MINISTRY OF EDUCATION AND TRAINING HO CHI MINH CITY UNIVERSITY OF TECHNOLOGY AND EDUCATION GRADUATION THESIS MAJOR: COMPUTER ENGINEERING TECHNOLOGY OPTIMIZE DEFOG PROCESSING FOR OBJECT RECOGNITION APPLICATIONS INSTRUCTOR: PHAN VAN KHOA PHD. STUDENT: NGUYEN HUY HOANG PHAN MINH NHAT Ho Chi Minh city, July 2024 HCMC UNIVERSITY OF TECHNOLOGY AND EDUCATION FALCUTY OF INTERNATIONAL EDUCATION GRADUATION PROJECT OPTIMIZE DEFOG PROCESSING FOR OBJECT RECOGNITION APPLICATIONS NGUYEN HUY HOANG Student ID: 20119002 PHAN MINH NHAT Student ID: 20119147 Major: COMPUTER ENGINEERING TECHNOLOGY Advisor: PHAM VAN KHOA, PhD. Ho Chi Minh City, July 2024 HCMC UNIVERSITY OF TECHNOLOGY AND EDUCATION FALCUTY OF INTERNATIONAL EDUCATION GRADUATION PROJECT OPTIMIZE DEFOG PROCESSING FOR OBJECT RECOGNITION APPLICATIONS NGUYEN HUY HOANG Student ID: 20119002 PHAN MINH NHAT Student ID: 20119147 Major: COMPUTER ENGINEERING TECHNOLOGY Advisor: PHAM VAN KHOA, PhD. Ho Chi Minh City, July 2024 THE SOCIALIST REPUBLIC OF VIETNAM Independence – Freedom– Happiness -------- Ho Chi Minh City, July 05, 2024 GRADUATION PROJECT ASSIGNMENT Student name: Phan Minh Nhat Student ID: 20119147 Student name: Nguyen Huy Hoang Student ID: 20119002 Major: COMPUTER ENGINEERING TECHNOLOGY Class: 20119CLA1,2 Advisor: PhD.
Pham Van Khoa Phone number: Date of assignment: Date of submission: 1. Project title: OPTIMIZE DEFOG PROCESSING FOR OBJECT RECOGNITION APPLICATIONS 2. Initial materials provided by the advisor: 3. Content of the project: This project focuses on improving defogging procedures with the primary goal is to enhance the efficiency of defogging for object recognition in real – world scenarios.
By integrating innovative techniques, the proposed method effectively handles varying haze densities, preserves image details, and enhances visual quality. Experimental results demonstrate that this approach outperforms existing dehazing algorithms, offering a robust solution for applications requiring reliable performance in adverse weather conditions. Final product: CHAIR OF THE PROGRAM ADVISOR (Sign with full name) (Sign with full name) I THE SOCIALIST REPUBLIC OF VIETNAM Independence – Freedom– Happiness -------- Ho Chi Minh City, July 05, 2024 ADVISOR’S EVALUATION SHEET Student name: Phan Minh Nhat Student ID: 20119147 Student name: Nguyen Huy Hoang Student ID: 20119002 Major: COMPUTER ENGINEERING TECHNOLOGY Project title: OPTIMIZE DEFOG PROCESSING FOR OBJECT RECOGNITION APPLICATIONS Advisor: PhD. Pham Van Khoa EVALUATION 1.
Content of the project:. Approval for oral defense? (Approved or denied) 5. Mark: - in words: Ho Chi Minh City, month day year ADVISOR (Sign with full name) II THE SOCIALIST REPUBLIC OF VIETNAM Independence – Freedom– Happiness -------- Ho Chi Minh City, July 05, 2024 PRE-DEFENSE EVALUATION SHEET Student name: Phan Minh Nhat Student ID: 20119147 Student name: Nguyen Huy Hoang Student ID: 20119002 Major: COMPUTER ENGINEERING TECHNOLOGY Project title: OPTIMIZE DEFOG PROCESSING FOR OBJECT RECOGNITION APPLICATIONS Name of Reviewer:. Content and workload of the project.
Approval for oral defense? (Approved or denied) .) Ho Chi Minh City, month day , year REVIEWER (Sign with full name) III THE SOCIALIST REPUBLIC OF VIETNAM Independence – Freedom– Happiness -------- Ho Chi Minh City, July 05, 2024 EVALUATION SHEET OF DEFENSE COMMITTEE MEMBER Student name: Phan Minh Nhat Student ID: 20119147 Student name: Nguyen Huy Hoang Student ID: 20119002 Major: COMPUTER ENGINEERING TECHNOLOGY Project title: OPTIMIZE DEFOG PROCESSING FOR OBJECT RECOGNITION APPLICATIONS Name of Defense Committee Member:. Content and workload of the project .) Ho Chi Minh City, month day , year COMMITTEE MEMBER (Sign with full name) IV DISCLAIMER This graduation thesis was completed as part of the requirements for the Bachelor's degree in Computer Engineering Technology at the Ho Chi Minh City University of Technology and Education. The content presented in this thesis is the result of our independent research and development, conducted under the supervision of Pham Van Khoa, PhD. We have made every effort to ensure the accuracy and completeness of the information included.
All sources of information, data, and research used in this thesis have been properly cited and acknowledged in accordance with academic standards. The findings, interpretations, and conclusions expressed in this thesis are those of the authors and do not necessarily reflect the views or opinions of the Ho Chi Minh City University of Technology and Education, its faculty, or any other associated entities. This text is intended for educational and informative purposes only and does not constitute professional advice. The author does not guarantee the completeness, correctness, reliability, appropriateness, or availability of the information, goods, services, or visuals included in this thesis for any purpose.
Any reliance on such material is solely at your own risk. The university and the supervising faculty are not responsible for any errors or omissions in this work or for the consequences of any actions taken based on the information provided herein. The responsibility for the content and the outcomes of this research rests solely with the authors. This thesis is intended as a contribution to the field of image processing and object recognition.
It should be noted that the methodologies and results presented are based on specific experimental conditions and parameters and may not be directly applicable to all real-world scenarios. Further verification, validation, and possibly adaptation of the proposed methods are required before they can be implemented in practical applications. The authors recommend that readers and future researchers exercise critical judgment and conduct additional studies to validate the findings and ensure their applicability in diverse contexts. The author maintains the right to make changes to the content without prior notification and is not responsible for any mistakes or omissions in the content of this thesis.
V ACKNOWLEDGEMENT First of all, The team would like to express our sincere gratitude to our supervisor, Dr. Pham Van Khoa, for his unwavering guidance, support, and mentorship throughout the course of this research project. His experience, important insights, and ongoing support were critical in influencing the creation and successful completion of my thesis. The team is particularly appreciative to Dr.
Pham Van Khoa for providing us with the required resources and facilities to carry out this research. We would also want to express our sincere gratitude to my senior, Pham Nguyen Hoang Hai, for his invaluable advice, support, and guidance along this journey. The team would also like to thank our colleague Tran Tuan Kiet, who collaborated with us on a different thesis project. His efforts and ideas have had an indirect and positive impact on the work presented here.
The team is extremely grateful to Dr. Pham Van Khoa for providing us with a stimulating intellectual atmosphere that was critical to the success of our attempt. A particular thank you to my fellow graduate students and coworkers, whose camaraderie, thought-provoking talks, and readiness to provide a helping hand have made this trip both rewarding and engaging. Last but not least, the team would like to thank our family and friends for their unconditional love, encouragement, and steadfast support during this journey.
Without them, this thesis could not have been completed. VI Contents CHAPTER 1 INTRODUCTION. SCOPES OF TOPIC. SUBJECT AND SCOPE OF THE RESEARCH.
4 CHAPTER 2 BACKGROUND KNOWLEDGE. PHYSICAL SCATTERING MODEL. DARK CHANNEL PRIOR. IMAGE ENHANCEMENT METHODS.
FOG REMOVAL EVALUATION. Human – subjective scoring. 15 CHAPTER 3 DESIGN AND IMPLEMENTATION. SOFTWARE FLOW CHART.
DETAILS OF DARK CHANNEL PRIOR. Dark Channel Normalization. Dark Channel Extraction. Transmission Map Estimation using Anisotropic Diffusion.
Apdaptive Fog Factor. Cumulative Distribution Function. Constrast Stretching Process. 37 VII CHAPTER 4 RESULT.
THE RESULTS OF THE SYSTEM. 48 CHAPTER 5 CONCLUSION AND FUTURE WORK. 56 VIII TABLE OF FIGURES Figure 2.1-1 Physical Scattering model.1-2 Variable in the haze imaging equation .1-3 Light and scattering types .2-2 Output estimation using different 𝝎 .2-3 Non – sky image quality after recovery using transmission map .2-4 Backlight image after recovery using transmission map .2-5 SkyScene image after recovery using transmission map .1-1 Sample of comparing foggy contexts .1-2 Fog removal comparision on different situations .1-3 Comparision of Mean BRISQUE for various image fog removal methods .1-4 Comparision of mean PSNR values for image fog removal methods .1-5 Comparision of mean PSNR values for image fog removal methods .2-1 Time complexity comparision of all methods .2-2 Object Detection using YOLOv4 on original image (left) and defogged image (right). 51 IX ABSTRACT Image dehazing plays a crucial role in enhancing object recognition systems, particularly in adverse weather conditions.
This paper introduces a novel hybrid algorithm designed to effectively remove haze from images, thereby improving the clarity and detail necessary for accurate object recognition. Our proposed method integrates an Adaptive Dark Channel Prior (DCP) with anisotropic diffusion for transmission map refinement, dynamic atmospheric light estimation, and advanced post- processing techniques such as histogram normalization and contrast stretching. The adaptive mechanisms in our method, including variable window sizes and dynamic fog factor adjustments, allow it to handle a wide range of haze densities more efficiently than traditional methods. Additionally, the use of anisotropic diffusion helps preserve edge details and reduce noise, addressing common shortcomings in existing dehazing algorithms.
Experimental results demonstrate that our proposed algorithm outperforms conventional techniques, including standard DCP and various deep learning-based approaches, in terms of both visual quality and object recognition performance. Key metrics such as Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), and object recognition accuracy are significantly improved, validating the effectiveness of our approach. INTRODUCTION Fog is a common meteorological phenomenon, occurring when small water droplets suspended in the air reduce visibility and obscure scenery. In modern technological applications, especially in the field of object recognition, fog causes significant difficulties.
For example, in self-driving vehicle systems, the ability to accurately detect and recognize objects such as pedestrians, other vehicles, and traffic signs is extremely important to ensure safety. Fog degrades image quality, causing blurriness and reduced contrast, thereby reducing the ability of these systems to accurately detect and identify. In security surveillance systems, cameras installed outdoors often face changing weather conditions, including fog. Blurred images caused by fog can reduce the effectiveness of image analysis systems, making it difficult to identify objects or detect suspicious activity.
For outdoor robot applications, visibility in foggy conditions is important to ensure robots can move safely and perform tasks effectively. To solve this problem, dehazing algorithms have been developed that aim to improve image quality in foggy conditions. Dehazing is the process of removing or minimizing the effects caused by fog on images, thereby enhancing image clarity and contrast. Dehazing methods are often based on mathematical models and image processing techniques to reconstruct the original image from an image blurred by fog.
However, developing effective dehazing algorithms is not a simple task. Fog is caused by the scattering and absorption of light, making modeling these effects complex. An effective dehazing algorithm needs to address challenges such as: balancing between enhancing clarity and maintaining natural image details, ensuring feasibility in real-time processing, and integrates seamlessly with existing object recognition systems. In addition, differences in fog conditions in reality also place high demands on the flexibility of the algorithm.
For example, fog may be denser in the early morning or thinner in the afternoon and may vary by geographical area. Therefore, an optimal dehazing algorithm needs to be able to adapt to diverse fog conditions. This research focuses on optimizing dehazing algorithms to improve object recognition in foggy conditions. The goal is to develop an algorithm that is not only effective in removing fog but can also perform well in real-time applications.
By improving image quality, this research hopes to contribute to improving the accuracy and reliability of object recognition systems, thereby expanding their application range in real-time conditions. PROJECT OBJECTIVES Developing dehazing algorithms – Dark Channel Prior Algorithm 1 Improving image after processing Evaluating performance of algorithms Reviewing impact of algorithms on different scenarios 1. SCOPES OF TOPIC First, it is necessary to design an effective algorithm capable of significantly improving the clarity and contrast of images affected by fog. To achieve this, the algorithm must accurately analyze and model the effects caused by fog such as light scattering and absorption.