rainfrost1 / py-cbir

Demo of Content Based Image Retrieval in python

Demo of Content Based Image Retrieval, implemented by Python and Tornado.

Image descriptors

• perceptual hash
• Otsu's method
• gray/RGB/YUV/HSV histograms
• GIST
• HoG and LSH (built by Kmeans clustering)
• SIFT and LSH (built by Kmeans clustering)
• Dense SIFT

Distance functions

• Hamming distance, or norm0 distance (L0)
• abs distance (L1)
• Eculidean distance (L2)

Simple re-ranking

• blending: mix results
• ensembling: weighted sum

Code structure

• util/: feature descriptors, feature and LSH preparation
• app/: http server, matching and retrieval
• templates/: html templates
• static/: datasets, js, css
• conf/: log.conf, and for feature data
• logs/: for log data
• settings.py: http port, common setting
• urls.py: server url path

Dependencies

• Tornado
• Image
• numpy, scipy

Run (Linux or Mac)

• cd util/pyleargist-2.0.5/lear_gist/ && make && cp compute_gist ../../ && cd -
• cd util && python prepare.py && cd -
• python main.py
• access http://localhost:19999/cbir

How to change dataset

• add a new image folder in static/dataset/
• in util/prepare.py, change dataset to the folder name, like dataset = 'ferrari'
• run as previous section

Author

Any question, please contact: Zuotao Liu([email protected])