Project 4 Image Search based on BoW model with Inverted File System
Experiment Setting Image database – UKBench dataset: images from 2550 categories – DupImage dataset: images from 33 categories – Holidays dataset: images from 500 category Feature extraction – SIFT feature ( Source code:
Task: develop an image search demo Alternative 1: Codebook-based image search – Train a 1-million visual codebook hierarchically 5M Training sample: – Index database images with the codebook With inverted file structure – Do image search with any query If possible, please add an interface based on MFC, Qt, etc. Alternative 2: Codebook training-free image search – Transform each SIFT to binary SIFT by scalar quantization (MM’12) – Index database images with 32-bit codeword With inverted file structure – Save the remaining bits in the inverted list – Do image search with any query If possible, please add an interface based on MFC, Qt, etc.
Implementation Program with C++ or Matlab – You may need Open CV when programing with C++ OpenCV 210 library files are provided – You do not need to install the OpenCV source file Refer to OpenCV China for instructions to set programming environment – Refer to OpenCV China for instructions to process images : –