M4

Detecting Edges with Canny

• Representing an image by its edges reduces the amount of data and retains the image information.
• The shape of an image is enclosed in edges.

Edge Detection

• Sobel filtering technique detects edges.
• Canny edge detection produces higher accuracy in detecting edges compared to the Sobel algorithm.
• Sobel is a good choice for quick and general edge detection.
• Canny is a better option for more precise and clean edge detection, especially in noisy images.

Canny Edge Detector

• The Canny edge detection algorithm applies a Gaussian filter to remove noise from the image.
• The sigma attribute in the Canny function adjusts the filter's intensity
• Setting a lower sigma value, such as 0.5, in the Gaussian filter for the Canny edge detection algorithm results in less filtering, and allowing more edges to be detected.
• A higher sigma value removes more noise, resulting in fewer detected edges.
• The default value for sigma is 1.

Corner Detection

• Corner detection extracts certain types of features and infers the contents of an image.
• Corner detection is frequently used in motion detection, image registration, video tracking, panorama stitching, 3D modelling, and object recognition.
• A corner is the intersection of two edges.
• Corners can also be a junction of contours.

Points of Interest

• Robust features in an image that are insensitive to changes in rotation, translation, brightness, and scale are known as points of interest.
• Corner detection identifies these interest points, focusing on corners as a key type of feature alongside edges.

Matching corners

• Detecting corners as interest points can match objects across images from different perspectives.
• Corners identified in an original image can be matched to those in a downscaled version of the same image.
• This demonstrates the technique's utility in recognizing and aligning features across varying scales.

Harris Corner Detector

• The Harris Corner Detector is a corner detection operator in computer vision that identifies regions where the intensity shows significant changes in multiple directions, indicating the presence of corners.
• The corner_harris function computes the Harris response for each pixel, producing a 2D array (measure_image) where each value represents the "corner strength" at that pixel location.
• Higher values in measure_image indicate areas that are more likely to be corners.
• The corner_peaks algorithm identifies points in the image that have strong corner responses and are at least min_distance pixels apart.
• The output of the corner_peaks algorithm is a 2D array where each row represents the coordinates of one detected corner point.

Face Detection Use Cases

• Face detection technology is used for filters, autofocus, recommendations, blur for privacy protection and to recognize emotions later on.

Detecting Faces with Scikit-Image

• Scikit-image can detect faces using a machine learning classifier
• Import the classifier class from skimage.feature
• Load the trained file from the module root
• Initialize the detector cascade
• The Cascade class is imported from the feature module to use the face detector.
• This detection framework needs an xml file, from which the trained data can be read.

Cascade Classifier

• Each stage is a filter that decides whether a region of the image likely contains the object of interest (such as a face or a specific pattern).
• The detect_multi_scale method from the Cascade class can be used.
• This method moves a window across the image to find areas resembling a human face.
• Searching happens on multiple scales to spot the small or far-away faces.
• the window will have a minimum size
• a maximum size to also find the larger faces in the image
• The detect_multi_scale method for face detection adjusts the window size at each step by a scale factor and moves the window by a step ratio.
• 1 indicates an exhaustive, though slow, search,
• Values between 1 and 1.5 balance speed and accuracy well.
• The min and max sizes define the range of detection.
• The detector returns a dictionary of coordinates:
  • r represents the row position of the top-left corner of the detected window
  • c is the column position pixel
  • width = width of detected window
  • height= the height of the detected window.

Applications

• Applications for processing the images:
  • Turning to grayscale before detecting edges/corners
  • Reducing noise and restoring images
  • Blurring faces detected
  • Approximation of objects' sizes

Privacy Protection

• A privacy protection case can be implemented by detecting faces and then anonymizing them.
• Cascade of classifiers and gaussian filter can be imported.
• Use gaussian filter to extract face.
• Result is an image in which personal data is anonymized.

OpenCV

• OpenCV (Open Source Computer Vision Library) is a comprehensive library primarily aimed at real-time computer vision.
• OpenCV offers additional functionalities and optimizations.
• Features include:
  • Real-Time Processing
  • Video Analysis
  • Machine Learning and Deep Learning
  • Camera Calibration and 3D Reconstruction
  • Motion Analysis
  • GUI Features
  • Wide Range of Algorithms for Object Detection
  • Integration with Other Technologies

Scikit Image vs OpenCV

• Scikit-image is mainly used for:
  • Image processing for scientific research
  • Suitable for offline analysis
  • Limited to frame extraction
  • Limited built-in support
  • Basic 3D image processing
  • Minimal GUI Capabilities
  • Extensive algorithms for image analysis
  • Python
  • Limited Hardware Acceleration
• OpenCV:
  • Real-time computer vision
  • Optimized for real-time applications
  • Comprehensive video processing capabilities
  • Extensive integration and support\
  • Advanced 3D reconstruction and camera calibration
  • Advanced GUI features for real-time interaction
  • Extensive algorithms, including ML and object detection
  • C++, Python, Java, and more
  • Supports CUDA, OpenCL, and more