Floor Plan Retrieval

An Android application for retrieving similar floor plans and architectural images using graph-based structural matching and semantic ranking.

Overview

This project helps architects and architecture students search for similar floor plans from a digital repository of architectural layouts. The system allows users to upload images to a cloud-hosted Firebase database and retrieve the most similar floor plans for a given query image. The retrieval pipeline performs feature extraction, matching, and ranking on the server side to return relevant results.

Features

• Reverse image search for floor plans
• Upload floor plan images to a Firebase database
• View uploaded images inside the application
• Retrieve similar layouts for a given query image
• Rank results using structural similarity and semantic difference

Tech Stack

• Android/Kotlin for the mobile application
• Firebase Realtime Database and Firebase Storage for cloud-hosted image storage
• Python for server-side retrieval and matching
• Android Studio for development

Dataset

The project uses the SESYD floor plan dataset. The database contains 5 layout types, with 20 floor plan images per type, and includes 12 standard decor symbols.

Retrieval Pipeline

The retrieval process is divided into three main stages:

1. Feature extraction

   • Layout segmentation
   • Boundary extraction and morphological operations
   • Connected component analysis for room segmentation
   • Adjacent Room Detection (ARD)
   • Graph construction from room adjacency

2. Structural matching

   • Graph spectral embedding is used to represent each layout in a pattern space
   • Similarity is computed by measuring distances between embedded graph representations

3. Semantic ranking

   • Retrieved floor plans are re-ranked based on semantic differences, such as the arrangement and number of furniture objects inside rooms

App Workflow

The Android app includes the following screens:

• Boot activity
• Home screen
• Search screen
• Upload pictures screen
• Show uploads screen

Users can select a query image, upload images to the database, and view ranked retrieval results directly in the app.

Results

The application successfully implements:

• Reverse image search
• Image upload to the database
• Display of uploaded images
• Retrieval of the top similar floor plans for a query image

Use Cases

This project can be useful for:

• Architects searching for similar existing layouts during design
• Architecture students storing and retrieving project layouts efficiently
• Exploring content-based retrieval techniques for architectural floor plans