Computer Vision and Recognition Systems

Research Innovations and Trends

Chiranji Lal Chowdhary, G. Thippa Reddy, B. D. Parameshachari

• Publisher: Apple Academic Press
• Year: 2022
• Print ISBN: 9781774630150
• eBook ISBN: 9781000401028
• Edition: 1
• Topic: Computer Science
• Subtopic: Software Development
• Index: Computer Science

This cutting-edge volume focuses on how artificial intelligence can be used to give computers the ability to imitate human sight. With contributions from researchers in diverse countries, including Thailand, Spain, Japan, Turkey, Australia, and India, the book explains the essential modules that are necessary for comprehending artificial intelligence experiences to provide machines with the power of vision. The volume also presents innovative research developments, applications, and current trends in the field. The chapters cover such topics as visual quality improvement, Parkinson's disease diagnosis, hypertensive retinopathy detection through retinal fundus, big image data processing, N-grams for image classification, medical brain images, chatbot applications, credit score improvisation, vision-based vehicle lane detection, damaged vehicle parts recognition, partial image encryption of medical images, and image synthesis. The chapter authors show different approaches to computer vision, image processing, and frameworks for machine learning to build automated and stable applications. Deep learning is included for making immersive application-based systems, pattern recognition, and biometric systems. The book also considers efficiency and comparison at various levels of using algorithms for real-time applications, processes, and analysis.

CHAPTER 1 Visual Quality Improvement Using Single Image Defogging Technique

PRITAM VERMA and VIJAY KUMAR*

Department of Computer Science and Engg., National Institute of Technology, Hamirpur, Himachal Pradesh, India

* Corresponding author. E-mail: [email protected]

ABSTRACT

In the wintry weather period, haze is the prime confront during driving. It eliminates the visibility of an image. Fog removal techniques are required to improve the visibility level of the image. In this chapter, a hybrid approach is implemented for fog removal. The expected approach utilizes the basic concepts of Dark Channel Prior and Bright Channel Prior. Apart from this, order statistic filter would use to refine the transmission map. The bright channel prior to boundary constraints would use to restore the edges. The proposed technique has been compared with existing techniques over a set of well-known foggy images. The proposed approach outperforms the predefined techniques in terms of average gradient and percentage of saturated pixels.

1.1 INTRODUCTION

Additional climate-related incidents are happened due to fog. In the year 2016, around 9000 peoples were died due to intense fog. The visual quality of images² is ruined due to the being there of dust, smoke, etc. Differences between fog, haze, and rain are described in Table 1.1.¹⁵ The core cause for fog in the environment due to water droplets suspension.^1,3 The water droplets are the reason for consumption and dispersion. When the light comes toward the camera or the viewer is incapacitated due to scattering through droplets and distort the visual quality of the image.^4,6,7 To conquer this problem, some sophisticated systems have been developing to maximize visibility during restraining the strong and dazzling light for oncoming vehicles.¹² For the recognition of fog the motor vehicle detection system was developed^8,13,21 but the main tribulations would have occurred that could not be able to remove the sky visibility. The automatic fog detection could detect only daytime fog but it would not able to detect the nighttime fog. To conquer this problem, computer vision techniques have been started to use.^11,14 These techniques also helped to cut down the operating cost and accommodated a better visual system.^10,25, He et al.¹⁶ planned a Dark Channel Prior (DCP) that would have utilized image pixels with low-intensity value in at least one of the color channels. Nevertheless, this value could be lessened in contrast due to additive air light. DCP commonly use to evaluate the transmission map and atmosphere shroud.9,20

TABLE 1.1 Weather Conditions and the Corresponding Particle Size.

	Condition	Type	Radius (in µm)	Concentration (cm^−3)
	Fog	Water droplet	1–10	100–10
	Haze	Aerosol	10^−2–1	10^3–10
	Rain	Water droplet	10^2–10^4	10^−2–10^−5
	Cloud	Water droplet	1–10	300–10

Fattal¹⁸ described the local color line prior to re-establish hazy images. Nandal and Kumar (2018) proposed a novel image defogged model that would use fractional-order anisotropic diffusion. They would have used the air light map that would have been evaluated from the hazy model as the picture in the anisotropic dissemination development. However, it went through halo artifacts. To reduce this problem,¹⁹ implemented a technology that would use improved DCP and contrast adaptive histogram equalization that would able to remove the halo artifact with a new median operator in the DCP. They would use a guided filt...

Table of contents

• Cover Page
• Half-Title Page
• Title Page
• Copyright Page
• About the Editors
• Table of Contents
• Contributors
• Abbreviations
• Preface
• 1. Visual Quality Improvement Using Single Image Defogging Technique
• 2. A Comparative Study of Machine Learning Algorithms in Parkinson’s Disease Diagnosis: A Review
• 3. Machine Learning Algorithms for Hypertensive Retinopathy Detection through Retinal Fundus Images
• 4. Big Image Data Processing: Methods, Technologies, and Implementation Issues
• 5. N-grams for Image Classification and Retrieval
• 6. A Survey on Evolutionary Algorithms for Medical Brain Images
• 7. Chatbot Application with Scene Graph in Thai Language
• 8. Credit Score Improvisation through Automating the Extraction of Sentiment from Reviews
• 9. Vision-Based Lane and Vehicle Detection: A First Step Toward Autonomous Unmanned Vehicle
• 10. Damaged Vehicle Parts Recognition Using Capsule Neural Network
• 11. Partial Image Encryption of Medical Images Based on Various Permutation Techniques
• 12. Image Synthesis with Generative Adversarial Networks (GAN)
• Index