Computer Vision Projects

This repository contains a collection of computer vision projects developed as part of the Foundations of Intelligent Systems Vision course at University of Cordoba. Each project demonstrates a key concept in image processing or computer vision, implemented in C++ using OpenCV. Below are detailed descriptions, usage instructions, and techniques used in each project.

---

1. Camera Calibration

Objective

Estimate the intrinsic and extrinsic parameters of a camera using chessboard images and apply image undistortion based on the results.

Techniques Used

• Chessboard corner detection
• cv::calibrateCamera() for calibration
• cv::undistort() for rectifying images

Input/Output

• Input: Multiple chessboard images (e.g. calib1.png, calib2.png, ...)

• Output:

  • calibration.yml: Calibration parameters
  • Undistorted versions of images

Run Instructions

./calibrate calib1.png calib2.png ... calibration.yml
./undistort calibration.yml input.png output.png

---

2. CBG Process (Color By Gray)

Objective

Create stylized versions of images where the grayscale intensity controls the coloring effect.

Techniques Used

• RGB to Grayscale conversion
• Channel manipulation
• Color masking based on intensity

Run Instructions

./cbg_process input.png output.png

---

3. Chromakey (Green Screen Removal)

Objective

Remove a green screen background and replace it with a custom image.

Techniques Used

• Convert to HSV color space
• Create masks for green background
• Composite two images using the mask

Input/Output

• Input: Green screen subject image and a background image
• Output: Composited image with background replaced

Run Instructions

./chromakey subject.png background.jpg output.png

---

4. Image Equalization

Objective

Improve image contrast using histogram equalization.

Techniques Used

• Convert to grayscale
• Apply cv::equalizeHist()
• Visualize histogram before/after

Input/Output

• Input: Grayscale or color image
• Output: Contrast-enhanced image

Run Instructions

./img_equalization input.jpg output_equalized.jpg

---

5. Pollen Grain Classification

Objective

Train and evaluate a machine learning model to classify 15 types of pollen grains.

Techniques Used

Feature Extraction:

• [0,1] normalized grayscale features
• Mean-stddev normalization

Classifiers:

• K-Nearest Neighbors (KNN)
• Support Vector Machines (SVM)
• Random Trees (RTrees)

Evaluation:

• Confusion matrix computation
• Recognition rate per class
• Overall accuracy and mean recognition rate

Dataset

• Images in data/train/, data/test/, data/valid/
• Labels provided via corresponding .csv files

Run Instructions

# train a model
./train_clf data/train knn_model.xml knn 5

# test the model
./test_clf data/test knn_model.xml predictions_test.csv

---

6. Sharpening and USM Enhance

Objective

Apply sharpening filters to enhance image details and edges.

Techniques Used

Sharpening:

• Laplacian kernel convolution
• Subtracting Laplacian result from original image

USM (Unsharp Masking):

• Apply Gaussian blur
• Subtract blurred version from original
• Add weighted mask back to the original

Input/Output

• Input: Raw image
• Output: Sharpened and USM-enhanced versions

Run Instructions

./sharpen input.jpg output_sharp.jpg
./usm_enhance input.jpg output_usm.jpg

---