Image Classification and Noise Reduction GUI Application

A Python-based GUI application for Image Noise Reduction and Image Classification (Cats vs Dogs).

The application allows users to:

• Apply multiple noise reduction filters to images
• Classify images using multiple machine learning algorithms
• Interact with a clean GUI built using customtkinter

---

How to Get Started

Prerequisites

• Python 3.6+
• Required libraries:
  • numpy
  • tkinter
  • customtkinter
  • scikit-learn
  • opencv-python
  • matplotlib
  • Pillow (PIL)
  • pickle

---

Installation

1. Create a virtual environment

python3 -m venv env

2. Activate the virtual environment

source env/bin/activate

3. Install required packages

pip install -r requirements.txt

---

Running the Application

python3 App.py

---

Project Documentation

Overview

This project implements an Image Processing GUI Application that supports:

• Image Noise Reduction
• Image Classification (Cat vs Dog)

The application integrates multiple classical image filters and multiple machine learning algorithms for classification.

---

Project Structure

IMAGE_PROCESSING/
├── Image_Classification_App/
│ ├── models/
│ ├── data_loader.py
│ ├── decision_tree_model.py
│ ├── knn_model.py
│ ├── svm_model.py
│ ├── Prediction.py
│ └── main.py
│
├── Image_Denoising_App/
│ ├── images/
│ ├── Average_Filter.py
│ ├── Gaussian_Filter.py
│ ├── Median_Filter.py
│ └── main.py
│
├── App.py
├── README.md
├── requirements.txt
└── .gitignore

---

Noise Reduction Algorithms

Median Filter

Purpose: Removes salt-and-pepper (impulse) noise.

How it works:

• Replaces each pixel with the median of neighboring pixels.
• Non-linear filtering approach.

Strengths:

• Preserves edges effectively
• Very robust against impulse noise

Weaknesses:

• Computationally expensive
• Less effective for Gaussian noise

---

Average (Mean) Filter

Purpose: General image smoothing.

How it works:

• Replaces each pixel with the average of its neighborhood.
• Linear filter.

Strengths:

• Simple and fast
• Reduces random noise

Weaknesses:

• Blurs edges and fine details
• Weak against salt-and-pepper noise

---

Gaussian Filter

Purpose: Smooths images while preserving structural details.

How it works:

• Uses a Gaussian-weighted average of neighboring pixels.
• Linear filter.

Strengths:

• Produces smoother results than the average filter
• Better edge preservation

Weaknesses:

• Not suitable for impulse noise
• Parameter tuning required (kernel size, sigma)

---

Image Classification Algorithms

The application now supports three machine learning algorithms for image classification:

• Support Vector Machine (SVM)
• K-Nearest Neighbors (KNN)
• Decision Tree (DTree)

---

Feature Extraction

All classifiers require numerical feature vectors, not raw images.

Supported feature extraction techniques include:

• Histogram of Oriented Gradients (HOG)
• SIFT (Scale-Invariant Feature Transform)
• ORB (Oriented FAST and Rotated BRIEF)
• Flattened pixel intensities (baseline approach)

---

Dataset Preparation

• Labeled images (Cats / Dogs)
• Dataset split into:
  • Training set
  • Testing set

---

Support Vector Machine (SVM)

Training:

• Uses extracted feature vectors
• Optimized using hyperparameters:

Key Parameters:

• Kernel: Linear, RBF, Polynomial
• C: Regularization parameter
• Gamma: Controls influence of data points (RBF kernel)

Strengths:

• High accuracy for well-separated data
• Effective in high-dimensional spaces

Weaknesses:

• Sensitive to parameter tuning
• Slower training for large datasets

---

K-Nearest Neighbors (KNN) — New Algorithm

How it works:

• Classifies an image based on the majority class of its K nearest neighbors .

Key Parameters:

• K: Number of neighbors
• Distance metric: Euclidean (default)

Strengths:

• Simple and intuitive
• No training phase (lazy learning)

Weaknesses:

• Slow prediction for large datasets
• Sensitive to noisy data and feature scaling

---

Decision Tree (DTree) — New Algorithm

How it works:

• Builds a tree of decisions based on feature thresholds.
• Each leaf node represents a class.

Key Parameters:

• Max depth
• Minimum samples per split

Strengths:

• Easy to interpret and visualize
• Fast inference
• Handles non-linear decision boundaries

Weaknesses:

• Prone to overfitting
• Less stable with noisy data

---

Testing and Prediction

• Models are evaluated on the test dataset
• The trained model predicts whether the input image is:
  • Cat
  • Dog

---

Summary of Supported Algorithms

Task	Algorithms
Noise Reduction	Median, Gaussian, Average
Image Classification	SVM, KNN, Decision Tree