techniques. Section 3.

Experimental methodology. Section 4. Deals with

the results and discussions and the last Section is the

conclusion of the study.

2 BLUR DETECTION

TECHNIQUES

Blur detection is one of the interesting research areas

in computer vision and image processing like in

(Cruz-Roa et al, 2013; Yang, Zhou and Bai, 2013;

Fang, Tao and Jia-Lin, 2017), and (Ancheta et al,

2018). Most of the captured images usually contain

two types of regions: blurred and sharp. Blur can be

categorized into two types: a) defocus blur or also

known as out-of-focus blur, which is caused by the

visual imaging system and b) motion blur or also

known as camera-shake blur, which is caused by the

relative motion between camera and scene objects

(Ali and Mahmood, 2018).

Study (Pertuz et al, 2013), reviews 36 different

techniques or focus measure operators to compute

the blurriness metric of an image, some of them are

simple and straight forward using just grayscale

pixel intensity statistics, other are more advanced

and feature-based that evaluate the local binary

patterns of an image.

A total of 32 different blur measure operators

was reviews for single image blur segmentation in

(Ali and Mahmood, 2018). Some number of measure

operators reviews included are originally developed

for autofocus and shape from focus (SFF) techniques

by (Abdel-Qader et al, 2003).

While in (Bansal et al, 2017), reviews 3 different

blur detection techniques such as laplacian operator,

fast fourier transform, and haar wavelet transform.

In their study, Laplacian operator was selected for

testing and successfully identify if the image is

blurred or sharp.

In other literature, Tenengrad technique is used

to extract the degradation degree of each target part

in the image. Tenengrad technique was used in

(Gao, Han, and Cheng, 2018) as operator used to

evaluate the iris image’s definition.

iBlurDetect: Image Blur Detection Techniques Assessment and Evaluation Study

287

With the help of the related research papers

available in public domain, conducting about blur

detection techniques, we consider some related

image blur detection techniques in our study and test

the performance of each technique.

There are many image blur detection techniques

to detect whether an image is blurred or sharp. Some

of them are:

2.1 Fast Fourier Transform (FFT)

In Fourier transform, this method calculates the

frequencies in the image at different points and

based on the set level of frequencies it decides

whether the image is blurred or sharp. When there is

a low amount of frequency based on the set level of

frequencies then it declares that the image is blurred

otherwise, if the computed frequencies is high then

the image is sharp. The decision that will be the

value of low and high frequencies is based on the

programmer. (Pertuz et al, 2013).

2.2 HaarWavelet Transform (HWT)

In this method, the images are split into NxN by

iterating on each tile of the 2Dimensional HWT, and

grouping diagonally, vertically, or horizontally

connected tiles into clusters containing images are

then declared blurred (Tong, Li, Zhang, and Zhang,

2004).

2.3 Laplacian Operator (LAP)

This method is implemented to discover edges in a

picture. It is additionally a derivative operator but

the basic contrast between different operators like

Sobel, Kirsch and Laplacian operator is that all other

derivatives are first order derivative mask. Laplacian

operator is further separated into two classification

which are the positive Laplacian operator and

negative Laplacian operator.

2.4 Modified Laplacian (MLAP)

The modified laplacian is developed to compute

local measures of the quality of image focus. By

getting the absolute values of the second derivatives

in x and y directions (Pech-Pacheco et al, 2000).

2.5 Tenengrad (TEN)

The well-celebrated focus measure based on image

gradients obtained by the convolving the image with

sobel operator that can also be considered as blur

measure operator (Pech-Pacheco et al, 2000).

Figure 2: Flowchart of Image Blur Detection Techniques

Processes.

Figure 2 shows the processes performed by the

different blur detection techniques for testing. The

user should input the value for threshold and

Gaussian kernel for assessing and computing the

score values of the inputted image. Based on the set

threshold and Gaussian kernel, the calculated score

value will be the basis of the image input is blurred

or sharp.

3 EXPERIMENTAL

METHODOLOGY

In this section, we describe the methodology we

followed to perform a comparative analysis of image

blur detection techniques. This study was

programmed and tested in Python 3.6, using

notebook computer, which has Intel Core i7-8750H

CPU @ 2.20GHz and 8.0 GB RAM with the

Windows 10, 64bit operating system.

3.1 Dataset

To quantitatively evaluate the performance of the

different blur detection techniques, we randomly

selected 200 blur and sharp images from the dataset

provided in the study of [23]. The RGB image is 640

x 480 pixels. The blur images may have motion blur,

out-of-focus blur, and synthetic blur.

CESIT 2020 - International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies

288

Selecting proper threshold value totally depends

on the domain. If the selected threshold is too high

or too low then the images would be marked falsely,

for example, if an image is sharp and the threshold is

too high then the image will be marked blurry.

3.2 Evaluation Measures

Table 1 shows the confusion matrix model used to

evaluate the accuracy of the different blur detection

techniques.

Table 1: Confusion Matrix.

Predicted Value

Negative (N) Positive (P)

Actual

Value

Negative

(N)

True Negative

(TN)

False Positive

(FP)

Positive

(P)

False Negative

(FN)

True Positive

(TP)

where:

• Positive (P): Observation is positive (image is

blurred)

• Negative (N): Observation is not positive

(image is not blurred (sharp))

• True Positive (TP): Observation is positive,

and is predicted to be positive (image is

blurred and predicted as blurred)

• False Negative (FN): Observation is positive,

but is predicted as negative (image is blurred

and predicted as sharp)

• True Negative (TN): Observation is negative,

and is predicted to be negative (image is sharp

and is predicted to be sharp)

• False Positive (FP): Observation is negative,

but is predicted positive (image is sharp, but is

predicted as blurred)

This different blur detection techniques are

also measure based on the following criteria:

1) Precision: a measure of relevance between the

retrieved result and the observation. It refers

to the fraction of the detected blurred (sharp)

pixels which are actually blurred (sharp).

Precision, P =













(24)

Where 𝑇



means that the blurred (sharp) pixel

has been correctly detected as blurred (sharp) pixel

and 𝐹



expresses that a pixel has been inaccurately

detected as blurred (sharp) but it was sharp (blurred)

actually.

2) Recall: also called as sensitivity in binary

classification, it is a measure of the ability to

retrieve the relevant results. It depicts the

fraction of the actual blurred (sharp) pixels

which are detected.

Recall, R =













(25)

Where Fnnmeans that a pixel has been

inaccurately detected as sharp (blurred) but it was

blurred (sharp) actually.

3) F-measure: is a measure of a test’s accuracy

and is defined as the weighted harmonic mean

(average) of the precision and recall of the

test.

F-measure, F =2 x

  



(26)

4) Accuracy: the ratio between the number of

blurred (sharp) images correctly classified.

Accuracy, A =





 











 



 



(27)

5) Execution Time: total number of run-time

during the execution of images

These quantitative measures provide an

appropriate tool for analysis and evaluation of

dataset.

4 RESULTS AND DISCUSSIONS

In this section, we discussed the results conducted

during experimentation to be able to analyze the

comparative performance of the different image blur

detection techniques. Figure 3 shows a sample result

of blur image after using blur detection techniques.

While, figure 4 shows a sample result of sharp

image after using blur detection techniques when

evaluated and tested using Python 6.3. The Gaussian

kernel of all techniques was set to three (3) and set

the proper threshold value.

Figure 3: Result of BlurImages using Blur Detection

Techniques; Result of (a) Fast Fourier Transform; (b)

Laplacian Operator; (c) Modified Laplacian;

(d)Tenengrad; and (e) HaarWavelett Transform.

iBlurDetect: Image Blur Detection Techniques Assessment and Evaluation Study

289

Figure 4 Result of Sharp Images using Blur Detection

Techniques; Result of (a) Fast Fourier Transform; (b)

Laplacian Operator; (c) Modified Laplacian;

(d)Tenengrad; and (e) HaarWavelett Transform.

Table 2: This caption has one line so it is centered.

Blur

Detecti

FN TP Accura

cy (%)

Total

Time

(sec)

FFT 10

0 13 87 93.5% 6.2001

LAP 73 2

2 98 85.5% 1.1482

MLAP 95 5 27 73 84% 0.8951

TEN 94 6 6 94 94% 5.6921

HWT 99 1 5 95 97% 6.0370

Table 2 shows the confusion matrix results of the

performance comparison of different blur detection

techniques. Provided the assessment results, in terms

of accuracy rate, HWT leads the best results follows

by TEN, FFT, LAP, and MLAP sequentially. In

terms of execution time, MLAP leads the best results

follows by LAP, TEN, HWT, and FFT sequentially.

Table 3: Comparison of Blur Detection Techniques.

Blur

Detecti

Precisio

n Score

(%)

Recall

Score

(%)

F-

Measure

Score (%)

Total

Time

(sec)

FFT 1.0 0.87 0.93048 6.2001

LAP 0.784 0.98 0.87111 1.1482

MLAP 0.9358 0.73 0.82022 0.8951

TEN 0.94 0.94 0.94 5.6921

HWT 0.9895 0.95 0.96938 6.0370

Table 3 shows the summary results of the

performance comparison of different blur detection

techniques. Provided the assessment results to

measure the scores are the precision score, recall

score, and F-measure score. Also, we considered the

total processing time (execution time) of each

technique. FFT got the highest precision score, while

LAP got the highest recall score, and HWT got the

highest f-measure score. In terms of execution time,

MLAP performs the fastest processing time.

5 CONCLUSIONS

The study aims to conduct comparative analysis

about the different image blur detection techniques.

Based on the results, in terms of accuracy rate, HWT

leads the best result. Based on the computed scores,

FFT got the highest precision score, while LAP got

the highest recall score, and HWT got the highest f-

measure score. In terms of execution time, MLAP

performs the fastest processing time among them all.

The next stage, as part of our long term project

goal, we planned to conduct a comparative analysis

of the different image restoration or deblurring

techniques that can be used in our long term goal.

REFERENCES

Cruz-Roa, A.A., Arevalo Ovalle, J. E., Madabhushi,, A.,

González Osorio, F. A., 2013 A deep learning

architecture for image representation, visual

interpretability and automated basal-cell carcinoma

cancer detection,” in Lecture Notes in Computer

Science (including subseries Lecture Notes in

Artificial Intelligence and Lecture Notes in

Bioinformatics).

Yadav, S., Singh, A., 2016. An image matching and object

recognition system using webcam robot,” in 2016 4th

International Conference on Parallel, Distributed and

Grid Computing, PDGC 2016.

Yang, C., Zhou, F., Bai, X., 2013. 3D Reconstruction

through Measure Based Image Selection, 2013 Ninth

International Conference on Computational

Intelligence and Security, Leshan.

Fang, L., Tao, J., Jia-Lin, G., 2017. Analysis of 3D

Reconstruction of Large-Scale Architectural Scene

Based on Virtual Reality, 2017 International

Conference on Computer Systems, Electronics and

Control (ICCSEC), Dalian.

Chen, G., Liu, Y., 2018. The study on motion message of

rotational motion with eye tracking, 2018 IEEE

International Conference on Applied System Invention

(ICASI), Chiba.

Ancheta, R. A., Reyes, F. C., Caliwag, J. A., Castillo, R.

E., 2018.FEDSecurity: Implementation of computer

vision thru face and eye detection, Int. J. Mach. Learn.

Comput.

Dharavath Amarnath, G., Talukdar, F. A., Laskar, R. H.,

2013. Impact of image preprocessing on face

recognition: A comparative analysis, in International

Conference on Communication and Signal Processing,

ICCSP 2014 - Proceedings.

CESIT 2020 - International Conference on Culture Heritage, Education, Sustainable Tourism, and Innovation Technologies

290

Su, B., Lu, S., Tan, C. L., 2011. Blurred image region

detection and classification, in MM’11 - Proceedings

of the 2011 ACM Multimedia Conference and Co-

Located Workshops.

Sprawls, P, 1995. Physical principles of medical imaging.

Bovik, A., Gibson, J., 2000. Handbook of Image and

Video Processing, Academic Press, Inc. Orlando, FL,

USA.

Yang, F., Lin, H. J., Chuang, H., 2017. Image deblurring,

2017 IEEE SmartWorld, Ubiquitous Intelligence &

Computing, Advanced & Trusted Computed, Scalable

Computing & Communications, Cloud & Big Data

Computing, Internet of People and Smart City

Innovation(SmartWorld/SCALCOM/UIC/ATC/CBDC

om/IOP/SCI), San Francisco, CA.

Bansal, R., Raj, G., Choudhury T., 2017. Blur image

detection using Laplacian operator and Open-CV,

Proc. 5th Int. Conf. Syst. Model. Adv. Res. Trends,

SMART 2016, pp. 63–67.

Huang, R., Fan, M., Xing, Y., Zou, Y., 2019. Image Blur

Classification and Unintentional Blur Removal, IEEE

Access.

Koik, B. T., Ibrahim, H., 2014. A literature survey on blur

detection algorithms for digital imaging, Proc. - 1st

Int. Conf. Artif. Intell. Model. Simulation, AIMS

2013, pp. 272–277.

Ali, U., Mahmood, M. T., 2018. Analysis of Blur Measure

Operators for Single Image Blur Segmentation.

Pertuz, S., Puig, D., Garcia, M. A., 2013. Analysis of

focus measure operators for shape-from-focus, Pattern

Recognit., vol. 46, no. 5, pp. 1415–1432.

Abdel-Qader, I., Abudayyeh, O., Kelly, M. E., 2003.

Analysis of edge-detection techniques for crack

identification in bridges, J. Comput. Civ. Eng.

Gao, S., Han, M., Cheng, X., 2018. The fast iris image

clarity evaluation based on Tenengrad and ROI

selection.

Tong, H., Li, M., Zhang, H., Zhang, C., 2004. Blur

detection for digital images using wavelet transform,

in 2004 IEEE International Conference on Multimedia

and Expo (ICME).

Pech-Pacheco, P. L., Cristöbal, G., Chamorro-Martínez, J.,

Fernândez-Valdivia, J., 2000. Diatom autofocusing in

brightfield microscopy: A comparative study, Proc. -

Int. Conf. Pattern Recognit.

Liu, Y., Zhang, H., Guo, H., Xiong, N. N., 2018. A FAST-

BRISK feature detector with depth information,

Sensors (Switzerland).

Lin, J., Ji, X., Xu, W., Dai, Q., 2013. Absolute depth

estimation from a single defocused image, IEEE

Trans. Image Process.

Tang, C., Hou, C., Song, Z., 2013. Defocus map

estimation from a single image via spectrum contrast,

Opt. Lett.

Chen, D. J., Chen, H. T., Chang, L. W., 2016. Fast defocus

map estimation, in Proceedings - International

Conference on Image Processing, ICIP.

iBlurDetect: Image Blur Detection Techniques Assessment and Evaluation Study

291