11
Periscope: A Content-Based Image Retrieval Engine By Antigoni M. Founta Student ID: 647

Periscope: A Content-based Image Retrieval Engine

Embed Size (px)

Citation preview

Periscope: A Content-Based Image Retrieval Engine

By Antigoni M. FountaStudent ID: 647

Goal: Search using an Image as a query

The Project

ComponentsThe essential components of the

project!

● Python

● PostgreSQL

● OpenCV

● Scikit-Image

● Python Libraries: ScıPy, NumPy

● Flask

● Bootstrap

Algorithm1. Create schema

2. Upload image and calculate histograms

3. Add image and calculated features to the database

4. Search for similarity in saved images

5. Return a subset with the most similar according to a distance measure

The TasksSome required tasks for the

project.

1. Extract Features

2. Add Image to Database

3. Search through Images and Compare Features

4. Return Similar Images

1. Extract Features

● Color Vector: CalcHist()OpenCV method - HSV

● Texture Vector: LBP (Local Binary Patterns)Scıkıt-Image Lıbrary - GrayScale

● Shape Vector: Hu Moments OpenCV method - GrayScale

Normalize all!

2. Add Image to DB

● Save image to local folder

● Get path of image

● Save path and features on Database

3. Search through Images

● Concatenate vectors

● Calculate ChiSquare (x2) distance

● Get results

4. Return Similar

● Sort Results

● Keep 8 best

● Present→

Add Many Images

More FeaturesAnnotations

Video

Audio

Ontology

Crawling

Future Work

Text-Based Retrieval

Image Segmentation

Resources● PyImageSearch (http://www.pyimagesearch.com/)

● HanzraTech (http://hanzratech.in/2015/05/30/local-binary-patterns.html)

● Unsplash (https://unsplash.com/)