SIFT based Instance Search

Introduction

This project aims to develop a program to detect objects with datasets(a random set of images). The first method is SIFT Feature Matching. SIFT is a feature detection algorithm to detect the image features. This project used the feature matching method to find out the similarity. The second method is SSIM who consider image luminance, contrast and structure to evaluate similarity.

How to use

1. Clone this project
2. Put your datasets and test images into folder Images and Queires respectivily
3. Run the main program Object_Detection.ipynb with jupyterlab

Dataset Image Format: 0001.jpg, 0002.jpg, 0003.jpg, ... , 9999.jpg

Test Image Format : 01.jpg, 02.jpg, 03.jpg, ... , 99.jpg

File Structure

.
├── Images                  # Datasets folder
│   ├── 0001.jpg
│   ├── 0002.jpg         
│   ├── ...         
├── Queries                 # Test files folder
│   ├── 01.jpg
│   ├── 02.jpg         
│   ├── ...      
└── Object_Detection.ipynb  # Main Program

Result

Summary

Dataset Size: 5000 images

Item	SIFT	SSIM
Mean average precision	0.332950	0.011360
Time	2.5 hr	20 mins

Details Comparison

Short Review

1. SIFT is too slow for matching. It should combine with Bag of Word.