Computer Vision Flashcards
(31 cards)
What is a digital image fundamentally?
An array of pixels, where each pixel has a numeric value
What is the typical value range for grayscale images?
0 to 255
What do color images consist of?
Three RGB channels (red, green, blue)
What does most image processing involve?
The application of filters
What are filters used for in image processing?
To modify pixel values and create visual effects
Give three examples of visual effects that filters can create
Blurring, sharpening, or color inversion
What is the result of applying a filter called?
A filtered image
What do filters consist of?
Kernels, which are arrays of weighted values
How is a filter kernel applied to an image?
The filter kernel is convolved (passed across and down) over the image
How does convolutional filtering calculate new values?
By taking the weighted sum of original pixel values multiplied by the corresponding kernel weights
What is the process of applying kernels to images called?
Convolutional filtering
Give an example of a specific filter and what it does
A Laplace filter can highlight the edges of shapes in an image
What does CNN stand for?
Convolutional Neural Networks
What are CNNs commonly used for?
Computer vision tasks as a machine learning model architecture
How do CNNs use filters?
CNNs use filters to extract numerical feature maps from images
What happens to feature maps in CNNs?
Feature maps are fed into deep learning models as input (X values) to generate predictions or labels
How are filter kernels initially defined in CNN training?
Using randomly generated weight values
What happens to filter weights during CNN training?
They are adjusted to improve accuracy based on model predictions evaluated against known labeled values
Give an example of labeled values used in CNN training
Whether an image is an apple or orange
What does the iterative weight adjustment process help the model learn?
The best features to identify different kinds of objects
Why might relationships be too complex for basic algorithms in deep learning?
Because deep learning involves neural networks with complex relationships
What architecture is used in Image Analysis 4.0?
Transformer architecture
What advantage does Transformer architecture offer?
Higher performance with less training data
