Furniture Extrac-on From Designed Home Decora-on and its Matching Guang Yang1, Chunliang Zheng2!
Department of Energy Resources Engineering1, Department of Electrical Engineering2, Stanford University
Motivation Work Flow & Methodology
Related Work
• Parse furniture items from designed indoor scheme and find similar ones for users’ own decoration in laptop.
• Use a handy android mobile application to quickly get information of the furniture items of interest in real life .
Android Client
Feature Matching / Image Retrieval
Image Segmenta5on
Object Recogni5on
Server
1. Harris Keypoint detec5on 2. Watershed. 3. Coutours based.
SVM classifiers.
Matlab
Contour based shape matching MexOpenCV
Vlfeat
Experimental Results
1. MexOpenCV library, Kota Yamaguchi, Stony Brook University
2. Caltech 101 Object Recognition, L. Fei-Fei, R. Fergus and P. Perona, CalTech University
3. Contour Correspondence via Colony Optimization, Oliver van Kaick, Simon Fraser University
4. Berkeley Segmentation Benchmark, UC Berkeley
Intermediate Results
Right: Watershed
Left: Harris Keypoint Detection
Combining three segmenta5on techniques, we get preKy nice results.(shown in demo) Contour shape matching is computa5onal efficient and works properly for furniture matching.