Ứng dụng đánh giá tình trạng mụn trứng cá và gợi ý lộ trình hỗ trợ điều trị

TR NG I H C BÁCH KHOA ­ HQG­HCM KHOA KHOA H C VÀ K THU T MÁY TÍNH LU N V N T T NGHI P IH C TÀI NG D NG ÁNH GIÁ TÌNH TR NG M N TR NG CÁ VÀ G I Ý L H TR TRÌNH I U TR Ngành: Khoa h c K thu t Máy tính H i đ ng: KHMT4 Gi ng viên h ng d n: ThS Tr n Ng c B o Duy Gi ng viên ph n bi n: ThS L u Quang Huân Sinh viên th c hi n: 1712393 ­ Ph m Nguy n Xuân Nguyên 1713214 ­ Tr n Th Th m 1713217 ­ Minh Th ng TP H CHÍ MINH ­ THÁNG N M 2021 NAM KHOA KH & KT MÁY TÍNH Ph MSSV: 1712393 SV: Tr Th MSSV: 1713214 SV: MSSV: 1713217 - - phúc ân Nguyên tính tính h Duy toán: - A1: A2: LVTN: - ý Play 10 2021 LVTN: 10 /10 L ic m n L i đ u tiên, nhóm xin đ tr c bày t lịng bi t n sâu s c đ n Ban Giám hi u nhà ng, khoa Khoa h c K thu t Máy tính tồn th th y/cơ t n tình gi ng d y, truy n đ t ki n th c kinh nghi m quý báo cho nhóm su t quãng th i gian h c t p làm vi c t i tr Nhóm đ c bi t xin đ ng i th y, ng ng i h c Bách Khoa ­ HQG­HCM c g i l i c m n sâu s c đ n ThS Tr n Ng c B o Duy, i gi ng viên h ng d n giám sát tr c ti p trình th c hi n đ tài Nh có nh ng ch d n, góp ý t n tình nhi u ki n th c b ích, quý giá c a th y mà nhóm m i có th hồn thành t t đ Nhóm c ng xin đ l c c m n TS Tr n c Lu n v n t t nghi p c D ng ThS L u Quang Huân, l n t gi ng viên ph n bi n c a đ tài trình c ng lu n v n gi ng viên ph n bi n trình Lu n v n t t nghi p, v i nh ng góp ý đ n mang tính đ nh h ng cao cho nhóm Ngồi ra, nhóm c ng xin đ c c m n Bác s Tào H ng Nga t B nh vi n Da Li u t nh Khánh Hòa v nh ng h tr vi c t v n, b sung ki n th c v Da Li u đ nhóm có đ hồn thành đ c đ tài Cu i cùng, nhóm c ng xin g i l i c m n đ n t t c anh/ch /em b n sinh viên s d ng, góp ý cho s n ph m c bi t c m n b n Ph m c Duy Anh, Tr n Th Thanh Kim Hu , Tr n Th Thanh Trúc đóng góp m t ph n quan tr ng khâu x lý d li u đ đ tài có th th c hi n theo đ nh h v ch ng ban đ u c bi t c m n b n oàn­H i khoa Khoa h c K thu t Máy tính ki m th s n ph m đ s n ph m hoàn thi n nh t có th tài đ c lên ý t ng hoàn thi n thành s n ph m hoàn c nh d ch b nh c ng th ng g p nhi u khó kh n giãn cách xã h i t i Vi t Nam Vì th , m i đóng góp cho đ tài đ kh e đ v c nhóm trân tr ng xin chúc m i ng t qua đ i d ch l n i có d i s c Tóm t t n i dung Xã h i ngày phát tri n, cu c s ng tr nên đ y đ ti n nghi h n ng i bi t ch m lo cho b n thân c a h n Nh ng vi c xã h i phát tri n c ng góp ph n n cho ng i tr nên b n r n cho th i gian ch m sóc cho b n thân mình, đ c bi t h c sinh, sinh viên công nhân viên ch c v n phòng B n thân bác s da li u c ng tr nên b n r n h n th i đ i nhu c u c i thi n nét đ p g ng m t tr nên ngày ph bi n h n Vi c có m t ng d ng đ có giúp đ ph n cơng vi c nh ng bác s hay góp ph n c i thi n ch t l cu c s ng c a b n thân nh ng ng i b n r n u c n thi t Nh n th y nh ng nhu c u th c ti n đó, nhóm n y sinh m t ý t thân m i ng ngày d i t c i thi n ch t l ng ng nh m giúp b n ng đ i s ng vi c t ch m sóc da m t m i i hình th c ng d ng n tho i thơng minh có ng d ng s c m nh c a Trí tu nhân t o (AI) th vi n h tr s n đ giúp đ đ c ph n gánh n ng phía ng i dùng tr nên thu n ti n h n i t ch c c ng nh giúp cho vi c s d ng c a ng ng d ng đ c kì v ng có kh n ng đ a k t lu n v tình tr ng da m t, đ ng th i g i ý l trình u tr phù h p v i tình tr ng m i ng i M CL C I L i cam đoan L ic m n Tóm t t n i dung Gi i thi u 14 tv nđ 14 M c tiêu, gi i h n giai đo n th c hi n lu n v n 14 2.1 M c tiêu c a đ tài 14 2.2 Gi i h n c a đ tài 15 2.3 Các giai đo n th c hi n lu n v n 16 C u trúc lu n v n 16 II Ki n th c n n t ng Kh i ki n th c v y h c 17 1.1 D u hi u nh n bi t v m n 17 1.2 Ph 19 1.3 T l vàng khn m t ng 1.4 Kích th ng pháp đánh giá tình tr ng m n i 21 c m n tr ng cá 23 1.5 Ki n th c v s n ph m ch m sóc da 25 1.6 Ki n th c v l trình s d ng s n ph m ch m sóc da 27 1.7 c m không gian màu phân đo n màu da (Skin Color Segmentation) 28 Kênh màu RGB 29 1.8.1 Không gian màu Lab 29 1.8.2 Không gian màu YCrCb 31 1.8.3 Không gian màu HSV 32 Kh i ki n th c v h c máy 32 2.1 M ng n ­ron tích ch p ­ Convolution Neural Network (CNN) 32 2.1.1 Gi i thi u 32 2.1.2 Các công th c toán h c 34 2.1.3 Kh n ng c a m ng CNN 34 2.1.4 M t s hàm kích ho t (activation function) phi n 34 2.1.5 Hu n luy n CNN 1.8 17 35 2.2 Transfer Learning 35 2.2.1 Khái ni m pretrain model 35 2.2.2 Khái ni m Transfer Learning 35 2.2.3 VGG16 37 Ki n th c v h th ng g i ý 39 2.3.1 L c n i dung (Content­based Filtering) 40 2.3.2 L c c ng tác (Collaborative Filtering) 41 2.3.3 L c d a mi n ki n th c (Knowledge Filtering) 41 2.3.4 H ng ti p c n lai 42 Kh i ki n th c v công ngh 43 3.1 Ngơn ng l p trình 43 3.1.1 Python 43 3.1.2 Javascript 43 3.1.3 Dart 43 Các n n t ng framework th vi n 44 3.2.1 Flask 44 3.2.2 Keras 44 3.2.3 Tensorflow 44 3.2.4 Pandas 44 3.2.5 Scikit­learn 45 3.2.6 Pymongo 45 3.2.7 Flutter 45 3.2.8 Selenium 45 Công c 46 3.3.1 Google Colab 46 3.3.2 Mongo Atlas 46 3.3.3 Full­Text Search 46 2.3 3.2 3.3 III Cơng trình nghiên c u liên quan 47 Các cơng trình liên quan đ n phát hi n m n 47 1.1 Khái quát hóa gi i pháp phát hi n m n 47 1.2 X lý hình nh y t phát hi n m n t đ ng đ u tr y t [2014] 1.3 48 Theo dõi da li u v đ hi u qu vi c u tr m n mãn tính [2016] 50 1.4 ánh giá m n t đ ng khn m t t hình nh n tho i thông minh [2018] 51 Các công trình liên quan đ n phân lo i m n 53 2.1 Khái quát gi i pháp phân lo i m n 53 2.2 Mơ hình phân lo i m n tr ng cá khuôn m t s d ng h c sâu Jung Cheeoh c ng s đ xu t [2019] 54 Mơ hình phân lo i m n tr ng cá khuôn m t s d ng h c sâu Xiaolei Shen c ng s đ xu t [2018] 56 Các cơng trình liên quan đ n đánh giá đ m n 59 Các công trình liên quan đ n h th ng g i ý 62 5.1 Khái quát gi i pháp g i ý s n ph m 62 5.2 H th ng khuy n ngh s n ph m ch m sóc da d a n i dung [2020] 5.3 H th ng khuy n ngh đ mua m ph m b ng cách s d ng l c d a n i dung [2018] IV Ph 62 ng pháp th c hi n 65 68 Phát hi n m b t th ng cho khuôn m t 69 1.1 T ng quan v phát hi n m n 69 1.2 Phát hi n khuôn m t ng i 71 1.3 Phát hi n vùng quan tâm (ROI) 71 1.4 Cân b ng histogram 72 1.5 Phát hi n m n 72 1.5.1 Phát hi n m n đ u tr ng 73 1.5.2 Phát hi n m n đ u đen 74 1.5.3 Phát hi n m n đ u viêm 74 1.6 Trích xu t ng ng (Threshold extraction) 76 1.7 X lý hàng đ i (Queue handling) 76 1.8 X lý lo i b nhi u 76 1.9 K t qu đánh giá 77 1.10 Th o lu n v ph ng pháp phát hi n m n 81 ng 82 2.1 T p d li u 82 2.2 Mơ hình l a ch n 83 2.3 ánh giá mơ hình 86 Phân lo i m b t th 2.3.1 it ng đánh giá 87 TÀI LI U THAM KH O [1] Ms.Watcharaporn Sitsawangsopon, Ms Maetawee Juladash, ”Medical Image Processing in Automatic Acne Detection for Medical Treatment” School of Information, Computer and Communication Technology, Sirindhorn International Institute of Technology, Thammasat University, March 2, 2014 [2] Lucut, S., & Smith, M R (2016) Dermatological tracking of chronic acne treatment effectiveness Proceedings of The 2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) held in Orlando, 16 20 August 2016 (pp 5421 5426) Florida, USA: Disney s Contemporary Resort at Walt Disney World® Resort [3] Mohammad Amini, Fartash Vasefi, Manuel Valdebran, Kevin Huang, Haomiao Zhang, et al., ”Automated facial acne assessment from smartphone images,” Proc SPIE 10497, Imaging,Manipulation, and Analysis of Biomolecules, Cells, and Tissues XVI, 104970N(22 February 2018) [4] Xiaolei Shen, Jiachi Zhang, Chenjun Yan Hong Zhou, et al., ”An Automatic Diagnosis Method of Facial Acne Vulgaris Based on Convolutional Neural Network,” Scientific Report 2018, (11 April 2018) [5] Jung Cheeoh et al., ”Classification Model of Facial Acne Using Deep Learning” Journal of the Korea Institute of Infomation and Communication Engineering, (30 April 2019) [6] Gyeongeun Lee, Department of Computer Science, Earlham College ”A Content­based Skincare Product Recommendation System” [7] Jiwon Jeong 2018 For Your Skin Beauty: Mapping Cosmetic Items with Bokeh https://towardsdatascience.com/for­ your­ skin­ beauty­ mapping­ cosmetic­ items­ with­ bokeh­ af7523ca68e5 [8] Joanna Cristy Patty, Elika Thea Kirana and Made Sandra Diamond Khrismayanti Giri, J C Patty et al, Recommendations System for Purchase of Cosmetics Using Content­Based Filtering, International Journal of Computer Engineering and Information Technology (IJCEIT), Volume 10, Issue 1, January 2018 [9] Nguyen Pham Nguyen Xuan, Tham Tran Thi, Thang Do Minh, Duy Tran Ngoc Bao, ”A Multilevel Thresholding Approach for Acne Detection in Medical 140 Treatment”, 3th International Conference on Image Processing and Machine Vision (IPMV), 2021 [10] Gabriele Maroni, Michele Ermidoro, and Fabio Previdi, ”Automated Detection, Extraction and Counting of Acne Lesions for Automatic Evaluation and Tracking of Acne Severity” in 2017 IEEE Symposium Series on Computational Intelligence (SSCI), 2017 [11] N Alamdari, K Tavakolian, M Alhashim, and R Fazel­Rezai, Detection and classification of acne lesions in acne patients: A mobile application in Electro Information Technology (EIT), 2016 IEEE International Conference on IEEE 2016, pp 0739 0743 [12] Thanapha Chantharaphaichit, Bunyarit Uyyanonvara, Chanjira Sinthanayothin, Akinori Nishihara, ”Automatic Acne Detection for Medical Treatment”, 2015 6th International Conference of Information and Communication Technology for Embedded Systems (IC­ICTES) [13] Ramli, R., Malik, A S., Yap, F B ”Identification of acne lesions, scars and normal skin for acne vulgaris cases” Proceedings of The 2011 National Postgraduate Conference held in Perak, 19 20 September 2011 (pp.1 4) Malaysia: Universiti Teknologi PETRONAS (UTP) ­ Tronoh [14] Chen, D., Chang, T., & Cao, R ”The development of a skin inspection imaging system on an Android device.” Proceedings of The 7th International Conference on Communications and Networking in China held in Kun Ming, 10 August 2012 (pp 653 658) China: Dianchi Garden Hotel & Spa [15] Fujii, H., Yanagisawa, T., Murakami, Y., & Yamaguchi M ”Extraction of acne lesion in acne patients from multispectral images” Proceedings of an Annual International IEEE EMBS Conference held in Vancouver, 21 24 August 2008 (pp.4078 4081) British Columbia, Canada: The Vancouver Convention and Exhibition Center [16] Chang, C.­Y Liao, H.­Y ”Automatic facial skin defects detection and recognition system.” In Genetic and Evolutionary Computing” (ICGEC), 2011 Fifh International Conference on, 260 263 (IEEE, 2011) [17] Malik, A S et al ”Digital assessment of facial acne vulgaris.” In Instrumentation and Measurement Technology Conference Proceedings, 2014 IEEE International, 546 550 (IEEE, 2014) 141 (I2MTC) [18] Ciregan, D., Meier, U Schmidhuber, J ”Multi­column deep neural networks for image classifcation.” In Computer vision and pattern recognition (CVPR), 2012 IEEE conference on, 3642 3649 (IEEE, 2012) [19] LeCun, Y et al ”Backpropagation applied to handwritten zip code recognition.” Neural computation 1, 541 551 (1989) [20] Sermanet, P., Chintala, S LeCun, Y ”Convolutional neural networks applied to house numbers digit classifcation.” In Pattern Recognition (ICPR), 2012 21st International Conference on, 3288 3291 (IEEE, 2012) [21] Hinton, G E., Srivastava, N., Krizhevsky, A., Sutskever, I Salakhutdinov, R R ”Improving neural networks by preventing co­adaptation of feature detectors.” arXiv preprint arXiv:1207.0580 (2012) [22] Krizhevsky, A., Sutskever, I Hinton, G E ”Imagenet classifcation with deep convolutional neural networks.” In Advances in neural information processing systems, 1097 1105 (2012) [23] He, K., Zhang, X., Ren, S Sun, J ”Delving deep into rectifers: Surpassing human­level performance on imagenet classifcation In Proceedings of the IEEE international conference on computer vision, 1026 1034 (2015) [24] Xiaoping Wu et al., ”Joint Acne Image Grading and Counting via Label Distribution Learning” (2019) [25] Villia Putriany, Jaidan Jauhari, Rahmat Izwan Heroza Faculty of Computer Science, Sriwijaya University, Indralaya, Indonesia ”Item Clustering as An Input for Skin Care Product Recommended System using Content Based Filtering” [26] Tingting Zhao et al., ”A Computer Vision Application for Assessing Facial Acne Severity from Selfie Images” (2019) [27] Ziying Vanessa Lim et al., ”Automated grading of acne vulgaris by deep learning with convolutional neural networks” (2019) [28] Nguy n Thanh Tu n, Sách Deep Learning c b n, ch ng (2019) [29] K Simonyan and A Zisserman, University of Oxford, Very Deep Convolutional Networks for Large­Scale Image Recognition 142 , ILSVRC (2014) [30] TS Nguy n Th Hi n, ”Nghiên c u đ c tính hi u qu c a ACNECA vi c u tr b nh tr ng cá thông th ng th v a”, Lu n án Ti n s Y h c, Hà N i, 2020 [31] Jeremy A.H, Holland D.B, Roberts S.G et al (2003) Inflammatory events are involved in acne lesion initiation J Invest Dermatol, 121(1), 20­7 [32] Thiboutot D, Gollnick H, Bettoli V et al (2F009) New insights into the management of acne: an update from the Global Alliance to Improve Outcomes in Acne group J Am Acad Dermatol, 60(5 Suppl), S1­50 [33] Holland DB, Jeremy AH, Roberts SG et al (2004) Inflammation in acne scarring: a comparison of the lesions from patients prone and not prone to scar Br J Dermatol, 150, 72­81 [34] Simpson NB, Cunliffe WJ Disorders of sebaceous glands In: Burns T, Breathnach S, Cox N, Griffiths C, editors Rook’s Textbook of Dermatology, 7th ed., Oxford: Blackwell publishing; 2004 p 43.1 ­ 43.75 [35] Nobukazu Hayashi, Hirohiko Akamatsu, Makoto Kawashima, and Acne Study Group Establishment of grading criteria for acne severity The Journal of Dermatology, 35(5):255 260, 2008 [36] Christos C Zouboulis Andreas D Katsambas Albert M Kligman Editors Pathogenesis and Treatment of Acne and Rosacea p 317 [37] R Ramli, A S Malik, A F M Hani, and A Jamil, Acne analysis, grading and computational assessment methods: an overview Skin Res Technol., vol 18, no 1, pp 14, Feb 2012 [38] Hideaki Fujii, Takashi Yanagisawa, Masanori Mitsui, Yuri Murakami, Masahiro Yamaguchi, Nagaaki Ohyama, Tokiya Abe, Ikumi Yokoi, Yoshie Matsuoka, and Yasuo Kubota ”Extraction of acne lesion in acne patients from Multispectral Images” Annual International IEEE EMBS Conference Vancouver, British Columbia, Canada pp.4078­4081, 2008 [39] Siddharth K, Madan and Kristin J, Dana ”Learning­Based Detection of Acne­like Regions Using Time­Lapse Features” Department of Electrical and Computer Engineering, Rutgers University NJ, USA 2011 143 [40] Phannapat S, Watcharaporn S, Maetawee J, Guntachai O ”Face Detection” School Of Information Computer and Communication Technology Sirindhorn International Institute of Technology, Thammasat University, Thailand 2013 [41] Anne Kaspers ”Blob detection” Biomedical Image Sciences,Image Sciences Institute, UMC Utrecht, 2011 [42] Chai D, Ngan K.N., ”Face segmentation using skin­color map in videophone applications,” Circuits and Systems for Video Technology, IEEE Transactions on , vol.9, no.4, pp.551,564, Jun 1999 [43] Chuan­Yu, Shang–ChengLi, Pau–Choo Chung, Jui–Yi Kuo, Yung–Chin Tu ”Atomatic Facial Skin Defect Detection System.” Dept of Computer Science Information Engineering, National Yunlin University of Science Technology, Taiwan pp.527—532, 2010 [44] Kazemi, V., & Josephine, S., One millisecond face alignment with an ensemble of regression trees In 27th IEEE Conference on Computer Vision and Pattern Recognition, CVPR 2014, Columbus, United States, (pp 1867­1874) IEEE Computer Society (2014) [45] Emmanuel P Prokopakis, Ioannis M Vlastos, Valerie Picavet, Gilbert Nolst Trenité, Regan Thomas, Cemal Cingi, Peter W Hellings ”The golden ratio in facial symmetry”, 05/07/2012 [46] Leon J.Kamin, ”Race, head size, andintelligence”, 28/03/1998 [47] Gross G (2007) ”Benzoyl Peroxide and Salicylic Acid Therapy ’Acne and Its Therapy’”, 117­132 [48] Waugh J, Noble S, Scott L.J (2004) ”Adapalene: a review of its use in the treatment of acne vulgaris.” Drugs, 64(13), 1465­78 [49] Mills O.H, Kligman A.M, Pochi P et al (1986) ”Comparing 2.5%, 5%, and 10% benzoyl peroxide on inflammatory acne vulgaris.” Int J Dermatol, 25(10), 664­7 [50] Webster G (2000) ”Combination azelaic acid therapy for acne vulgaris.” J Am Acad Dermatol, 43(2 Pt 3), S47­50 [51] Cove J.H, Holland K.T (1983) ”The effect of benzoyl peroxide on cutaneous micro­organisms in vitro.” J Appl Bacteriol, 54(3), 379­82 144 [52] Hegemann L, Toso S.M, Kitay K et al (1994) Anti­inflammatory actions of benzoyl peroxide: effects on the generation of reactive oxygen species by leucocytes and the activity of protein kinase C and calmodulin Br J Dermatol, 130(5), 569­75 [53] Sagransky M, Yentzer B.A, Feldman S.R (2009) ”Benzoyl peroxide: a review of its current use in the treatment of acne vulgaris.” Expert Opin Pharmacother, 10(15), 2555­62 [54] Thiboutot D, Gollnick H, Bettoli V et al (2009) ”New insights into the management of acne: an update from the Global Alliance to Improve Outcomes in Acne group.” J Am Acad Dermatol, 60(5 Suppl), S1­50 [55] Pediatric New, ”Acne vulgaris and rosacea: Optimizing therapy” (2005), Produced in affiliation with the 29th Annual Hawaii Dermatology Seminar [56] Benjamin Barankin, Anatoli Freiman, ”Derm notes Dermatology Clinical Pocket Guide”, p.54 [57] R F Rahmat, T Chairunnisa, ”Skin Color Segmentation Using Multi­Color Space Threshold”, 2016 3rd International Conference On Computer And Information Sciences (ICCOINS), Aug 2016 [58] R Rothe, M Guillaumin, ”Non­Maximum Suppression for Object Detection by Passing Messages between Windows” Conference: Asian Conference on Computer Vision (ACCV 2014), Apr 2015 [59] S Albawi, T A Mohammed, S Al­Zawi, ”Understanding of a convolutional neural network”, 2017 International Conference on Engineering and Technology (ICET), Aug 2017 [60] S L Phung, A Bouzerdoum, and D Chai, Skin segmentation using color pixel classification: analysis and comparison, IEEE transactions on pattern analysis and machine intelligence, vol 27, no 1, pp 148 154, 2005 [61] Vikas Gupta, ”Color spaces in OpenCV”, [62] L.R.RoopeshandB.Tulasi, A survey of travel recommender system Dept.Computer Science, CHRIST Deemed to be University, Bengaluru, India, vol 9, Sept 2018 145 [63] J Leskovec, A Rajaraman, and J D Ullman, Mining of massive datasets, vol 9, 2012 [64] V H Ti p, Content­based recommendation systems, May 2017 [65] Francesco Ricci, Lior Rokach, ”Recommender System Handbook” (2010) [66] S S Sohail, J Siddiqui, and R Ali, Classifications of recommender systems: A review, JOURNAL OF Engineering Science and Technology Review, Sep 2017 146 PH L C Bài báo khoa h c công b : Nguyen Pham Nguyen Xuan, Tham Tran Thi, Thang Do Minh, Duy Tran Ngoc Bao, A Multilevel Thresholding Approach for Acne Detection in Medical Treatment, 3th International Conference on Image Processing and Machine Vision (IPMV), 2021 147 A Multilevel Thresholding Approach for Acne Detection in Medical Treatment Nguyen Pham Nguyen Xuan∗ Tham Tran Thi Faculty of Computer Science and Engineering, Ho Chi ℧inh City University of Technology, VNU-HC℧, Vietnam Faculty of Computer Science and Engineering, Ho Chi ℧inh City University of Technology, VNU-HC℧, Vietnam Thang Do ℧inh Duy Tran Ngoc Bao Faculty of Computer Science and Engineering, Ho Chi ℧inh City University of Technology, VNU-HC℧, Vietnam Faculty of Computer Science and Engineering, Ho Chi ℧inh City University of Technology, VNU-HC℧, Vietnam ABSTRACT In the quantitative assessment on the success of treatment, the automatic detection of acne pixels from digital color images would be helpful In this paper, we proposed an automatic acne detection method through the processing of facial images taken by the smartphone based on the image processing In this approach, the RGB image is transformed into various color spaces based on the diferences between features of each acne lesion type This method has been used the a* cha♪♪el of the CIELab color space to detect the i♪lammatory ac♪e (papules a♪d pustules) The S cha♪♪el of HSV color space was used to detect the ♪o♪-i♪lammatory ac♪e (whiteheads a♪d blackheads) A multi-level threshold is the♪ used to make ac♪e extractio♪ a♪d blob detectio♪ The efective♪ess of the proposed procedure is show♪ by experime♪tal results Ωe showed the possibility of detecti♪g types of ac♪e lesio♪s (whiteheads, blackheads, papules, pustules) with difere♪t ski♪ colors a♪d differe♪t smartpho♪es i♪ this experime♪t by applyi♪g a combi♪atio♪ of several color spaces The result shows a recall of about 85.71% i♪ detecti♪g difere♪t ac♪e types at a reaso♪able processi♪g time This is the remise to help doctors to assess the level of ac♪e o♪ the patie♪t's face i♪ a♪ efective a♪d time-savi♪g way Ac♪e is the most commo♪ ski♪ disease a♪d the leadi♪g reaso♪ for visiti♪g a dermatologist It afects approximately 50 millio♪ people i♪ the USA, with a♪♪ual costs for health care i♪ ac♪e exceedi♪g $1 billio♪ i♪ the USA alo♪e I♪ adolesce♪ts, ac♪e ca♪ afect self-image, psychological well-bei♪g, feeli♪gs, perso♪al relatio♪ships, sports, social life, a♪d may eve♪ precipitate suicide It has bee♪ suggested that it is importa♪t to ide♪tify adolesce♪ts that are afected by their ac♪e early to reduce the future socioeco♪omic burde♪ of their disease [2] Therefore, there are a massive ♪umber of ac♪e patie♪ts that ♪eed speciic treatme♪t immi♪e♪tly U♪til ♪ow, the popular ski♪ a♪alysis is take♪ picture a♪d ma♪ually cou♪t a♪d mark but this is subjective, tedious a♪d time co♪sumi♪g To address these problems, there is a♪ evolutio♪ o♪ proposi♪g computatio♪al image processi♪g method for ac♪e detectio♪ i♪ rece♪t years I♪ 2018, Ami♪i et a [10] studied used the a* cha♪♪el i♪ CIE L*a*b* color space to ide♪tiies a♪d classiies i♪lammatory ac♪e lesio♪s (papules a♪d pustules) The algorithm used to ide♪tify ac♪e lesio♪s was able to detect 92% of lesio♪s (withi♪ ROIs) for difere♪t lighti♪g, a♪d ski♪ color variatio♪ They observed that ac♪e detectio♪ missed ac♪e blemishes with a pale red♪ess (i♪ very early or late stages) ℧aro♪i et al [4] developed a prototype applicatio♪ for automatic ac♪e detectio♪ through the processi♪g of dista♪ce picture take♪ by mobile devices They used a* cha♪♪el of the CIELab color space for e♪ha♪ce the visual co♪trast betwee♪ i♪lammed zo♪es a♪d healthy ski♪ Fi♪ally, ac♪e extractio♪ a♪d blob detectio♪ were accomplished through Adaptive Thresholdi♪g a♪d Laplacia♪ of Gaussia♪ ilteri♪g Accordi♪g to the authors, their result was acceptable Alamdari et al [11] is the irst work that developed a detectio♪ a♪d classiicatio♪ ac♪es by processi♪g close-up images take♪ by smartpho♪e They prese♪ted several image segme♪tatio♪ methods to detect ac♪e k-mea♪s clusteri♪g, HSV model segme♪tatio♪ techiques had accuracy of about 70 CCS CONCEPTS · Computing methodologies; · Artiicial intelligence; · Computer vision; · Computer vision problems; · Object detection; KEYWORDS Ac♪e detectio♪, Image processi♪g, ℧ulti-level threshold ACM Reference Format: Nguye♪ Pham Nguye♪ Xua♪, Tham Tra♪ Thi, Tha♪g Do ℧i♪h, a♪d Duy Tra♪ Ngoc Bao 2021 A ℧ultilevel Thresholdi♪g Approach for Ac♪e Detectio♪ i♪ ℧edical Treatme♪t I♪ 2021 3rd International Conference on Image Processing and Machine Vision (IPMV) (IPMV 2021), May 22ś24, 2021, Hong Kong, China AC℧, New York, NY, USA, pages https:⁄⁄doi.org⁄10.1145⁄3469951.3469955 ∗ Correspo♪di♪g E-mail: ♪ickpham2808@gmail.com Permissio♪ to make digital or hard copies of all or part of this work for perso♪al or classroom use is gra♪ted without fee provided that copies are ♪ot made or distributed for proit or commercial adva♪tage a♪d that copies bear this ♪otice a♪d the full citatio♪ o♪ the irst page Copyrights for compo♪e♪ts of this work ow♪ed by others tha♪ AC℧ must be ho♪ored Abstracti♪g with credit is permitted To copy otherwise, or republish, to post o♪ servers or to redistribute to lists, requires prior speciic permissio♪ a♪d⁄or a fee Request permissio♪s from permissio♪s@acm.org IPMV 2021, May 22ś24, 2021, Hong Kong, China © 2021 Associatio♪ for Computi♪g ℧achi♪ery AC℧ ISBN 978-1-4503-9004-0⁄21⁄05 $15.00 https:⁄⁄doi.org⁄10.1145⁄3469951.3469955 INTRODUCTION Cha♪tharaphaichit et al [13] proposed a♪ automatic detectio♪ o♪ part of face image by the processi♪g of ℧ATLAB program Firstly, RGB image cha♪ged i♪to gray-scale a♪d HSV image The♪ gray data was ♪ormalized a♪d subtract from HSV model to obtai♪ ROI They applied a method of small spots a♪d large regio♪ elimi♪atio♪ to classify ac♪e or ♪ot The system had fair accuracy with se♪sitivity 66 - 100%, precisio♪ 56.52 - 100%, accuracy 52-87% Ramli et al [12] addressed the segme♪tatio♪ of ac♪e lesio♪ usi♪g the CIELAB color space Firstly, sample images are collected i♪ IPMV 2021, May 22–24, 2021, Hong Kong, China RGB Seco♪d, the RGB images are co♪verted to CIELAB Third, the ac♪e lesio♪s, scares a♪d ♪ormal ski♪ are ide♪tiied usi♪g Euclidea♪ dista♪ce The result had se♪sitivity, speciicity greater tha♪ 80% Lucut et al [9] proposed a♪ ac♪e tracki♪g to exte♪d previous work by ramli et al They used histogram equalizatio♪ to improve co♪trast, a combi♪atio♪ of the Hough tra♪sformed a♪d image chroma to better ide♪tify a♪d classify ac♪e features The experime♪t showed result i♪ 59-99% Characterizatio♪ Se♪sitivity Kho♪gsuwa♪ et al [7] prese♪ted cou♪ti♪g ♪umber of poi♪ts for ac♪e usi♪g UV Fluoresce♪e a♪d image processi♪g The UV image cropped is resized a♪d co♪verted to gray image Quality of this gray image will be improved for image e♪ha♪ceme♪t usi♪g adaptive histogram equalizatio♪ The♪ exte♪ded maxima is used for cou♪ti♪g ♪umber of ac♪e poi♪ts The result show that 83.70% of accuracy, 98.22% of se♪sitivity, 85.04% of precisio♪ Che♪ et al [1] developed a ski♪ i♪spectio♪ imagi♪g system o♪ a♪ A♪droid device A series of ski♪ images are pre-processed to extract image feature to classify the ski♪ co♪ditio♪ The YCbCr color space a♪d simple thresholdi♪g to extract ac♪e feature The result showed 82 Fuljii et al [3] proposed a♪ extractio♪ method usi♪g the spectral i♪formatio♪ of the vairous type of ac♪e ski♪ lesio♪s calculated from the multispectral images of the lesio♪ I♪ the experime♪t, the results o♪ real cli♪ical images showed viability of the method ℧ost of previous work detected the ski♪ lesio♪ by the processi♪g of cha♪♪el of color space which there is o♪ly o♪e type ac♪e lesio♪ However, i♪ this paper, we used images that take♪ from smartpho♪e a♪d combi♪e several color spaces to extract more type ac♪e lesio♪s (whitehead, blackhead, papules, pustules) from difere♪t smartpho♪e images with difere♪t color ski♪, which makes the problem more challe♪gi♪g Fi♪ally, the multilevel threshold is used to extract ac♪e PROPOSED METHOD Here, our algorithm has importa♪t parameters: a ♪umber of regio♪s of i♪terest, called R, a le♪gth of list of thresholdi♪gs, called T a♪d a ♪umber of the ac♪e, called A From that, the complexity of this is O (3 * R * T * A) Ωe ca♪ see that the time complexity belo♪gs to A or a ♪umber of the ac♪e o♪ huma♪'s face a♪d T or a le♪gth of list of thresholdi♪gs, because a ♪umber of regio♪s of i♪terest (R) is co♪sta♪t If someo♪e's face is experie♪ci♪g the severe level of ac♪e, it will take some time to detect fully ac♪e o♪ this face O♪ the co♪trary, the mild level of ac♪e is faster tha♪ the severe o♪e 2.1 Human face detection Huma♪ face detectio♪ is the irst step of process The image of the face must be take♪ from the fro♪t camera The e♪tire face must be withi♪ the image a♪d the eyes must be ope♪ Ωe used a♪ available algorithm that detects localize a♪d represe♪t salie♪t regio♪s of the face (Eyes, Eyebrows, Nose, ℧outh, Jawli♪e) base o♪ 68 la♪dmark poi♪ts [8] This result is used later i♪ the process for facial regio♪s of i♪terest determi♪atio♪ The followi♪g illustratio♪ will provide more details of this process NGUYEN PHAM NGUYEN XUAN et al For this step, face image will remove the backgrou♪d parts a♪d focuses o♪ly o♪ the real face part Si♪ce the♪, the determi♪atio♪ of 68 la♪dmark poi♪ts becomes more accurate 2.2 Facial regions of interest detection The 68 la♪dmark poi♪ts is mapped to the speciic facial structures Based o♪ them, we ca♪ i♪d regio♪s of i♪terest (ROI) o♪ the face such as forehead, ♪ose, left a♪d right cheek, a♪d chi♪ The co♪siste♪t speciicatio♪ of ROI is importa♪t to provide a qua♪titative assessme♪t of ac♪e (size a♪d ♪umber) Each ROI is depe♪de♪t o♪ the speciic face of each huma♪, the dista♪ce from camera to the face, the a♪gle of the camera a♪d the face with respect to each other The i♪put is the image which from the mobile pho♪e The resolutio♪ i♪ the camera pho♪e is ♪ot as perfect as the image from the professio♪al camera The i♪put image is i♪lue♪ced by light i♪te♪sity a♪d the a♪gle of the camera a♪d the face with respect to each other This causes the areas ♪ear the bou♪dary to be afected a♪d give erro♪eous results This is the reaso♪ why we use o♪ly ROIs a♪d the rest of the face is ♪ot used 2.3 Acne detection The ♪ext step of the pipeli♪e is ac♪e detectio♪ process I♪ this system, the data a♪alysis methods have bee♪ used for ac♪e detectio♪ that is ski♪ spectral distributio♪ (color a♪alysis) Ac♪e is categorized i♪to difere♪t type but i♪ this paper, we chose to detect type i♪cluded: whitehead (♪o♪-i♪lammatory) are ac♪e lesio♪s that are ti♪y (1-2mm diameter), whitish bumps, same color with ski♪ Blackhead (♪o♪-i♪lammatory) are similar to whitehead with ti♪y size (1-2mm diameter), but the surface of the ski♪ remai♪s ope♪ with a dark appeara♪ce such as black a♪d brow♪ color Papules (i♪lammatory) are ac♪e lesio♪s that are visible o♪ the ski♪ as small (less tha♪ 10 mm i♪ diameter) a♪d irm pi♪k bump Pustules are full of visible pus, which is emerges red at the base with a yellowish or a whitish ce♪ter with small rou♪d lesio♪s that are i♪lamed To e♪sure the good level of recall i♪ detect ac♪e, several colour space were combi♪ed together o♪ segme♪tatio♪ of ski♪ pixel For detail, the cha♪♪el a* of the CIELab model showed to be the most i♪formative feature for i♪lammatory ac♪e (papules, pustules) Because the a* cha♪♪el represe♪ts the red♪ess of the pixel color i♪depe♪de♪t of its lumi♪a♪ce a♪d ca♪ be used to robustly detect the i♪crease of ski♪ red♪ess The S cha♪♪el i♪ HSV colour spaces suitable to e♪ha♪ce discrimi♪atio♪ betwee♪ healthy ski♪ a♪d white head Fi♪ally, ROI threshold was selected to detect black head Inlammatory (papules and pustules) i♪ this cha♪♪el a*, red or ora♪ge color of i♪lammatory ac♪e are cha♪ged to white color, they ca♪ be difere♪tiated from other colors The sample images applied cha♪♪el a* are show♪ i♪ igure The S cha♪♪el was chose♪ based o♪ the previous dei♪itio♪ of blackhead to emphasize black⁄brow♪ areas prese♪ted i♪ the ski♪ the sample images are show♪ i♪ igure I♪ this cha♪♪el S, black color of blackheads is cha♪ged to white color, they ca♪ be difere♪tiated from other colors The sample images applied cha♪♪el S are show♪ i♪ igure Non-inlammatory acne - whiteheads ca♪ be ide♪tiied by their visible o♪ ski♪ a♪d have same color with ski♪ Ωe combi♪ed cha♪♪el S i♪ the HSV color space to detect whitehead I♪ this cha♪♪el A Multilevel Thresholding Approach for Acne Detection in Medical Treatment IPMV 2021, May 22–24, 2021, Hong Kong, China Algorithm: Automatic Ac♪e Detectio♪ Input: Face Image Output: Coordi♪ates of ac♪e spots Recog♪ize Face creates 68 poi♪ts based o♪ Face Image I♪itialize list L Specify Regio♪s of I♪terest creates regio♪s of i♪terest (ROI) based o♪ Face Image a♪d 68 poi♪ts the♪ appe♪d each ROI to L while L is not empty I♪itialize a♪ empty queue A while Not yet use all methods ROI ← The_irst_eleme♪t_of_L I♪itialize a queue Q a♪d place ROI i♪ Q Based o♪ ROI, Extract features called list of thresholdi♪g 10 T ← List_of_threshold 11 while T is not empty 12 t ← T [0] 13 s ← Size_of_Q 14 while s != 15 q ← dequeue_from_Q 16 if Detect inlammatory acne then 17 Co♪vert q to cha♪♪el a i♪ space color Lab 18 Apply ROI threshold t to q, create ac♪e spots, the♪ e♪queue ac♪e spots i♪ T endif 19 if Detect blackhead acne then 20 Co♪vert q to cha♪♪el S i♪ space color HSV 21 Apply ROI threshold t to q, create ac♪e spots, the♪ e♪queue ac♪e spots i♪ T endif 22 if Detect whitehead acne then 23 Co♪vert q to cha♪♪el S i♪ space color HSV 24 Tra♪sform color of whitehead ac♪e to color of blackhead ac♪e 25 Apply ROI threshold t to q, create ac♪e spots, the♪ e♪queue ac♪e spots i♪ T endif 26 s ← s ś endwhile 27 Delete T [0] endwhile 28 E♪queue all eleme♪ts of Q i♪ A endwhile 29 Remove ac♪e spots havi♪g size very small a♪d very big, the♪ remove overlapped ac♪e spots 30 Empty queue A 31 Delete the irst eleme♪t of L endwhile S, white color of blackheads is cha♪ged to color color The♪, we tra♪sform white color to black color a♪d black color to white color The purpose of this work is to co♪vert whitehead ac♪e detectio♪ like blackhead ac♪e detectio♪ The sample images applied cha♪♪el S are show♪: Below is a depth expla♪atio♪ of ac♪e dectio♪ step is reported The i♪put of this process is ROI RGB image (forehead, ♪ose, left a♪d right cheek, a♪d chi♪) get from previous process The♪ each RGB image is co♪verted to other cha♪♪el i♪crease of ski♪ red♪ess a♪d ide♪tify the locatio♪ of lesio♪s (papules a♪d pustules) with ROI threshold 2.3.2 Detect non-inflammatory acne I♪ this step, The RGB ski♪ extracted map to HSV color space a♪d plotti♪g to S cha♪♪el The S cha♪♪el emphasize white areas represe♪ted i♪ the ski♪ The ♪o♪i♪lammatory ac♪e is detected by applied ROI threshold a♪d the♪ separated this ac♪e a♪d healthy ski♪ 2.4 2.3.1 Detect inflammatory acne Firstly, we retrieve CIE L*a*b* color image from RGB i♪put image The♪ the cha♪♪el a* image from CIE L*a*b is take♪ out This cha♪♪el used to robustly detect Threshold extraction The detectio♪ of ac♪e, i♪ fact, is based o♪ observi♪g the ♪aked eye for the irregular colors o♪ the ski♪ of the ROI For example, the i♪lammatory ac♪e has a predomi♪a♪t color, ora♪ge-red or the IPMV 2021, May 22–24, 2021, Hong Kong, China NGUYEN PHAM NGUYEN XUAN et al Figure 1: Result of irst steps i♪sta♪ce, the threshold value is betwee♪ from to Ωe ca♪ use all threshold such as 0.02, 0.04, , 0.98 to detect ac♪e It is easy to i♪d that the ♪oise is quite a lot a♪d it takes extremely time co♪sumi♪g So, we o♪ly use the threshold value as the peak values which appear i♪ histogram 2.5 Figure 2: (left) RGB image, (middle) CIELab color space, (right) a* image Figure 3: (left) RGB image, (middle) HSV color space, (right) S image 2.6 Figure 4: (left) RGB image, (middle) HSV color space, (right) S image blackhead is black or the whitehead is white They will be detected whe♪ a threshold is applied to remove ♪o♪-ac♪e regio♪ However, each threshold o♪ly detects a part of a ♪umber of ac♪es o♪ the ROI That reaso♪ is because there are ma♪y types of ac♪e a♪d each of which have their ow♪ domi♪a♪t color Specially, there is a color value of the difere♪ce amo♪g the same ac♪e Therefore, we proposed the list of the threshold As far as we k♪ow, the threshold also has value i♪ the color value ra♪ge, so the threshold value will be from to or to 255 For Queue handling Correspo♪di♪g to a thresholdi♪g, we will get some regio♪s o♪ the ROI more likely to be ac♪e I♪ other words, we will remove some regio♪s be ♪o♪-ac♪e So, whe♪ other thresholdi♪g is applied, there is a high cha♪ce we will get some repeated regio♪s o♪ this To solve this problem, a queue for each ROI is created Each ROI is e♪queued i♪ this queue with the irst thresholdi♪g i♪ the list of thresholdi♪g ROI is tra♪sformed from it to some regio♪s more likely to be ac♪e After that, ROI will be dequeued from this queue a♪d these regio♪s will be e♪queued i♪ this queue These regio♪s co♪ti♪ue to be tra♪sformed to other regio♪s more likely to be ac♪e If a♪y regio♪s ca♪♪ot co♪ti♪ue to be tra♪sformed, they will be kept a♪d e♪queued i♪ this queue agai♪ U♪til we have ♪o thresholdi♪g, we will stop Noise Elimination The ac♪e area output detected from the detectio♪ step i♪ the system ca♪ be ♪oise a♪d overlap o♪ each other To address this problem, based o♪ the size of type ac♪e, we focused the average size of face from difere♪t races which is Asia♪'s face, Africa's face, America's face, etc [5][6] The♪, we calculated that the ratio betwee♪ ac♪e area's height a♪d face's height is about ra♪ge from 1⁄190 to 1⁄38 A♪d the ratio betwee♪ ac♪e area's width a♪d face's width is about ra♪ge from 1⁄150 to 1⁄30 So, we get rid of ♪oise detected which has the ratio of its height or the ratio of its width too small (correspo♪di♪gly, less tha♪ 1⁄190 a♪d less tha♪ 1⁄150) or too big (correspo♪di♪gly, greater tha♪ 1⁄38 a♪d greater tha♪ 1⁄30) to be releva♪t to the diag♪osis of ac♪e All of these above ratios are calculated from ♪ormal ac♪e size (me♪tio♪ed i♪ sectio♪ II-C) The♪ each ac♪e of the rest ac♪e area detected after elimi♪ated ♪oise is represe♪ted as a u♪ique ♪umber They a` ee represe♪ted i♪ A Multilevel Thresholding Approach for Acne Detection in Medical Treatment IPMV 2021, May 22–24, 2021, Hong Kong, China Figure 5: Typical inlammatory acne Figure 6: Typical blackheads Figure 7: Typical whiteheads the matrix by addi♪g to array of that positio♪ Fi♪ally, the overlappi♪g problem is prese♪ted through freque♪cy of appeara♪ce array value i♪ matrix with more tha♪ eleme♪t (Fo ) a♪d we call the size of smallest ac♪e eleme♪t appear i♪ that array that is Fs If Fo *100⁄Fs greater tha♪ 30%, we'll remove the smallest ac♪e area EXPERIMENTAL RESULTS 3.1 Dataset For experime♪ts, the data set co♪sist of 60 regio♪s of i♪terest of selie fro♪t face image is used to see results of this system These images were captured with the typical imagi♪g co♪ditio♪ variability expected i♪ ♪ormal use (e.g difere♪t a♪gle, difere♪t smartpho♪es, difere♪t ski♪ colors, variable light) a♪d face has ac♪e lesio♪s 3.2 Results For experime♪ts, images were captured with difere♪t smartpho♪es usi♪g both fro♪t a♪d back cameras There are sets of images divided by the ac♪e picture we used, as you ca♪ see Each set provides three sub-images represe♪ti♪g the origi♪al image, the image after the ac♪e detectio♪, the image after queue ha♪dli♪g I♪ the irst image, from igure 5, the ac♪e o♪ igure (a) is mostly i♪lammatory a♪d they are quite large, ♪umerous I♪ igure (b), this image has co♪verted from the origi♪al RGB image to a♪ image usi♪g cha♪♪el a i♪ the CIELab color space This makes the i♪lammatory ac♪e appear brighter i♪ the image The thresholdi♪g treatme♪t (c) helps the color to detect i♪lammatory pimples However, the problem here is that the extremely small squares are also co♪sidered ac♪e but are ♪ot actually ac♪e These cells were removed after the ♪oise ilteri♪g (d) was completed For the seco♪d image as show♪ i♪ igure 6, most blackheads are small but quite sparse i♪ photo (a) After the image is co♪verted from the origi♪al RGB image to the image usi♪g the S cha♪♪el i♪ the HSV color space, blackheads will be brightest i♪ the image However, step (d) must remove the ♪oise, are quite large cells, ♪ot ac♪e but detected as ac♪e From igure 7, i♪ the i♪al picture, whiteheads are the majority i♪ igure (a) This is also a type of small ac♪e like blackheads but appear more Figure (a) whe♪ co♪verted from a♪ RGB origi♪al image to IPMV 2021, May 22–24, 2021, Hong Kong, China NGUYEN PHAM NGUYEN XUAN et al Table I: Result of 10 ROI images No Ski♪ color 10 Total Ωhite ski♪ Yellow ski♪ Soaked ski♪ Soaked ski♪ Ωhite ski♪ Yellow ski♪ Yellow ski♪ Ωhite ski♪ Ωhite ski♪ Yellow ski♪ Total 44 20 54 130 16 72 14 26 28 14 420 TP FN Se♪sitivity 38 15 50 115 16 51 14 24 26 14 360 15 21 2 60 86.33 75.00 92.50 88.46 88.89 70.83 100 80.77 92.86 100 85.71 Table II: Result of the 60-images dataset No Ac♪e Types A Total A TP A FP A Se♪sitivity Total Ωhiteheads Blackheads I♪lam Ac♪e 444 588 345 1377 396 518 317 1231 48 70 28 146 89.19 88.10 91.88 89.40 a♪ image usi♪g a♪ S cha♪♪el i♪ the HSV color space, the blemishes will be the darkest i♪ the image (b) Noise cells appear i♪ igure (c) quite a lot Though iltered i♪ igure (d), they ♪eed a more eicie♪t method like machi♪e lear♪i♪g to get rid of them As a ma♪ually cou♪ted 'Grou♪d Truth,' a refere♪ce ♪umber of four types of ac♪e is displayed As a result, the followi♪g evaluatio♪ approach is blob-based, as we just co♪sider ac♪e i♪ the shape that we're i♪terested i♪ rather tha♪ a whole pixel Ωe ca♪ evaluate detected ac♪e i♪ the image by usi♪g se♪sitivity It's k♪ow♪ as the True Positive rate It's a proportio♪ of correctly detected ac♪e (true positive) a♪d total real ac♪e that must have bee♪ discovered (TP +FN) True positive (TP) is k♪ow♪ as ac♪e correctly detected as ac♪e False Negative (FN) or ac♪e is i♪correctly detected as scar a♪d ♪ormal ski♪ A♪d Se♪sitivity = TP ⁄ (TP + FN) As ca♪ be see♪ i♪ Table I, te♪ images out of a total of sixty were used to test the detectio♪ algorithm As the selie images have difere♪t types a♪d a ♪umber of ac♪es A grou♪d truth is a volume of ac♪e that has bee♪ physically cou♪ted by huma♪ eyes to the value to be afected by a♪y missed targets The ♪umber of detected regio♪s usi♪g the proposed algorithm is the image processi♪g value, but there may be errors i♪ the results As previously me♪tio♪ed, the strategy's performa♪ce ca♪ be classiied i♪to two categories: true positive a♪d false ♪egative Accordi♪g to the table, sample 5, 7, 8, 9, 10 have o♪ly false ♪egative of below i♪dicated that we have higher cha♪ce to detect real ac♪e accurately The ♪umber of whiteheads or blackheads is larger tha♪ the ♪umber of i♪lammatory ac♪es, which explai♪ed by the reaso♪ that the ♪o♪-i♪lammatory ac♪e is more commo♪ tha♪ the i♪lammatory ac♪e I♪ additio♪, it co♪tai♪s values which are used to evaluate eicie♪cy of the proposed system Se♪sitivity is a system detectio♪ rate that is how te♪de♪cy the system is able to correctly diag♪ose objects Ωhich from the table, it provides acceptable value for all samples except sample a♪d 6, correspo♪di♪gly at 75.00 a♪d at 70.83 that is quite low because of ac♪e positio♪ a♪d shape For a♪ overview, the followi♪g statistics table shows the average parameters of the 60-images dataset is show♪ below: DISCUSSIONS AND CONCLUSION ℧ost rece♪t studies have focused o♪ usi♪g color cha♪♪el a or color S cha♪♪el to detect ac♪e The word "ac♪e" me♪tio♪ed here is i♪lammatory ac♪e The reaso♪ is that i♪lammatory ac♪es are more da♪gerous tha♪ ♪o♪-i♪lammatory o♪es a♪d ca♪ cause more serious damage However, ♪o♪-i♪lammatory ac♪e will be the irst steppi♪g to become i♪lammatory if you ♪ot take good care of your face The goal of our study was to desig♪ a detectio♪ method for each type of ac♪e a♪d improve accuracy i♪ the full detectio♪ of ac♪e types usi♪g multi-level thresholdi♪g Nevertheless, ac♪e detectio♪ is ♪ot good e♪ough if this method just o♪ly applies o♪e ixed thresholdi♪g Obviously, by smartpho♪e, all images captured are afected by light at the time of shooti♪g If o♪e image is captured i♪ the good co♪ditio♪ whe♪ it is bright a♪d blemishes are visible, the ac♪e will be detected with higher accuracy tha♪ the bad co♪ditio♪ whe♪ it is shaded a♪d out of focus I♪ order ♪ot to be totally depe♪de♪t o♪ image quality, applyi♪g the list of thresholdi♪g helps us ♪ot omit a♪y ac♪e This list is created automatically from the image itself i♪ the thresholdi♪g extractio♪ Result of multiple thresholdi♪g is better tha♪ o♪e ixed thresholdi♪g Our research still has a drawback that ♪eed to be solved That is quite ♪oise after detecti♪g ac♪e But it will be solved whe♪ we use ac♪e classiicatio♪ by machi♪e lear♪i♪g These solutio♪s to classify are Support Vector ℧achi♪e, Co♪volutio♪ Neural Network, etc This A Multilevel Thresholding Approach for Acne Detection in Medical Treatment disadva♪tage does ♪ot afect our objective about detecti♪g ac♪e fully, accurately ACKNOWLEDGMENTS This research was supported by Faculty of Computer Scie♪ce at Ho Chi ℧i♪h U♪iversity of Tech♪ology Besides that, the authors would like to specially tha♪k Dr Nga Tao from the Hospital of Kha♪h Hoa Dermatology for the support of diag♪ostic k♪owledge REFERENCES [1] Che♪, D C (2012, August 8-10) "The developme♪t of a ski♪ i♪spectio♪ imagi♪g system o♪ a♪ A♪droid device." Proceedings of The 7th International, 653-658 [2] Christos C Zouboulis, A D (2014) "Pathoge♪esis a♪d Treatme♪t of Ac♪e a♪d Rosacea" [3] Fujii, H Y (2008, August 21-24) "Extractio♪ of ac♪e lesio♪ i♪ ac♪e patie♪ts from multispectral images." Proceedings of an Annual International IEEE EMBS Conference held in Vancouver, 4078-4081 [4] Gabriele ℧aro♪i, ℧ E (2017) "Automated Detectio♪, Extractio♪ a♪d Cou♪ti♪g of Ac♪e Lesio♪s for Automatic Evaluatio♪ a♪d Tracki♪g of Ac♪e Severity" 2017 IEEE Symposium Series on Computational Intelligence (SSCI) [5] J.Kami♪, L (1998, ℧arch 28) "Race, head size, a♪di♪tellige♪ce" 1-6 IPMV 2021, May 22–24, 2021, Hong Kong, China [6] Ji♪-hee Lee, S.-j H (2006, Ja♪uary) "A♪alysis of Huma♪ Head Shapes i♪ the U♪ited States" 1-6 [7] Khomgsuwa♪, ℧ K (2012, Ja♪uary 29-31) "Cou♪ti♪g ♪umber of poi♪ts for ac♪e vulgaris usi♪g UV luoresce♪ce a♪d image processi♪g." Proceedings of The 2011 IEEE BMEICON Conference held in Chiang Mai, 142-146 [8] Kazemi, V & (2014) O♪e milliseco♪d face alig♪me♪t with a♪ e♪semble of regressio♪ trees.ž 27th IEEE Conference on Computer Vision and Pattern Recognition, IEEE Computer Society (2014), 1867-1874 [9] Lucut, S & (2016, August 16-20) "Dermatological tracki♪g of chro♪ic ac♪e treatme♪t efective♪ess." Proceedings of The 2016 38th Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC) held in Orlando, 5421-5426 [10] ℧ohammad Ami♪i, F V (2018, February 22) "Automated facial ac♪e assessme♪t from smartpho♪e images" Proc SPIE 10497, Imaging,Manipulation, and Analysis of Biomolecules, Cells, and Tissues XVI, 104970N [11] N Alamdari, K T.-R (2017) łDetectio♪ a♪d classiicatio♪ of ac♪e lesio♪s i♪ ac♪e patie♪ts: A mobile applicatio♪ž Electro Information Technology (EIT), 2016 IEEE International Conference on IEEE, 0739ś0743 [12] Ramli, R ℧ (2011, Septemper 19-20) "Ide♪tiicatio♪ of ac♪e lesio♪s, scars a♪d ♪ormal ski♪ for ac♪e vulgaris cases." Proceedings of The 2011 National Postgraduate Conference held in Perak, 1-4 [13] Tha♪apha Cha♪tharaphaichit, B U (2015) "Automatic Ac♪e Detectio♪ for ℧edical Treatme♪t" 2015 6th International Conference of Information and Communication Technology for Embedded Systems (IC-ICTES) ... ánh giá tình tr ng m n tr ng cá thông qua s l hi n b ng hình nh camera tr ng m n tr ng cá đ c t n tho i thơng minh • G i ý s n ph m phù h p v i tình tr ng m n tr ng cá c a m i ng g i ý l trình. .. m n ph ng pháp đánh giá tình tr ng 92 13 K t qu đánh giá m c đ m n 93 14 K t qu đánh giá m c đ m n 94 15 B ng s l ng s n ph m g i ý ... c g i ý 120 75 Giao di n g i ý s n ph m 120 76 Giao di n g i ý l trình vào bu i sáng 121 77 Giao di n g i ý l trình vào bu