Comvis Core is a project that explores the core techniques of Computer Vision (CV), including image processing, feature extraction, and pattern recognition. It demonstrates the foundational steps required to enable a machine to interpret and analyze visual data.
This project was developed as a qualification case to teach Computer Vision as a laboratory assistant. The code showcases how to transform raw pixel data into meaningful information using classic CV algorithms. By working through these cases, I gained a deeper understanding of image manipulation, spatial filtering, feature detection, and classical machine learning application in image recognition.
Tech Stack: Python with libraries OpenCV (cv2), NumPy, and Matplotlib.
Key Features
Image Processing with Thresholding and Filtering
- Grayscale Conversion → Transforms the color image into a single-channel grayscale representation to simplify processing.
- Image Thresholding → Implements various techniques (
BINARY,TRUNC,OTSU, etc.) to segment an image based on pixel intensity. - Spatial Filtering → Applies different blurring techniques (Mean, Gaussian, Median, Bilateral) to smooth the image and reduce noise.
Edge and Feature Detection
- Laplacian Detection → Uses the Laplacian operator to identify areas of rapid intensity change, highlighting edges in the image.
- Sobel Filters → Applies Sobel operators in the X and Y directions to calculate the gradient and separately detect vertical and horizontal edges.
- Canny Edge Detection → Implements the advanced Canny algorithm, which is highly effective at detecting a wide range of edges by suppressing noise and using hysteresis thresholding.
Shape and Object Detection
- Adaptive Thresholding → Uses Gaussian Adaptive Thresholding to create a sharp binary image, which is crucial for contour detection, even under uneven lighting.
- Contour Finding → Identifies and extracts contours (boundaries of shapes) in the binary image using
cv2.findContours. - Geometric Classification → Implements a custom function that uses the number of contour vertices (
cv2.approxPolyDP) and the aspect ratio to accurately classify shapes (triangle, square, rectangle, pentagon, circle, oval).
Pattern Recognition with Face Recognition
- Face Detection → Uses a pre-trained Haar Cascade classifier to quickly and accurately detect faces within images.
- Model Training (LBPH) → Trains a Local Binary Pattern Histograms (LBPH) Face Recognizer using a collection of grayscale face images.
- Real-time Prediction → Predicts the identity of faces in test images, drawing bounding boxes and displaying a confidence score.