Effective Image Processing Procedure for Skin Lesion Recognition in

Contactless Skin Diagnosis Devices

Hansoo Kim

Department of Information Security, Seowon University, 377-3, Mushimseoro, Seowongu, Cheongjusi, Choongbuk, Korea

Keywords: Skin Diagnosis, Lesion Recognition, Image Processing, Contactless Skin Diagnosis Device.

Abstract: An image analysis procedure for recognizing various skin lesions under contactless skin diagnosis

environment is proposed. The proposed procedure is composed of five stages, and experimental results

show that issues such as uneven distribution of light are properly addressed, and various skin lesions are

effectively discriminated according to their characteristics using the image processing technology and the

shadow analysis.

1 INTRODUCTION

Composed of epidermis, dermis, and subcutaneous

tissues, skin is the largest organ of human body (Wei

et al., 2018). Containing blood vessels, lymphatic

vessels, nerves, and muscles, which can perspire,

perceive the external temperature, and protect the

body. Covering the entire body, the skin can protect

multiple tissues and organs in the body from

external invasions including artificial skin damage,

chemical damage, adventitious viruses, and

individuals’ immune system. Besides, skin can also

avoid the loss of lipids together with water within

epidermis and dermis so that skin barrier function

can be stabilized (Hu and Yu, 2013). Skin is the first

defender of human body, from various external

hazards.

It is of great theoretical significance and practical

value to study how to extract symptoms of diverse

skin diseases on the basis of modern science and

technology. Under this circumstance, effective and

accurate identification of the types of skin diseases

can be achieved to prescribe treatment according to

patients’ symptoms (Vezhnevets et al., 2003).

Television, social media, and advertising have

had a tremendous impact on consumers’ paying

increased attention to physical appearances and

aesthetics. This has also raised interest in a variety

of cosmetic and aesthetic surgeries and procedures.

This, supported by surge in consumer disposable

income and introduction of technologically

advanced solutions, has had a major impact on

market demand (Market Research Report, 2018). As

a standard of beauty, skin increases its importance as

society develops. However, various skin disease

have emerged due to air pollution, micro dust, and

unnecessary UV exposure due to the ozone depletion.

With the development of dermatology and skin

medicine, the skin analysis technology with image

processing and computer vision gains its importance

recent years. In addition, research has been

conducted to measure skin condition quickly, easily,

and accurately.

In this paper, an image processing procedure for

recognizing skin lesions over contactless skin

imaging device is proposed. Experiments show that

analyzing the shadow of the skin lesions can

effectively detect and classify the lesion of concern.

The rest of this paper is as follows. In Section 2 ,

recent studies and technologies for skin analysis are

introduced. Details of the proposed procedure are

shown in Section 3 . Experimental Results are

shown in Section 4 and Section 5 concludes with

future work.

2 RELATED WORKS

2.1 Skin Diagnosis Device

With the development of semiconductor technology

and small electronic devices industry, devices for

measuring skin using various computerized and

electronic technologies have emerged.

Kim, H.

Effective Image Processing Procedure for Skin Lesion Recognition in Contactless Skin Diagnosis Devices.

DOI: 10.5220/0007948403370342

In Proceedings of the 16th International Joint Conference on e-Business and Telecommunications (ICETE 2019), pages 337-342

ISBN: 978-989-758-378-0

337

The global dermatology devices market size was

valued at over USD 9.6 billion in 2017 and is

expected to showcase lucrative growth over the

forecast period, registering a CAGR of 13.5%.

Increasing prevalence of skin cancer and other skin

diseases has significantly contributed to high

demand for various dermatological diagnostic and

treatment procedures in recent times, thereby

spurring product demand (Market Research Report,

2018). Figure 1 shows one of the representative

personal skin diagnosis devices (Ahn, 2019).

Figure 1: Sample of skin diagnosis device.

To get a clear image, most of these devices make

contact with the skin. Contacting the cylindrical

body with the skin enables to get a clear skin image

with fixed focal length and stable light source.

However, contact with the skin can cause

possible skin contamination due to uncleanness and

corrosion of the contact part. To prevent this, the

contact part needs to keep clean and sterilized which

lead to additional costs and actions.

2.2 Skin Analysis Technology

There have been many researches on skin analysis

using image processing. Various statistical methods

for segmentation and classification of skin lesions in

dermoscopic images is developed (Zaqout, 2016).

Texture and color features of skin diseases are

analysed (Wei et al., 2018; Yang et al., 2018).

Also, applying neural networks including the

Artificial Neural Network (ANN) and the

Convolutional Neural Network (CNN) are

conducted for effective skin recognition (Abbadi et

al., 2010; Menaka and Rohini, 2014; Yang et al.,

2018).

2.3 Object Detection and Recognition

For a long time, algorithms for finding objects in

images have been studied extensively. Numerous

researches have been done in object detection,

recognition, classification and discrimination.

Scale-Invariant Feature Transform (SIFT) has

been studied and applied various applications

(Gonzalez and Woods, 2018). Speeded Up Robust

Features (SURF) and RANdom Sample Consensus

(RANSAC) are also well-studied algorithms (Kim

and Kim, 2014). Circular Hough Transform (CHT)

is one of the widely used algorithms (Vegt, 2015).

Also, light source detection has been researched

in a number of areas, including human vision,

optical science and photometric engineering (Nillius

and Eklundh, 2001; Funk and Yang, 2007)

3 PROPOSED PROCEDURE

The proposed procedure is composed of five stages:

Preprocessing, Parameter Configuration, Abnormal

Area Separation, Lesion Recognition, and

Categorization and Visualization. The flowchart for

the overall procedure is shown in Figure 2, followed

by its pseudocode shown in Figure 3.

3.1 Preprocessing

To analyse skin and its lesions, an image is taken by

skin diagnosis device. As the device does not touch

the skin and is exposed to external light source

(sunlight, indoor lamp, etc.), various factors such as

direction and amount of light, vignetting and lens

distortion should be considered. So, color

temperature and lens distortion calibration,

vignetting and chromatic abberation correction and

other necessary processing should be applied on the

stage of imaging the skin.

After the correction and calibration of the image,

basic properties such as the image size, resolution

and the image file type are acquired for further

analysis.

3.2 Parameter Configuration

Detailed and practical properties such as the average

brightness for entire image and the image histogram

are obtained in this stage.

In addition, the image is divided into number of

image blocks. For each image block, average

brightness is obtained so that the threshold for each

block is configured using the average brightness.

The threshold for each block is used to separate the

lesion part from the skin part. This adaptive

threshold configuration enables the separation of the

skin and its lesions effectively.

SIGMAP 2019 - 16th International Conference on Signal Processing and Multimedia Applications

338

3.3 Abnormal Area Separation

Using the parameters and properties acquired from

Section 3.1 and 3.2, the lesion part is separated from

the skin part.

By the threshold value on each designated image

block, each pixel is classified upon the binarization

method (black and white manner).

3.4 Lesion Recognition

Among the number of image processing and object

recognition technologies described in Section 2.3,

CHT is one of the most appropriate algorithms to

detect and recognize the skin lesions.

In general, lesions are dark and circular in shape

compared to the skin, so it is expected that CHT is

possible to detect and recognize the skin lesions

effectively.

Furthermore, as there are lesions in a concave

shape such as pores and a convex shape such as

moles, the location of dark and bright part of those

lesions is different according to the direction of light.

Therefore, it is possible to distinguish lesions by the

direction of the shadow. In this stage, the Smart

Shadow analysis is performed; the discrimination of

the lesion by its shadow.

3.5 Categorization and Visualization

After obtaining the concave and convex lesions

according to the shadow analysis, each feature is

analysed and marked in the final stage.

Since the information obtained by the CHT

includes the position (coordinates) and size of each

lesion, the detected lesions can be displayed in an

image for visual understanding, and statistical

features such as area can also be analysed.

Figure 2: Flowchart for the proposed procedure.

Effective Image Processing Procedure for Skin Lesion Recognition in Contactless Skin Diagnosis Devices

339

INPUT : skin_Image

OUTPUT : p_center and p_radius according to lesions

/* parameter definition & configuration */

skin_Image // skin image to analyze

skin_h, skin_w // size of skin_Image

n_blocks // number of blocks (horizontal and vertical)

th_lesion // threshold value to classify lesions

p_comp // compensation parameter

p_center // array of p_center coordinate values of lesions

p_radius // array of radii of lesions

LESION // value for separating lesions

n_blocks_h=skin_h/n_blocks

n_blocks_w=skin_w/n_blocks

/* th_lesion acquisition */

for k=1 to n_blocks

s=(k-1)*n_blocks_h

for m=1 to n_blocks

n=(m-1)*n_blocks_w

g=1

for ni=1+s to n_blocks_h+s

e=1

for nj=1+n to n_blocks_w+n

tmp_Image(g,e)=skin_Image(ni,nj,1)

// the R value

e=e+1

g=g+1

th_lesion(k,m)=mean(mean(tmp_Image))-p_comp

/* lesion (abnormal area) separation */

for k=1 to n_blocks

s=(k-1)*n_blocks_h

for m=1 to n_blocks

n=(m-1)*n_blocks_w

for ni=1+s to n_blocks_h+s

for nj=1+n to n_blocks_w+n

if (skin_Image(ni,nj,1)<=th_lesion(k,m))

proc_Image(ni,nj,1)=LESION

proc_Image(ni,nj,2)=LESION

proc_Image(ni,nj,3)=LESION

/* object detection & recognition */

// to find abnormal area (dark area regarded as lesions) and localize

// put the center points and radii into arrays of p_center and p_radius

// use imfindcircles in MATLAB (Circular Hough Transform is used)

/* enhanced object recognition: Smart Shadow */

for i=1 to size(p_center)

if (a(p_center(i,2)+p_radius(i)/4,p_center(i,1),1)

>=a(p_center(i,2)-p_radius(i)/4, p_center(i,1),1))

// assumed the light direction goes downwards

p_center(j,:)=p_center(i,:)

p_radius(j)=p_radius(i)

j=j+1

Figure 3: Pseudocode of the proposed procedure.

SIGMAP 2019 - 16th International Conference on Signal Processing and Multimedia Applications

340

4 EXPERIMENTAL RESULTS

The procedure proposed in Section 3 has been

implemented using Mathworks MATLAB R2016a

on Microsoft Windows 7. After a number of close

concern of various skin lesion images taken by skin

diagnosis devices, the skin images are artificially

produced using Adobe Photoshop CC 2018.

Figure 4: The original skin.

Figure 4 shows the original skin image, regarded

as taken from a skin diagnosis device. It is regarded

as previously processed to address the issues of

color temperature difference, lens distortion,

vignetting and chromatic abberation. It is set that the

light source is at the upper left (the brightness

gradually decreases from upper left to lower right)

with the light brown skin.

Figure 5 shows the result of separating abnormal

area which are regarded as skin lesions. Red circles

denote the abnormal area. After the analysis of

recognized lesions, the number of lesions is 74 with

the total area of 40787.6 (pixels

) and the average

area per lesion of 551.2 (pixels

After the enhanced object recognition using the

shadow analysis called Smart Shadow, possible

pores (concave objects) are shown in Figure 6 and

possible moles (convex objects) are shown in Figure

7. After the analysis, the number of possible pores is

65 and the number of possible moles is 8. The

overlapped parts in Figure 5 and Figure 6 are

identified as errors, mainly due to the CHT

parameter settings.

Figure 5: The result of recognizing skin lesions.

Figure 6: The result of recognizing possible pores.

Effective Image Processing Procedure for Skin Lesion Recognition in Contactless Skin Diagnosis Devices

341

Figure 7: The result of recognizing possible moles.

5 CONCLUSIONS

An effective image analysis procedure for

discriminating and recognizing various skin lesions

under contactless skin diagnosis environment is

proposed. Issues such as uneven distribution of light

are properly addressed, and various skin lesions are

effectively discriminated according to their

characteristics, using the image processing technique

and the shadow analysis.

To generalize and objectify the proposed

procedure, additional analysis and comparison of

detailed skin lesion would be performed under

additional light source including the variety of its

strength, angle, wavelength, etc. Further researches

after this work include detailed lesion discrimination

upon appropriate decision-making frameworks such

as deep learning, and Bid Data analysis over various

kinds of skin image.

ACKNOWLEDGEMENTS

This work was supported by the National Research

Foundation of Korea (NRF) grant funded by the

Korea government (MSIT) (No.

2018R1C1B5043326).

REFERENCES

Abbadi, N. K. A., Dahir, N. S., Al-Dhalimi, M. A., and

Restom, H., 2010. Psoriasis Detection Using Skin

Color and Texture Features. In Journal of Computer

Science, 6(6):648-652.

Ahn, Y. K., 2019. Genie Skin. In AMC Co., Ltd.,

http://www.amc11.com

Funk, N., and Yang, Y.-H., 2007. Using a Raster Display

for Photometric Video. In Proceedings of Canadian

Conference of Computer and Robot Vision (CRV

2007), 201-207, Montreal, Canada.

Gonzalez, R. C., and Woods, R. E., 2018. Digital Image

Processing, 4

Ed. By Pearson Education.

Hu, Z., and Yu, C. S., 2013. Functional Research and

Development of Skin Barrier. In Chinese Journal of

Clinicians, 7(7):3101–3103.

Kim, J., and Kim, D., 2014. Matching Points Filtering

Applied Panorama Image Processing Using SURF and

RANSAC Algorithm. In Journal of the Institute of

Electronics and Information Engineers, 51(4):820-835.

Market Research Report, 2018. Dermatology Devices

Market Size, Share & Trends Analysis Report By End

Use (Clinics, Hospitals), By Product & Application

(Diagnostic, Treatment), By Region, And Segment

Forecasts, 2018 – 2025. In Grand View Research,

2018.

Menaka, R., and Rohini, S., 2014. Efficient Detection of

Ischemic Stroke from MRI Images Using Wavelet

Transform. In International Journal of Computer

Science and Information Technology Research,

2(3):446-454.

Nillius, P., and Eklundh, J. O., 2001. Automatic

Estimation of the Projected Light Source Direction. In

Proceedings of the 2001 IEEE Computer Society

Conference on Computer Vision and Pattern

Recognition (CVPR 2001).

Vegt, S. E., 2015. A Fast and Robust Algorithm for the

Detection of Circular Pieces in a Cyber Physical

System. In ES Reports, Eindhoven University of

Technology, 1-5.

Vezhnevets, V., Sazonov, V., and Andreeva, A., 2003. A

Survey on Pixel-Based Skin Color Detection

Techniques. In Proceedings of Graphicon 2003, 85–

92, Moscow, Russia.

Wei, L., Gan, Q., and Ji, T., 2018. Skin Disease

Recognition Method Based on Image Color and

Texture Features. In Computational and Mathematical

Methods in Medicine, 2018(7):1-10.

Yang, J., Sun, X., Liang, J, and Rosin, P. L., 2018.

Clinical Skin Lesion Diagnosis Using Representations

Inspired by Dermatologist Criteria. In The IEEE

Conference on Computer Vision and Pattern

Recognition (CVPR), 1258-1266.

Zaqout, I. S., 2016. Diagnosis of Skin Lesions Based on

Dermoscopic Images Using Image Processing

Techniques. in International Journal of Signal

Processing, Image Processing and Pattern

Recognition, 9(9):189-204.

SIGMAP 2019 - 16th International Conference on Signal Processing and Multimedia Applications

342