COMPUTER VISION

workshop
Presented by Rania SAOUD
 Fundamentals of computer vision
& image processing
● Computer Vision:
○1. Gives computers the ability to see and understand images and videos like humans.
○2. Involves recognizing and comprehending the content of images or videos.
○3. Provides computers with "eyes" to perceive visual data.

● Image Processing:
○1. Advanced editing of images beyond basic adjustments.
○2. Enhances or alters images for clarity, size, or color.
○3. Similar to editing photos on a phone but with more sophisticated tools.

● Working Together:
1. Image processing prepares images for analysis in computer vision.
○

2. Computer vision interprets and identifies objects or faces within images.

○

3. Both are essential components for tasks like object recognition and scene understanding.
○
 Tools & Libraries

1. OpenCV:
- Handles tasks from basic operations like resizing to complex tasks like face recognition.

2. TensorFlow and PyTorch:

- Enable computers to learn from images.
- Ideal for enabling computers to not just see but also understand and learn from visual
data.

3. PIL/Pillow:
- Go-to library for opening, manipulating, and saving various image file formats in Python.
Pixels, Resolution & Color Models
● Pixels are tiny squares that create an image

● Resolution is how many pixels are packed together. More pixels mean more detail.

● Color Models :

○ RGB (Red, Green, Blue) : Combining different amounts of red, green, and blue gives us all the colors
we see in digital images. It's used in screens.

○ HSV (Hue, Saturation, Value) ; This model is about colors, their brightness, and how intense they are.
Hue is the color itself, saturation is how vivid the color is, and value is how light or dark it is.
Pixels, Resolution & Color Models

● Pixels are tiny squares that create an image

Pixels, Resolution & Color Models

● Resolution is how many pixels are packed together. More pixels mean more detail.
Pixels, Resolution & Color Models

○ RGB (Red, Green, Blue) : Combining different amounts of red, green, and blue gives us all the
colors we see in digital images. It's used in screens.
 Reading, Displaying & Writing

- Reading an Image:

- Similar to opening a photo album to find a picture.

- The computer or phone reads the image file, allowing it to be processed within a program.

- Displaying an Image:

- Like looking at a picture in your album.

- The program presents the image on the screen for visualization and analysis.

- Writing (Saving) an Image:

- Equivalent to putting a photo back in the album after editing.

- Choosing the format (box) for saving based on desired preservation or sharing requirements.
 Image Manipulation

- Resizing :
- Changing the size of a picture to fit a specific location or purpose.
- It's akin to zooming in or out, but permanently altering the dimensions.

- Cropping :
- Allows removing unwanted parts of an image, such as excess sky or photobombers.
- Focuses on the essential elements by eliminating distractions or irrelevant details.

- Rotation :
- Involves adjusting the orientation of an image, correcting sideways or upside-down views.
- Ensures the image is correctly oriented for optimal viewing, akin to adjusting a picture frame.
 Resizing

Importance :
- Resizing allows standardizing the size of images within a dataset, which is crucial for consistency in
model training.
- It helps in optimizing computational resources by reducing the image resolution while preserving essential
features, thus speeding up processing and improving efficiency.
 Cropping

Importance :
- Cropping enables focusing on specific regions of interest within an image, eliminating irrelevant
background or noise.
- It enhances the performance of object detection, recognition, and segmentation algorithms by removing
clutter and emphasizing relevant features, leading to more accurate results.
 Rotation

Importance :
- Rotation corrects the orientation of images that may be captured in different orientations, ensuring
uniformity in the dataset.
- It improves the robustness and accuracy of object detection, classification, and localization models by
presenting images consistently, irrespective of their original orientation, thus facilitating better
generalization and performance.
➔ RGB to grayscale
➔ RGB to HSV
 Image Filtering: Smoothing, Sharpening Filters
➔ Smoothing :
◆ Noise Reduction: Helps in reducing noise or unwanted variations
in pixel values.

◆ Image Enhancement: Improves overall image quality, making it

more pleasing to the eye.

◆ Preprocessing for Analysis: It’s crucial for preprocessing images

before further analysis, such as object detection or segmentation.
It helps in removing distractions and focusing on relevant
features.

➔ Sharpening :
◆ Detail Enhancement: Enhance the clarity and definition of edges
and fine details in an image.

◆ Feature Extraction: Aid in extracting important features and

improving the accuracy of algorithms for tasks like object
recognition or classification.

◆ Correcting Blurriness: Help in correcting slight blurriness or

softness in images caused by camera shake or lens imperfections,
leading to sharper and more professional-looking results.
 Edge detection

It looks at your picture and find where the edges of

objects are. It's like drawing lines around everything
to make it stand out more.

- Sobel Algorithm: This is a way of finding edges

that's pretty straightforward. It looks at the colors
and brightness to find where things change a lot,
which usually means there's an edge.

- Canny Algorithm: This is a more advanced way of

finding edges. It's really good at ignoring noise
(random changes in color or brightness that aren't
important) and finding the really important edges.
 Feature Detection & Matching

- Feature Detection:

- Finding landmarks or distinctive features in images.

- Algorithms like SIFT, VLAD, SURF help computers identify unique
features that aid in image recognition.

- Matching:

- Involves connecting identified landmarks or features across

different images.
- Enables recognition of the same object or scene in multiple images,
even with variations in angle or distance.
- Useful for tasks such as creating 3D models from photos or assisting
in robotic navigation.
Feature Detection & Matching
 Object detection
❏ Computer-based spotting of objects within an image.

❏ Identifies and labels objects like "dog," "car," or "tree."

❏ Focuses on not just detecting objects but also understanding their

identity.
 Face recognition
❏ Comparable to recognizing familiar faces among a crowd.

❏ Utilizes facial recognition technology to identify or verify individuals

from photos or videos.

❏ Examines unique facial features (e.g., eye distance, nose shape) for
differentiation between individuals.
 How it works ?

The model employs a multi-step process for object detection and identification. Initially, it
extracts relevant features from the input image to discern objects from the background.
Subsequently, it localizes these objects by predicting bounding boxes that outline their spatial
extent within the image. Through classification, the model assigns labels to the detected
regions, indicating the type of object present, such as "car" or "person." Post-processing
techniques are then applied to refine predictions and eliminate false positives, ensuring
accurate localization and classification. In certain applications like facial recognition, the
model may further identify specific attributes or instances within detected objects, such as
recognizing individuals by matching detected faces with known identities. Through this
iterative process of feature extraction, localization, classification, and post-processing, the
model effectively detects objects and identifies them within images or videos.
 Data augmentation & Data cleaning in
images
➔ Data Augmentation in Computer Vision:

- Techniques include flipping horizontally or vertically, rotating, adjusting

brightness/contrast, zooming, and cropping.

- Increases the diversity of data for better model learning and performance when
encountering new images.

➔ Data Cleaning in Computer Vision:

- Tasks involve removing duplicates, deleting low-quality images, correcting mislabeled

images, and ensuring labeling consistency.

- Enhances the model's learning by providing high-quality, accurate data, reducing

confusion, and improving recognition and understanding of images.