CSC2014 - Digital Image Processing

This project contains solutions for various digital image processing tasks focusing on video analysis and face detection.

Project Structure

CSC2014/
├── Task A/
│   ├── Section 1.py          # Day/Night detection and brightness adjustment
│   ├── Section 2.py          # Face detection and blurring
│   ├── Section 3.py          # Final output assembly
│   ├── Section 4.py          # Watermarking and final enhancement
├── Task B/
│   └── Task_B.py             # Paragraph extraction from scientific papers
├── Converted Paper (8)/      # Input images for Task B
├── Recorded Videos (4)/      # Input video files
├── Output/
│   ├── Task A/
│   │   ├── Section 1/        # Brightness comparison charts
│   │   ├── Section 2/        # Face-blurred videos
│   │   ├── Section 3/        # Intermediate outputs for final assembly
│   │   └── Final_Output_Videos/ # Watermarked and enhanced videos
│   └── Task B/               # Extracted paragraphs and histogram visualizations
├── Dependency/
│   └── face_detector.xml     # Haar cascade classifier for face detection
└── README.md

Task A

Section 1: Day/Night Detection & Brightness Enhancement

Objective: Detect whether a video is taken during daytime or nighttime and enhance brightness for nighttime videos.

Features:

• Automatic day/night classification based on average brightness and histogram analysis
• Adaptive brightness enhancement for nighttime videos
• Visual comparison charts showing before/after brightness levels
• Support for multiple video formats (.mp4, .avi, .mov, .mkv)

Algorithm:

• Analyzes average brightness across video frames
• Uses histogram distribution to determine lighting conditions
• Applies gamma correction and exposure adjustment for nighttime videos

Section 2: Face Detection & Blurring

Objective: Detect and blur all camera-facing faces in videos while maintaining tracking stability.

Features:

• Advanced Face Tracking: Uses KCF/CSRT trackers with motion prediction
• Scene Density Analysis: Automatically adapts to low/medium/high density scenarios
• Smooth Motion Handling: Reduces jitter and tracking discontinuities
• False Positive Filtering: Multi-layer validation (texture, brightness, aspect ratio)
• Adaptive Blur Intensity: Dynamic blur strength based on tracking confidence

Technical Highlights:

• Multi-scale face detection for crowded scenes
• Motion prediction for handling occlusion and profile faces
• Bounding box smoothing to eliminate visual artifacts
• Intelligent tracker management with confidence scoring
• Real-time processing with progress monitoring

Section 3: Final Output Assembly

Combines processed videos and prepares them for final watermarking and output.

Section 4: Watermarking & Final Enhancement

Objective: Overlay a watermark on all output videos and apply final brightness enhancement for improved visual quality.

Features:

• Automatic resizing of watermark images to match video resolution
• Adjustable watermark transparency (alpha blending)
• Brightness and contrast enhancement for all frames
• Appends a custom ending video to each output

Technical Highlights:

• Uses OpenCV's addWeighted for watermark blending
• Frame-by-frame brightness adjustment using convertScaleAbs
• Supports batch processing of multiple videos

Output:

• Watermarked and enhanced videos saved in Output/Task A/Final_Output_Videos/ with prefix watermarked_

---

Task B: Paragraph Extraction from Scientific Papers

Objective: Automatically detect, extract, and sort paragraphs from double/triple-column scientific paper images, filtering out tables and saving results for further analysis.

Features:

• Column detection using vertical whitespace analysis
• Paragraph segmentation based on row pixel density and spacing
• Table region filtering to exclude non-paragraph content
• Histogram projection visualizations for both rows and columns
• Batch processing of multiple input images

Technical Highlights:

• Uses OpenCV and NumPy for image processing and histogram analysis
• Table regions detected and masked using pixel density thresholds
• Paragraphs sorted top-to-bottom and left-to-right for correct reading order
• Output includes cropped paragraph images and histogram visualizations

Output:

• Cropped paragraph images saved in Output/Task B/ with filenames indicating source image, column, and paragraph order
• Row and column histogram images for each input saved in Output/Task B/

Usage:

python "Task B/Task_B.py"