Study Notes on Computer Vision

Chapter Overview

  • Second Domain of AI: Computer Vision

  • Learning Objectives

    • Understand the definition of Computer Vision.

    • Explore various applications of Computer Vision.

    • Understand key concepts related to Computer Vision.

Definition of Computer Vision

  • Computer Vision: A field of artificial intelligence that trains computers to interpret and understand the visual world, essentially imbuing machines with human-like visual perception.

  • It enables machines to see images or visual data and analyze them using algorithms and methodologies to understand real-world phenomena through images.

Human Vision System vs. Computer Vision System
Human Vision System

  • Comprises:

    • Eye: Acts as the sensing device capturing images of the surrounding environment.

    • Brain: Acts as the interpreting device responsible for understanding content in images (e.g., recognizing objects like bowls and fruits).

Computer Vision System

  • Defines: A segment of computer science enabling computers to 'see', identify, and process images similar to human vision, subsequently providing appropriate outputs.

  • Related to Artificial Intelligence as it necessitates interpreting visual stimuli and executing analysis or actions accordingly.

  • Operates by utilizing digital images from cameras, videos, and employing deep learning models to accurately identify and localize objects, enabling reactive capabilities.

  • The complete process involves:

    • Image Acquisition

    • Screening

    • Analysis

    • Identification

    • Information Extraction

Applications of Computer Vision

  • Applications span various fields, emerging since the concept's introduction in the 1970s. Advances in computation have made these technologies widely accessible. Examples of applications include:

    • Face Filters: Used by apps like Instagram and Snapchat to apply visual effects by recognizing facial dynamics.

    • Visual Navigation

    • Augmented Reality (AR)

    • Optical Character Recognition (OCR)

    • Panorama Stitching

    • 3D Reconstruction

    • Games

    • 3D Object Recognition

    • Surveillance

    • Segmentation

    • Human-Computer Interfaces

    • Inspection

    • Automation

Technology in Action - TikTok

  • Analyzes user-uploaded videos using natural language processing and computer vision, extracting metadata such as content descriptions, audio, and transcriptions to enhance user engagement.

Computer Vision in Retail

  • Retail rapidly integrates computer vision techniques for numerous advantages:

    • Tracking customer movements

    • Analyzing navigation behavior

    • Monitoring compliance and operations

    • Automating inventory management and payments

  • Example: Amazon Go stores utilize vision technology to allow shopping without traditional checkout processes.

Medical Imaging Applications

  • Computer vision assists in medical fields by providing diagnostic tools:

    • Image Creation and Analysis: Creates and interprets medical images, often rendering 2D scans into 3D models for deeper understanding.

    • Microsoft InnerEye: Uses 3D imaging algorithms to identify tumors and present anomalies more accurately.

    • MaxQ AI: Identifies rare abnormalities in brain scans through machine vision.

Image Fundamentals
Basic Concept of Pixels

  • Pixel: The smallest unit of a digital image, each representing information such as color, essential in forming images on displays.

  • Organized in a 2D grid where each pixel's arrangement creates the overall image.

Resolution

  • Resolution: Defined as the total pixel count in an image.

    • Can be expressed as dimensions (M×N), where M is the number of rows and N is the number of columns, for instance, a resolution of 1280×1024.

    • A camera with a resolution of 5 megapixels means it captures images with a total of five million pixels.

  • High pixel resolution corresponds to higher image quality (example: 2500 x 3192 pixels gives approximately 8 megapixels).

Pixel Value and Color Representation

  • Each pixel contains color value ranging from 0 (black) to 255 (white).

  • RGB color model utilizes three color channels (Red, Green, Blue) to represent varied colors by combining intensities of each channel.

    • Each of the primary colors can have values from 0 to 255, determining color representation in images.

Grayscale Images

  • Grayscale: Images composed of shades from black to white, lacking color. Ranges represented by values 0 (black) to 255 (white) signify depth of gray.

    • Storing grayscale images uses a single intensity level instead of three for RGB, with an array format Height x Width defining the image size.

Image Processing Concepts

  • Image Processing: Involves enhancing or extracting information from images through mathematical operations. Applications span sectors like computer vision, feature extraction, and optical character recognition.

  • Features critical in identifying information relevant to tasks: points, edges, corners, and objects aid computational tasks leveraging image data effectively.

Conclusion

  • Computer Vision represents a fascinating intersection of image processing, machine learning, and practical applications across various fields, showcasing its critical role in both technological advancement and real-world usability.