"cnn" 태그로 연결된 11개 게시물개의 게시물이 있습니다.

Type of ML Systems​

• Supervised Learning
• Unsupervised Learning
• Semi-supervised Learning
• Reinforcement Learning
• Batch and Online Learning
  • Whether system can learn on the fly
• Instance-based and Model-based Learning
  • Comparing data points or detect patterns in training data to build a predictive model

Evaluation Metrics​

Interesction over Union (IoU)​

• a metric used for the evaluation of object detection detectors
• how good is the predicted bounding box for an object detected closely matches.

$IoU = \frac{Area\ of\ Overlap}{Area\ of\ Union}$

Digital Image​

• made of picture elements (pixels)
• an array or a matrix of Pixels arranges in columns and rows
• Each Pixel has its own intesity value, or brightness
• Intensity values in digital images are defined by bits.
  • 8 bits image = 256 (2^8) intensity values (0-255)
• Black & White images have a single 8-bits intensity range.

Image Processing Basics​

• Image dimension = $5 \times 5 \times 3$
• Number of Channels = 3 (Red, Green, Blue)
• So, 24-bit color dpeth (8 bits per channel)
  • Each pixel has 3 intensity values (R, G, B) each in the range of 0-255
  • Total number of possible colors = $256^3$ (16,777,216)

Image Processing Types​

• Image Enhancement
• Image Restoration
• Image Segmentation
• Image Recognition & Classification
• Image Compression
• Image Transformation
• Image Filtering
• Morphological Processing
• Color Image processing
• 3D Image Processing

Image Thresholding (Segmentation)​

• Easist method for image segmentation
• Converts gray-scale image into a binary image
  • $f(x,y) < \text{Threshold} \Rightarrow 0$ (black)
  • $f(x,y) \geq \text{Threshold} \Rightarrow 255$ (white)

Image Thresholding methods​

• Histogram shape: Peaks, valleys, and curvature of the histogram are analyzed to determine the optimal threshold value.
• Clustering based: The Otsu method, good for bimodal distribution
• Adaptive thresholding: Instade of a single threshold, have thresholds for different regions in the image.

Image Transformation​

• Rotation
• Translation
• Uniform Scaling
• Non-Uniform Scaling
• Reflection
• Shearing

Edge Detection (Image Filtering)​

• Edge?
  • The points/pixels in an image where brightness/intensities changes sharply.
  • A simple and fundamental tools in image processing and computer vision, useful in feature detection/extraciton
• Canny Edge detection
• Sobel Edge detection
• How to detect edges?

Prewitt Filters​

• Vertical Edge detector: $\begin{bmatrix} 1 & 0 & -1 \\ 1 & 0 & -1 \\ 1 & 0 & -1 \end{bmatrix}$
• Horizontal Edge detetor: $\begin{bmatrix} 1 & 1 & 1 \\ 0 & 0 & 0 \\ -1 & -1 & -1 \end{bmatrix}$

Sobel Filters​

• Vertical Edge detector: $\begin{bmatrix} 1 & 0 & -1 \\ 2 & 0 & -2 \\ 1 & 0 & -1 \end{bmatrix}$
• Horizontal Edge detetor: $\begin{bmatrix} 1 & 2 & 1 \\ 0 & 0 & 0 \\ -1 & -2 & -1 \end{bmatrix}$

Morphological Processing​

• Dilation: Shrinks the image pixels from the boundaries of objects in an image, making them thicker.
• Erosion: Adds pixels to the boundaries of objects in an image, making them thinner.